Manifold Grace← Ventures

VENTURES / EDUCATIONAL SYSTEMS GUIDE

From venture idea to accountable systems.

Explore how a website, payments, business data, automation and AI could work together. Expand each section for architecture, controls, customisation options and delivery evidence.

Educational planning resource. Venture concepts are being explored collaboratively. This guide is not a claim of an operating company, active integrations or a consulting service. All paths below are reference designs; no live payments or AI actions are connected.

01 / PaymentsCheckout → ledger → settlement → account
02 / Security & complianceAuthority, privacy, risk and audit evidence
03 / GrowthMarketing, SEO, analytics and future sales
01 · Requirement coverage & review recordTrace the recorded brief to concrete explanations, examples and evidence.

This review maps 54 requirements from this chat and the legible relevant notebook items to the page. It covers the recorded brief; it is not a claim that one page exhausts AI engineering or establishes legal compliance. Unclear handwriting and personal planning notes are not silently turned into public claims. Decisions that need the actual venture, data, jurisdiction, provider or contract remain explicitly unresolved.

Recorded requirementWhere to inspectCoverage / evidence
Venture count versus income streamsOpen sectionDistinct ventures, streams and entity relationships; no invented count.
Payment gateways as first priorityOpen sectionCheckout, states, merchant ownership and gateway controls.
Money reaching the accountOpen sectionPayout, fees, bank reconciliation and exceptions.
Security as second priorityOpen sectionIdentity, permissions, security engineering and AI threats.
Market data and development evidenceOpen sectionDemand, campaigns, conversion and decision-quality data.
Marketing and promotionOpen sectionChannels, audience eligibility, measurement and approval.
SEO and future salesOpen sectionTechnical/search content checks and qualified conversion.
Predictive analyticsOpen sectionTraining, time-aware evaluation and forecasting limits.
Generic customisable automation presetsOpen sectionTwelve presets with explicit control fields.
Structured templates before engagementOpen sectionFourteen downloadable records and their fields.
GUI and backend relationshipOpen sectionTrust-boundary diagram, API, state, queue and connectors.
More than ten customisationsOpen sectionTwenty concrete GUI/application options.
Data platform clarificationOpen sectionCRM, ERP, database, warehouse, object storage, queues and BI.
Zapier-style trigger and website connectorOpen sectionWebhook, polling, batch, request/response and orchestration contracts.
Visual architectural relationshipsOpen sectionLabelled system diagram and operational path.
ML signal path for educationOpen sectionInputs, weights, calculation, output, policy, action and feedback.
Intelligent automation explainedOpen sectionRules, orchestration and evaluated AI roles.
AI engineering explainedOpen sectionRequirements through data/model/application/infrastructure and operations.
Secure AI and responsible innovationOpen sectionNeed, affected users, threats, controls and contestability.
Projects separate from governanceOpen sectionDistinct categories with lifecycle relationships.
Transformation trackingOpen sectionBenefits, adoption, baseline/target/actual and operating evidence.
Governance, ethics and risk managementOpen sectionOwners, decisions, risk treatment and assurance.
Data protectionOpen sectionPurpose, data roles, DPIA screening, rights, retention and transfers.
Salesforce, CRM, ERP and EinsteinOpen sectionPlatform distinctions, example path and licence limitations.
Educational, not consulting or active-company claimOpen sectionPage framing, collaborative role allocation and illustrative scope.
Expandable sections with detailOpen sectionRequirement links open associated sections; expand/collapse controls.
Milestones visually illustratedOpen sectionSix-stage roadmap, acceptance gates and iteration.
Delivery / business-development / support rolesOpen sectionDefined remit, authority, acceptance and commercial boundaries.
Government controls and auditOpen sectionApplicability register; law versus framework; evidence and ongoing review.
Workplace integration varietyOpen sectionSixteen categories with contracts and failure risks.
Customer and demand validationOpen sectionEvidence before investment and build choices.
Profitability and cash flowOpen sectionScenarios, costs, margins and settlement timing.
Build versus buyOpen sectionConfiguration, extensions, platform automation and bespoke comparison.
Ownership and exitOpen sectionAccounts, IP, exports, revocation and migration.
Customer journey and fulfilmentOpen sectionPre-purchase through support, cancellation and refund.
Ongoing operating responsibilitiesOpen sectionOwner, backup, support, monitoring and maintenance.
Continuity and recoveryOpen sectionBackup, restore, outage, safe manual route and tolerances.
Measurable acceptanceOpen sectionConcrete tests and responsible reviewers.
Accessibility and usabilityOpen sectionCore journeys, keyboard, devices and access needs.
AI suitability and limitsOpen sectionNon-AI baseline, data adequacy and prohibited actions.
Training and handoverOpen sectionDemonstrated independent operation and evidence bundle.
Ongoing benefit and cost reviewOpen sectionMeasured net value and improve/retire decisions.
Purpose, owner, cost, data, dependencies, evidence, exit per componentOpen sectionWorked reviewable scope and component proposal template.
Ten project outlinesOpen sectionTen scoped educational briefs with roles and acceptance gates.
Demo, prototype, implementation, login and refinementOpen sectionMaturity roadmap, existing workspace boundary and verified sign-in requirements.
Co-founder / partnership and commissionOpen sectionOwnership, authority, payment and exit parameters.
DPO role clarityOpen sectionNeed assessment, independence and conflict review.
Trade marks, domain and IPOpen sectionClearance questions, ownership and licensing records.
Business model / CIC and administrative readinessOpen sectionEducational options and applicability checks; no assumed formation.
Paid experience and recognitionOpen sectionFee, attribution, testimonial permission and evidence.
Learning / knowledge gaps and project rolesOpen sectionSkills matrix, practical review, PM and AI engineering distinctions.
Enterprise management, ERP, CRM, ERM, PPM, people and ITOpen sectionManagement umbrella, system relationships, intended benefits and source-specific LEADing Practice distinction.
×7 MSI means seven planned income streamsOpen sectionSeven planning slots; mechanisms kept distinct from ventures and legal entities.
Charts, dashboards and populated audit examplesOpen visual gallerySeven synthetic dashboard examples, chart data tables, explicit calculations and a failed-control-to-retest audit trail.

Review outcome: the earlier outline has been expanded with connector contracts, worked payment and ML examples, control tests, responsibility allocation and handover evidence. Not activated: payment collection, bank feeds, external connectors, real-data models or autonomous actions. Still to decide: actual venture register, suppliers/editions, budgets, named owners, applicable legal duties, model datasets and acceptance thresholds.

02 · Visual dashboards, charts & completed audit examplesRead the system at a glance, then inspect underlying evidence.

Seven visual examples show what the written controls look like in use. Charts can be scrolled horizontally on narrow screens; exact values remain available in the accompanying tables. All figures are invented for teaching; they are not forecasts, live dashboards or statements about operating businesses. Expand a dashboard for charts, underlying values and the decision it supports.

Seven-stream management dashboardSee which streams generate contribution before shared operating costs.

SYNTHETIC TEACHING DATA · invented scenario, not Grace’s revenue, clients, projects, forecasts or live operating results. Each dashboard has its own scenario and denominator; do not combine their totals.

Scenario sales£7,500Illustrative month
Contribution£4,240Sales less £3,260 variable costs
Shared overhead£1,200Assumed period costs
Illustrative surplus£3,040Before tax; teaching assumptions
Sales and contribution by income stream
£0£750£1,500£2,250£3,000MSI-01£2,400£1,440MSI-02£1,800£1,080MSI-03£1,200£600MSI-04£900£540MSI-05£600£300MSI-06£400£200MSI-07£200£80

Scenario salesContribution

View data, calculation and management decision
StreamSales £Variable costs £Contribution £
MSI-0124009601440
MSI-0218007201080
MSI-031200600600
MSI-04900360540
MSI-05600300300
MSI-06400200200
MSI-0720012080

Contribution = sales − defined variable costs. Scenario surplus = total contribution − shared overhead. The simplified example excludes tax, financing and accounting adjustments. Compare contribution, workload and risk before deciding to expand a stream; high sales alone do not establish profitability.

Payments and reconciliation dashboardDistinguish sales, gateway adjustments, cash received and pending settlement.

SYNTHETIC TEACHING DATA · invented scenario, not Grace’s revenue, clients, projects, forecasts or live operating results. Each dashboard has its own scenario and denominator; do not combine their totals.

Gross captured£10,000Scenario period total
Refunds / fees£400 / £288Hypothetical; not provider rates
Net payable£9,312Captured − refunds − fees
Bank received£9,000£312 remains pending
Where the £9,312 net payable sits
£9,000 bank received£312 pending
Reconciliation checkAmount £Result
Gross captured less refunds and fees10,000 − 400 − 288 = 9,312Agrees with net payable.
Bank received plus pending payout9,000 + 312 = 9,312Balances; pending cash is not bank cash.
Duplicate eventNo second financial effectRejected by deduplication test.
Late payout312 pendingReview against agreed payout timing; assign owner if overdue.

Read this as a cash-location view, not profit. Reconciliation exceptions require references, ownership and resolution evidence. See the worked payment path for event-level detail.

Sales, marketing and SEO dashboardSeparate visibility, qualified demand, purchases and acquisition economics.

SYNTHETIC TEACHING DATA · invented scenario, not Grace’s revenue, clients, projects, forecasts or live operating results. Each dashboard has its own scenario and denominator; do not combine their totals.

Search visibility5,000 impressionsSeparate search scenario; 250 clicks
Search CTR5%250 ÷ 5,000 × 100
Cohort conversion1%10 purchasers ÷ 1,000 visitors
Acquisition cost£30£300 campaign spend ÷ 10 new customers
One illustrative conversion cohort — unique people at each stage
02505007501,000Landing cohort1,000Enquiries100Qualified leads40Purchasers10

People

Stage transitionCalculationRate
Visitors → enquiries100 ÷ 1,00010%
Enquiries → qualified leads40 ÷ 10040%
Qualified leads → purchasers10 ÷ 4025%
Visitors → purchasers10 ÷ 1,0001%

The 250 search clicks are not added to the 1,000-person cohort: click counts and unique people are different measures. Assume the ten purchasers are new customers attributed to the £300 campaign solely for this teaching calculation. Real attribution, consent and cross-device gaps require explicit treatment. No sales or ranking outcome is promised.

Project and transformation dashboardShow dependency, spend and acceptance instead of an unsupported progress percentage.

SYNTHETIC TEACHING DATA · invented scenario, not Grace’s revenue, clients, projects, forecasts or live operating results. Each dashboard has its own scenario and denominator; do not combine their totals.

Accepted gates2 of 6Gate count is not effort completion
Budget cap£5,000Illustrative approved cap
Actual spend£1,500As of scenario snapshot
Forecast total£4,800£3,300 remaining; £200 headroom
Illustrative delivery roadmap — dependency-led, not a promised timetable
W1W2W3W4W5W6W7W8EndDiscoverAcceptedDesignAcceptedSandboxIn progressPilotBlocked by G2ReleaseNot authorisedBenefits reviewPlanned
Management viewScenario stateNext decision
Critical dependencyPilot waits for sandbox gate G2.Complete replay/recovery tests before granting pilot access.
Change requestA new CRM connector is proposed.Assess scope, cost, data access and delivery impact before approval.
Benefit baselineManual task: 20 minutes; pilot target: 8 minutes.Measure review and rework too; no benefit reported as achieved yet.
AcceptanceTwo gates accepted; sandbox work ongoing.Sponsor receives evidence, defects and residual risk before next gate.
Risk, controls and audit dashboardUse risk position to prioritise review, then inspect control evidence.

SYNTHETIC TEACHING DATA · invented scenario, not Grace’s revenue, clients, projects, forecasts or live operating results. Each dashboard has its own scenario and denominator; do not combine their totals.

Risks under review4Illustrative residual positions
Overdue treatments1R1 access-control retest
Controls tested8 of 10Six pass; two fail; two not tested
Failed controls2Failures remain visible until retest
Residual risk map — illustrative 1–5 ordinal scale
1122334455R1R2R3R4Likelihood →Impact →
RiskImpact / likelihoodRequired actionAccountable role
R1 · Data leakage4 / 3Close access-control defect and retest before pilot.Security/risk authority
R2 · Duplicate fulfilment3 / 2Verify effect-key uniqueness and recovery evidence.Integration owner
R3 · Supplier outage2 / 3Exercise manual fallback and restoration.Service owner
R4 · Forecast error2 / 2Compare baseline and subgroup errors.Model owner

Ordinal ratings are prioritisation aids, not measured probabilities or monetary loss. Labels, thresholds and risk appetite must be agreed for the actual use case. Colour never replaces the risk ID, score, treatment and authority.

AI quality and operations dashboardCompare models on the same test set and distinguish operational measures.

SYNTHETIC TEACHING DATA · invented scenario, not Grace’s revenue, clients, projects, forecasts or live operating results. Each dashboard has its own scenario and denominator; do not combine their totals.

Baseline success80%80 / 100 fixed test tasks
Candidate success86%86 / 100 same tasks
Improvement+6 percentage pointsNot proof of production uplift
Candidate p95 latency1.8 sFrom separate synthetic request sample
Held-out task success — same 100-task test set
0255075100Baseline80Candidate86

Success percentage

MetricIllustrative observationInterpretation / review
SeverityCandidate makes 14 errors; 1 is high severity.Higher average success alone does not justify deployment; inspect unacceptable errors.
Subgroup checkGroup A: 47/50 = 94%; Group B: 39/50 = 78%.Overall 86/100 hides a 16-point gap. Review representativeness, causes and consequences.
Human override12 overrides / 100 reviewed outputs.Override rate is not identical to error rate; record reasons and reviewer consistency.
Inference cost£0.02 average per request in a separate 1,000-request sample.£20 sample cost; include failed calls, retries and review cost in real total-cost measures.
Input shiftIllustrative feature PSI = 0.18.One indicator of distribution change, not a universal pass/fail threshold or proof of concept drift.
Release decisionHOLD: high-severity error and subgroup gap need review.Risk authority and release owner assess evidence; the dashboard cannot approve itself.

No real model performance is reported. Accuracy/task success, calibration, fairness, cost and latency answer different questions; do not compress them into one “AI score”.

Completed audit example and reusable audit recordSee the evidence behind a finding, corrective action and justified closure.

SYNTHETIC TEACHING DATA · invented scenario, not Grace’s revenue, clients, projects, forecasts or live operating results. Each dashboard has its own scenario and denominator; do not combine their totals.

Audit evidence chain — a failed control is corrected and retested
  1. Requirement → control
  2. Test → initial FAIL
  3. Finding → release blocked
  4. Correction → new evidence
  5. Independent retest → PASS
  6. Scoped closure → monitoring
Audit stageWorked exampleEvidence / record
A01 · RequirementOnly authorised venture users can access their own orders.Requirement SEC-01; role matrix v1; scope: synthetic ventures A/B.
A02 · Risk and controlCross-venture disclosure; backend must check actor, action and venture for every record request.Control C-ACCESS-01; owner: service/security lead.
A03 · Test procedureUser A attempts GUI and direct API access to A and B; cover read, update and export, including guessed IDs.Synthetic fixture set F-01; repeatable request script; expected denial without record content.
A04 · Initial evidenceFive of six checks pass; export endpoint returns a B record to user A.E-001 request/response fixture; synthetic record only; finding F-001 OPEN.
A05 · Finding and consequenceControl ineffective on export; block pilot/release for this scenario.High-severity teaching classification; no real incident asserted.
A06 · Corrective actionApply shared server-side authorisation to export and add regression checks.Ticket FIX-001; accountable owner assigned; due before next release gate.
A07 · RetestAll six original checks and six added regression checks pass after the fix.E-002 result set; 12/12 pass; code/config version recorded.
A08 · Independent reviewReviewer checks evidence scope, redaction, reproducibility and residual limitations.Reviewer distinct from implementer where feasible; review recorded, not assumed.
A09 · Closure / residual riskClose F-001 only after review; retain ongoing monitoring and periodic access tests.Synthetic example status: CLOSED AFTER RETEST; earlier failure remains in history.
A10 · Sign-off / limitationRelease authority accepts this control evidence within the agreed scope.No claim that one passed test certifies the complete system or legal compliance.

Evidence identifiers are fictional references, not downloadable logs. This example intentionally includes failure and remediation rather than presenting a perfect checklist. Review evidence coverage, reproducibility, dates, scope and independence before relying on a sign-off.

03 · Enterprise management & seven income streamsCoordinate strategy, resources, systems and controls across planned income streams.

Open the synthetic visual dashboard / worked example →

Enterprise management coordinates strategy, people, capabilities, processes, information, technology and controls across an organisation. It is an operating and decision-making approach, not a synonym for ERP software. For this venture concept, the planning target is seven streams of income (×7 MSI); that does not mean seven companies or seven operating revenue sources.

Management areaPurpose and relationshipEvidence / dashboard view
Strategy and capabilitiesDefine outcomes, customer value, required capabilities and investment priorities. Enterprise architecture describes how business, data, applications and technology fit together.Outcome → capability → process → owner → system → cost → control → measurable benefit.
ERP · enterprise resource planningCoordinate operational records such as finance, purchasing, inventory and fulfilment where needed. Scope depends on the selected platform and configuration.Order, stock, supplier, invoice and reconciliation states with a named system of record.
CRM · customer relationship managementManage relationship history, enquiries, opportunities and service interactions. Connect to ERP/payment state through controlled interfaces.Eligible leads, pipeline stages, conversion, service quality and preference/suppression status.
ERM · enterprise risk managementAssess risks across the enterprise, including strategic, financial, operational, technology and compliance risks and their dependencies.Risk appetite, inherent/residual exposure, treatments, control evidence and named acceptance authority.
PPM · project portfolio managementSelect, prioritise, balance and review projects against strategy, resources, dependencies, costs and expected benefits. Programme management coordinates related delivery; project management controls bounded work.Portfolio priority, capacity, dependencies, approved budget, milestones and benefit realisation. Choose a suitable delivery method rather than assume one framework fits every project.
People and workplace managementCoordinate role design, workforce capacity, onboarding, training, collaboration and fair working practices. Limit access to personnel information.Named roles, capability/skills gaps, access review, workload and training evidence.
IT and service managementOperate applications and infrastructure; manage incidents, service requests, changes, continuity, security and supplier dependencies.Service health, support owner, recovery results, change history, cost and end-of-life decisions.
Governance, compliance and auditProvide cross-cutting authority, obligations, policy and assurance. Controls must operate within workflows and be tested; no software product makes an organisation compliant by itself.Decision register, obligation/control/evidence links, audit findings and overdue actions.
Data and analyticsCreate consistent definitions, lineage, quality checks, permitted sharing and decision support across these functions.Refresh time, metric definitions, uncertainty, data-quality alerts and access-restricted views.
Enterprise coordination across seven planned income streams
  1. Strategy + seven-stream register
  2. Shared capabilities + accountable owners
  3. CRM / ERP / people / IT services
  4. Controlled APIs + workflow orchestration
  5. Payments / fulfilment / customer outcomes
  6. Portfolio + risk + benefit review

Governance, privacy, security, finance and audit apply across the path. A management dashboard provides a coordinated view; each underlying system retains its defined authority. Share a capability where appropriate while preserving venture, entity and personal-data boundaries.

Seven-stream register: count mechanisms, avoid invented operating claims
Stream IDOffer / payer / revenue mechanismReadiness
MSI-01To be defined and validated; link to venture and proposed legal entity.Planning target; no live-income claim.
MSI-02To be defined and validated; link to venture and proposed legal entity.Planning target; no live-income claim.
MSI-03To be defined and validated; link to venture and proposed legal entity.Planning target; no live-income claim.
MSI-04To be defined and validated; link to venture and proposed legal entity.Planning target; no live-income claim.
MSI-05To be defined and validated; link to venture and proposed legal entity.Planning target; no live-income claim.
MSI-06To be defined and validated; link to venture and proposed legal entity.Planning target; no live-income claim.
MSI-07To be defined and validated; link to venture and proposed legal entity.Planning target; no live-income claim.

For each stream record: customer and offer; payer; one-off/recurring/usage/licence/commission mechanism as applicable; pricing assumptions; validation evidence; owning venture/entity; payment and fulfilment path; direct/shared costs; contribution and cash-flow assumptions; owner; risks; dependencies and launch gate. These mechanisms are examples, not an assignment of your seven actual streams.

Silos, redundancy and a “360-degree” view: intended benefits and limits

Integration can reduce duplicate entry and improve coordination, but benefits require consistent definitions, field ownership, reliable synchronisation and adoption. A single shared database is not necessary: federated systems can integrate through APIs/events and governed analytical datasets. Combining unrelated personal records or granting everyone access would create new risks.

A dashboard is only as current and complete as its feeds. Show timestamps, lineage, missing systems, quality exceptions and the distinction between observed outcomes and predictions. “Real time” and “360-degree” are requirements to define and verify, not guaranteed properties of buying enterprise software.

LEADing Practice: source-specific reference

LEADing Practice’s Enterprise Management page describes reusable reference content covering management disciplines and relationships across business, information and technology. Treat this as the provider’s published approach. This educational guide does not reproduce its proprietary reference content, claim certification, endorse its superiority or turn its standards into statutory duties. Check access and licence terms before using provider materials.

NetSuite’s overview offers a software-provider perspective. Compare any vendor’s claims against the actual capability, interoperability, operating-cost and acceptance requirements in this guide.

04 · Ventures & income streamsSeparate venture economics, shared infrastructure, accountable decisions and staged growth.
ConceptMeaningPlanning fields
VentureA distinct business proposition; it may have several income streams.Owner, stage, customer, jurisdiction, budget and next validation gate.
Income streamA way revenue is earned; count validated streams separately from ideas.Product sales, subscriptions, services, licensing, events, sponsorship or commission; record evidence for each.
Legal entityThe contracting, tax and liability boundary; a brand is not automatically a separate company.Proposed merchant, bank account, controller, contracts and professional review.
Shared platformReusable components with isolation between ventures.Venture ID, access boundaries, separate ledgers, cost allocation and retention.

The clarified target is seven planned income streams; no fixed count of separate businesses or entities is inferred. A venture register should distinguish proposed, validated, building, operating and retired streams. Begin with one measurable workflow; extend shared infrastructure only after proving value.

05 · Payment gateways & money into the accountVerify payments, reconcile settlements, track fees and protect account access.

Open the synthetic visual dashboard / worked example →

For a complete synthetic transaction and failure path, see Worked payment example.

Money path — customer payment and bank settlement are separate events
  1. Customer → product or service GUI
  2. Backend → hosted gateway checkout
  3. Gateway → signed payment event
  4. Order ledger → fulfilment eligibility
  5. Gateway → settlement payout
  6. Bank + accounting → reconciliation

The browser success screen is not proof of payment. A server verifies the payment provider’s event and authoritative status before releasing goods or access. Payment collection, a recorded sale, gateway balance and bank payout are different states.

ControlSpecification / evidence
Merchant setupConfirm legal seller, supported business activity, currency, bank ownership, verification and settlement terms.
CheckoutUse provider-hosted payment fields or checkout; keep card numbers and secret keys out of website code, analytics and logs. Confirm applicable PCI scope with the provider.
LifecycleTrack pending, authorised, captured, failed, refunded and disputed states; fulfilment and subscription states are separate.
ReliabilityVerify raw-body signatures and timestamp freshness; deduplicate event IDs and business effects. Handle retries, delayed events and out-of-order delivery. Queue work durably before acknowledging receipt.
SettlementMatch payment IDs → order IDs → payout IDs → bank transactions; explain fees, refunds, currency differences and reserves. Flag unmatched items.
AuthoritySeparate view, refund and payout permissions; require approval for sensitive changes. Alert on bank-account changes.
Commercial metricsGross sales, net revenue, cash received, fees, refunds, disputes, settlement delay and contribution margin. Use an accountant-approved treatment for tax and revenue recognition.
06 · GUI, backend & connector architectureA usable interface controls workflows through authenticated, observable service boundaries.

For connector types, workplace integration categories and field ownership, see Integration catalogue.

System relationships and trust boundaries
Website, secure backend, payment gateway and automation integrationThe website calls an authenticated backend. Checkout goes to a payment gateway whose signed events reach the backend. The backend stores orders and publishes durable events. A workflow engine connects to CRM, ERP and marketing. Optional AI proposes decisions through policy and human approval gates. Audit and monitoring span all backend actions. Bank settlements are reconciled separately.Website GUIAuthenticated backendPayment gatewayOrders / payment ledgerQueue + workflow engineBank reconciliationCRM / ERP / marketingAI → policy / approvalAudit + monitoringAPI + identitycheckoutsigned eventdurable eventpayoutorder eventmatch recordsscoped connectorsoptional inferenceapproved proposaldecision evidenceAll backend actions: scoped identity · data minimisation · correlation ID · audit · recovery
Operational path — the GUI displays state; the backend enforces authority
  1. Website GUI → authenticated API
  2. API → validation + permissions
  3. Durable event → workflow engine
  4. Connector → CRM / ERP / email
  5. Results → ledger + audit log
  6. Status API → GUI dashboard

A Zapier-style connector subscribes to approved events or polls an API. A trigger starts a versioned workflow; filters check conditions; actions call approved services. The browser must never contain connector secrets or unrestricted payment credentials.

BoundaryWhat crosses itRequired protection
Public website → backendValidated request, session and venture identifierTLS, server-side authorisation, CSRF protection where relevant, rate limits and schema validation.
Provider → webhookEvent ID, signature, timestamp and minimum payloadVerify provider authenticity, reject stale/replayed requests and store durable processing state.
Workflow → external platformScoped API operation and correlation IDLeast-privilege OAuth/service account; secret vault; timeouts, backoff and quotas.
AI → action executorProposed structured actionAllowlisted tools, argument checks, explicit authority and human approval for consequential actions.
Services → GUISanitised result, processing state and errorsVenture-level isolation; no secrets, raw personal records or internal stack traces.
Failure and recovery path

Timeout → bounded retry with backoff → dead-letter queue → named operator → inspect and replay safely. Use transaction/outbox patterns where needed, idempotency keys for business effects, correlation IDs across services and an audit record of every replay. A connector does not guarantee exactly-once processing.

07 · Intelligent automation & customisable presetsRules coordinate actions; models handle uncertainty within explicit operational limits.

Intelligent automation combines business-process orchestration with AI where interpretation or prediction is useful. A rule such as “paid order → send receipt” does not require machine learning. A model may classify an enquiry or estimate demand; its output still passes policy and permission checks.

PresetTrigger → conditions → actionCustomise / safeguard
Paid-order fulfilmentVerified captured payment → order valid → release fulfilmentAllowed products; deduplication; exceptions queue.
Failed-payment recoveryFailure event → permitted contact → send recovery informationDelay, retry cap and communication eligibility.
Invoice reconciliationBank import → reference matches → reconcileTolerance, currency and manual approval exceptions.
Lead routingEnquiry → validated category → assign ownerRules first; model confidence threshold and human fallback.
Campaign follow-upEligible contact event → preferences permit → schedule messageChannel rules, suppression list and unsubscribe.
Stock alertInventory threshold → valid supplier → create draft reorderSpend limit and purchasing approval.
Support triageTicket arrives → urgency check → routeSensitive-case escalation; no unsupported promises.
Data-rights requestVerified request → identity adequate → assign workflowScope, deadline tracking and redaction approval.
Security incidentAlert → severity triage → contain and notify ownerRunbook, evidence preservation and recovery authority.
SEO reviewScheduled crawl → issue detected → create taskHuman review before public content changes.
Forecast refreshApproved dataset snapshot → quality gate → scoreBaseline comparison, drift check and publication approval.
Partner handoverMilestone evidence → acceptance check → grant next accessTime-limited permissions and revocation date.

Every preset has: ID, version, owner, trigger schema, conditions, action permissions, data fields, approval threshold, retry policy, failure route, retention, expected cost, test cases and disable switch. Customisation changes configuration; a new purpose or privilege needs a fresh assessment.

08 · Data platforms, CRM, ERP & analyticsChoose platforms by data purpose, ownership, integration effort and evidence.
LayerPurposePossible platform / decision
CRMContacts, lead stages, opportunities and relationship historySalesforce, HubSpot or Dynamics; verify licensing, APIs, permissions and export.
ERP / accountingOrders, procurement, stock, finance and operational recordsERP platforms plus accounting connectors; define system of record per field.
Transactional databaseReliable application state and integrityRelational tables for ventures, orders, payments and workflow runs.
Warehouse / lakehouseHistorical cross-platform analysisCurated analytical data with lineage and access controls; not an operational payment authority.
Object storageDocuments and approved source filesEncryption, signed access, malware checks, lifecycle and deletion rules.
Event bus / queueDurable asynchronous workRetries, dead-letter handling, ordering requirements and observability.
BI and predictive analyticsDashboards, forecasts and outcome estimatesCRM Analytics, Einstein Discovery or equivalent; inspect feature and licence scope.
Vector indexSimilarity retrieval for relevant documentsOptional for retrieval-augmented generation; enforce source permissions and deletion propagation.
Salesforce / Einstein integration example

Website enquiry → authenticated ingestion → CRM record → approved transformation → analytics dataset → deployed prediction model → score displayed in a record or workflow → authorised follow-up. CRM Analytics connects and explores data; Einstein Discovery supplies predictive modelling capabilities in supported editions. The specific Sales Cloud Einstein analytics bundle has restrictions and is not interchangeable with the full analytics platform. Validate the exact licence, external-data permissions, costs, deployment method and connector contract.

Maintain a data catalogue: field definition, purpose, source, system of record, controller/processor role, sensitivity, quality threshold, refresh interval, recipients, location, retention and deletion route. Avoid silently combining unrelated ventures’ customer data.

09 · Custom application development: 20 optionsSpecify capabilities, acceptance tests and ownership before approving bespoke development.
CustomisationLikely parameters / acceptance
1. Venture selectorIsolated dashboards and permissions; unauthorised cross-venture requests fail.
2. Checkout experienceProduct, currency, checkout copy and return route; verified payment state drives fulfilment.
3. Subscription portalPlan change, cancellation, proration and access; test each lifecycle state.
4. Refund approvalsReason, authority and amount thresholds; approvals logged.
5. Reconciliation dashboardLedger-to-payout-to-bank matching; export exceptions.
6. Role-based accessOwner, operator, reviewer and partner scopes; deny by default.
7. AuthenticationMFA, session expiry and recovery; test revocation.
8. Consent and preferencesPurpose, channel and notice version; suppression propagates.
9. CRM forms and fieldsSchema, validation, deduplication and assignment.
10. Workflow editorApproved triggers/actions; versioning and test mode.
11. Connector adaptersOAuth scopes, pagination, webhooks, quotas and revocation.
12. Lead-scoring displayVersioned scores, explanation limits and human override.
13. Forecast dashboardTime horizon, uncertainty range, backtest and baseline.
14. Marketing attributionUTMs, source taxonomy, attribution window and consent state.
15. SEO content workflowMetadata, schema, canonical links and publishing approval.
16. Support assistantApproved knowledge, citations, escalation and tool limits.
17. Data-rights workspaceIdentity verification, search scope, review and completion evidence.
18. Audit explorerActor, action, record, time and correlation ID; protected access.
19. Partner workspaceScoped tasks, evidence submissions, expiry and handover.
20. Accessibility and localisationKeyboard access, labels, contrast, language, currency and timezone.

These are capability specifications for learning and procurement, not an offer to provide consulting or proof that integrations are active. Prefer configuration or a supported extension when it meets the acceptance criteria.

10 · Machine learning signal pathTraining learns parameters; inference applies them to new eligible inputs.

Inspect a numerical inference and the full approval boundary in the ML workbench.

Training path — example: probability that an eligible lead converts
  1. Approved historical records + outcomes
  2. Validate, minimise, label and time-split
  3. Train-only preprocessing + feature creation
  4. Fit model parameters on training data
  5. Validate, calibrate and test held-out data
  6. Version model + approve deployment
Inference path — learning does not happen automatically on every request
  1. New eligible lead → validate input
  2. Same versioned feature transformation
  3. Model → score and uncertainty checks
  4. Policy gate → human / authorised action
  5. Actual outcome → monitored feedback
  6. Review → separately approved retraining
Precisely what the model computes

An illustrative logistic model receives features x (for example enquiry age and expressed product interest). It computes z = b + Σwᵢxᵢ and p = 1 / (1 + exp(−z)). Training adjusts weights w and intercept b to reduce prediction loss on historical labelled examples. The output p estimates the chosen outcome under the training conditions; it is not a guarantee or an explanation of causation.

Fit imputers, encoders and scalers on training data only. Split by time and, where necessary, customer/group to prevent leakage. Never include a future purchase or post-outcome status among the features. Tune on validation data and evaluate the locked model on held-out test data.

Assess calibration, precision/recall and relevant subgroup error; compare with rules and a simple baseline. Choose thresholds by consequences and capacity, not an arbitrary confidence number. Keep a human route for missing inputs, out-of-distribution cases and consequential decisions. Obtain independent evaluation data rather than assuming past actions represent unbiased labels.

Forecasting, LLMs and small-model routing are different

Forecasting estimates future quantities using time-aware backtests, intervals and seasonality; track MAE or another appropriate error against a naïve baseline. An LLM predicts token sequences and may propose tool calls. Retrieval supplies permitted context; it does not itself retrain the LLM. A small-model router can delegate uncertain decisions to a larger model, but confidence must be evaluated against real task outcomes. Larger-model escalation is not human approval.

11 · AI engineering across the lifecycleEngineer complete systems around models, with evaluation, monitoring and recovery.

Open the synthetic visual dashboard / worked example →

DisciplineRequired work / review evidence
Problem and requirementsDefine user need, baseline, exclusions, acceptable harms and measurable success.
Data engineeringIngestion, contracts, quality, provenance, labels, permissions and retention.
Model engineeringBuild or buy decision, features/prompts, model selection, calibration and error analysis.
Application engineeringAPIs, GUI, retrieval, tool use, orchestration, state, validation and accessibility.
InfrastructureEnvironments, identity, compute, network boundaries, secrets and recoverable storage.
MLOps / LLMOpsVersion data/models/prompts, reproducible builds, CI checks, deployment registry and rollback.
EvaluationRepresentative test sets, adversarial tests, factuality, fairness, task success and human usability.
Reliability and costLatency, throughput, availability, token/compute budgets, caching and vendor limits.
Security engineeringThreat model, access enforcement, dependency checks and abuse testing.
Operations and retirementMonitoring, incidents, support ownership, retraining criteria, export and safe decommissioning.

AI engineering is the work that makes an AI-enabled system usable and maintainable. It includes software and data engineering, model evaluation and operations; connecting an API alone does not demonstrate end-to-end quality.

12 · AI governance, ethics, security & riskGovern decisions continuously; verify controls and preserve accountable human oversight.

Open the synthetic visual dashboard / worked example →

DomainControls and evidence
GovernanceAI inventory, accountable owner, intended use, approval authority, risk appetite and periodic review; use Govern, Map, Measure and Manage as connected functions.
Responsible innovationTest whether AI is needed; involve affected users; assess exclusion, accessibility, misuse, societal effects and contestability; pilot proportionately and record alternatives. Consider resource/energy use and environmental impact proportionately; distinguish measured outcomes from unsupported ESG claims.
Data protectionMap purposes and controller/processor roles; identify an applicable lawful basis and any additional conditions; minimise data; give appropriate notices; assess DPIA need, rights handling and transfer safeguards.
Marketing and trackingAssess UK GDPR and PECR requirements by channel, subscriber type and tracking technology; record eligibility and preferences; honour objections and suppression. Do not assume purchase permits all marketing.
SecurityMFA, least privilege, server-side authorisation, encryption, secret rotation, patching, tested backups and vendor due diligence.
AI-specific threatsPrompt injection, retrieval poisoning, model/data leakage, insecure outputs and tool abuse; treat retrieved text as untrusted; isolate execution and validate outputs.
Human authorityName who may approve publication, spending, refunds, access changes and consequential decisions. Provide override, appeal and emergency stop.
Risk assessmentRecord cause → event → harm, inherent likelihood/impact, control, owner, evidence, residual risk, treatment deadline and acceptance authority.
Assurance and auditTrace requirements to tests, approvals and operating evidence; record model/data versions and authorised actions. Protect logs while applying justified retention and access limits.
Government and sector controlsMaintain an applicability register for relevant data, consumer, payment, tax, equality and sector obligations; distinguish law, contractual duties and voluntary frameworks. Re-check territorial scope and commencement before launch.

This is an educational control framework, not a compliance certification. UK data-protection guidance is being updated following the Data (Use and Access) Act 2025. Confirm the current rules and relevant commencement provisions for the actual use case; do not infer that a generic template establishes compliance.

13 · Marketing, SEO & future salesMeasure acquisition and conversion before investing in predictive sales optimisation.

Open the synthetic visual dashboard / worked example →

AreaPractical specificationTrack
Market researchDefine customer problem, segment, alternatives, buying barriers and evidence quality.Interviews, qualified enquiries, willingness to pay and validation results.
PromotionChannel plan, approved content, audience eligibility, budget and campaign IDs.Spend, qualified leads, conversion, acquisition cost and payback.
SEOCrawlability, semantic pages, canonical URLs, titles, descriptions, structured data where appropriate and useful content.Search impressions, clicks, indexed pages, organic qualified enquiries and conversion; no ranking guarantees.
ConversionAccessible mobile GUI, clear offer, trustworthy checkout and measured experiments.Step conversion, abandonment, completion time and refund rate.
RetentionPermission-aware follow-up, support and appropriate product improvements.Repeat purchase, churn, cohort retention and customer value assumptions.
Predictive analyticsOnly after sufficient representative outcome data; benchmark incremental value.Calibration, drift, forecast error, human overrides and actual uplift from controlled experiments.

Predicted conversion is not observed revenue. Attribution is an estimate; consent, missing data and cross-device behaviour limit measurement. Track revenue and contribution margin per stream, rather than treating traffic as business success.

14 · Projects, transformation & milestone gatesDeliver incrementally; require evidence before extending scope, access or spend.

Open the synthetic visual dashboard / worked example →

Delivery gates require the control-test evidence and handover bundle.

Illustrative gated roadmap — stages may loop when evidence fails
  1. Discover → baseline + named owner
  2. Design → architecture + risk review
  3. Sandbox → synthetic-data prototype
  4. Pilot → agreed acceptance evidence
  5. Release → approval + rollback ready
  6. Operate → benefits + periodic review
GateAcceptance evidenceDecision
G0: discoveryProblem, scope/exclusions, venture register, success measures and budget ceiling.Sponsor approves a bounded test.
G1: designData flow, threat model, compliance applicability, interface contracts and ownership.Technical and risk reviewers accept design or require changes.
G2: sandboxSynthetic fixtures; payment signature, duplicate, denial and failure-path tests.Approve limited pilot access only.
G3: pilotUser acceptance, accessibility, security, data-quality and reconciliation results.Compare against pre-agreed thresholds; pause if unmet.
G4: releaseOperations owner, runbook, rollback, incident contacts, retention and handover evidence.Named release authority signs off.
G5: benefits reviewActual costs, adoption, outcome improvement, incidents and residual risk.Continue, improve, scale or retire.
Transformation tracking and project controls

Track scope, milestones, dependencies, accountable roles, risks, assumptions, issues, decisions, budget actual/forecast, change requests and evidence links. Separately track benefits: baseline → target → actual → measurement date → owner. Include adoption, training, time saved net of review, failure rate, service quality, accessibility, customer impact and operating cost.

Change requests record purpose, affected architecture/data, cost, delivery impact, new risk, approver and rollback. Procurement or payment milestones depend on accepted evidence, not a demonstration alone. Review unresolved defects and access expiry at every gate.

15 · Collaboration roles & delivery parametersAllocate responsibilities explicitly; retain ownership, evidence and controlled partner access.
RoleParametersAcceptance / boundaries
Delivery partnerOwns defined work packages under a statement of work, milestones and named delivery lead.Deliver code/configuration, tests, documentation and handover; scope changes need approval.
Joint business-development collaboratorJointly validates markets or develops opportunities under agreed contribution and commercial terms.Agree lead ownership, expenses, revenue attribution, permissions and conflicts; no implied equity or authority to bind.
Technical support partnerOperates within specified support hours, systems, severity levels and escalation routes.Agree response/restoration targets, approved access, incident records and termination handover.
Specialist reviewerProvides independent security, privacy, accounting or sector assessment as commissioned.Define independence, evidence, deliverable and limitations; cannot self-approve work where separation is required.
Venture owner / sponsorSets outcomes, budget and risk acceptance; appoints operational owners.Retains account ownership and final acceptance; tasks may be shared without implying a solo project.
Collaboration agreement checklist

Define parties, deliverables, exclusions, dependencies, RACI responsibilities, evidence-based acceptance, fees and expenses, IP/background IP and licences, confidentiality, data roles and processing terms where applicable, subcontractor approval, access restrictions, liability allocation, conflicts, dispute route, termination, export and revocation. Keep domains, payment accounts, code repositories and production credentials under the agreed owner’s control. Test competence through a bounded sandbox task before granting live access.

16 · Structured generic templatesUse consistent fields to compare proposals, implementation evidence and accountability.

See worked templates and responsibilities for an example of completing these records.

Download a blank structured pack, then require each prospective collaborator to complete the relevant records. Null fields mean “to be decided”, not “approved”. No payment, customer data or credentials belong in a public template.

Component proposal and comparison

component_id · purpose · owner · options_considered · build_buy_reason · setup_cost · recurring_cost · data_access · dependencies · acceptance_evidence · exit_route

Commercial validation and cash flow

customer · problem · demand_evidence · pricing_assumptions · volume_scenarios · direct_costs · platform_and_partner_costs · refunds_and_fees · settlement_timing · cash_flow · decision_gate

Customer journey and operating model

journey_stages · fulfilment_owner · exception_routes · accessibility_target · support_hours · operational_owner · backup_owner · maintenance · training · handover_tests

Continuity ownership and exit

asset_register · ip_and_licences · owner_admin_access · recovery_time_target · recovery_point_target · backup_schedule · restore_test · manual_fallback · export_format · migration_test · revocation · deletion_evidence · trademark_and_domain_checks · background_ip · new_ip_licence_or_assignment

Venture and income register

venture_id · name · stage · proposed_legal_entity · accountable_owner · customer_problem · income_streams · validation_evidence · gross_revenue_assumptions · direct_costs · platform_costs · margin_assumptions · next_gate · stream_ids · revenue_mechanism · owning_entity_id · capability_and_process_owner · system_of_record · shared_cost_allocation · portfolio_priority · enterprise_risk_dependencies

Payment and settlement specification

merchant_owner · gateway · currencies · products · checkout_flow · payment_states · webhook_verification · idempotency_strategy · refund_authority · dispute_process · payout_schedule · reconciliation_rules · test_cases · acceptance_owner

Security and privacy review

purpose · data_categories · controller_processor_roles · applicable_law_and_review_date · lawful_basis_assessment · dpia_screen · threats · access_roles · retention · transfers · vendor_checks · controls · residual_risk · approval · dpo_necessity_assessment · dpo_independence_and_conflicts

Workflow preset contract

workflow_id · version · owner · trigger_schema · conditions · allowed_actions · field_mapping · approval_rules · idempotency · timeouts · retry_policy · failure_queue · audit_fields · cost_limit · disable_switch · acceptance_tests

Integration and data contract

source · destination · system_of_record · auth_method · scopes · schema_version · mapping · pagination · rate_limits · event_ordering · error_handling · deletion_propagation · export_route · support_owner

AI engineering and model card

use_case · non_ai_baseline · intended_users · excluded_uses · dataset_provenance · splits · features_or_prompt · model_version · metrics_by_group · calibration · thresholds · human_escalation · monitoring · retraining_gate · rollback

Marketing and SEO plan

segment · offer · channel · eligibility_and_preferences · budget · campaign_id · content_approval · seo_checks · measurement_limits · conversion_goal · cost_per_acquisition · experiment · review_date

Project and benefits charter

sponsor · delivery_team · scope · exclusions · baseline · targets · budget_ceiling · milestones · dependencies · risk_register · change_control · acceptance_evidence · release_authority · benefits_owner · review_date · skills_gaps · learning_evidence · prototype_login_requirements · independent_reviewer

Partner statement of work

parties · role · deliverables · exclusions · milestones · acceptance_criteria · fees · ip_terms · data_roles · access_scope · subcontractors · support · handover · termination · revocation · cofounder_or_partner_status · equity_and_vesting_if_agreed · commission_rules · authority_to_bind · recognition_permissions

Risk control and audit register

risk_id · cause · event · harm · inherent_likelihood · inherent_impact · control · control_owner · evidence · residual_rating · treatment_due · acceptance_authority · next_review

17 · Commercial validation, economics & build decisionsValidate demand and sustainable economics before committing to system development.
AreaPlanning requirementsAcceptance evidence
Customer and demandDefine customer, problem, alternatives, willingness to pay, purchase barriers and the smallest useful pilot. Record assumptions separately from observed evidence.Interview or pilot results, qualified demand and a documented proceed/change/stop decision.
Income stream economicsRecord price, sales volume assumptions, direct costs, payment fees, refunds, partner fees and platform overhead. Distinguish revenue, contribution margin and cash received.Low/base/high scenarios and a cash-flow forecast with assumptions, settlement timing and owner.
Total cost of ownershipInclude setup, licences, connectors, compute, storage, support, security reviews, training, maintenance and exit. Set spend caps and alerts.Costed proposal and operating budget; review actual cost per completed business outcome.
Build versus buyCompare native configuration, supported extension, automation platform and bespoke development against the same requirements. Consider interoperability, skills, portability and maintenance.Decision record: options, costs, constraints, selected approach and review trigger.
AI suitabilityCheck whether rules solve the problem first. Assess available data, representative outcomes, error consequences and whether prediction adds measurable value.Baseline comparison, bounded experiment, explicit prohibited actions and named approval authority.
18 · Customer journey, accessibility & operational ownershipConnect acquisition to fulfilment, support and accessible ongoing service delivery.
AreaPlanning requirementsAcceptance evidence
Customer journeyMap discovery, enquiry, selection, checkout, confirmation, fulfilment, onboarding, support, cancellation and refund. Define state transitions and exceptions.End-to-end journey tests, including failed payment, unavailable stock and cancellation.
FulfilmentSpecify who releases goods or access, timing, inventory dependencies and evidence of delivery. Payment status alone does not prove fulfilment.Orders trace to fulfilment records and exceptions have an accountable owner.
Accessible GUIDesign keyboard navigation, meaningful labels, contrast, understandable errors, mobile layouts and appropriate language/timezone formats. Agree an accessibility target and assess with real users.Documented checks across agreed devices and assistive technologies; defects and remediation owner.
Service ownershipName the operational owner and backup, monitoring duties, support hours, escalation route, maintenance window and vendor contacts.Support rota or responsibility map, runbooks and tested escalation.
Training and adoptionProvide role-specific training, user guidance, administrator walkthroughs and accessible handover material. Measure whether people can complete real tasks independently.Recorded handover acceptance, task demonstrations, adoption measures and unresolved questions.
19 · Ownership, continuity, portability & partner exitRetain control and recover service when systems or relationships fail.
AreaPlanning requirementsAcceptance evidence
Asset ownershipAgree ownership of domains, payment accounts, repositories, cloud resources, data, code, configurations and documentation. Distinguish pre-existing IP from commissioned work.Asset register, agreed rights/licences, owner-controlled administration and recovery access.
Continuity and recoveryDefine tolerable downtime and data loss, backup schedule, restoration procedure, service dependencies and manual fallback. Test provider outage and connector failure scenarios.Restore exercise, recovery results, named incident lead and maintained runbook.
Safe manual operationDefine which actions may continue manually during an outage and how to reconcile them after recovery. Stop unsafe automation and prevent duplicate side effects.Demonstrated disable switch, manual records and safe replay/reconciliation test.
Portability and exitAgree usable export formats, schema documentation, migration assistance, licence constraints, retention/deletion evidence and termination timetable.Test export and re-import; revoke partner access and rotate affected secrets at exit.
Ongoing valueCompare actual benefits with all operating and checking costs. Review quality, customer impact, accessibility and incidents alongside time saved.Continue, improve, replace or retire decision supported by measured outcomes.
20 · Proposal review, acceptance & accountabilityRequire measurable evidence before accepting deliverables or extending partner authority.
AreaPlanning requirementsAcceptance evidence
Component reviewFor every component identify purpose, owner, setup and recurring cost, data access, dependencies, acceptance evidence and exit route.Completed component record; missing fields remain unresolved rather than implicitly approved.
Acceptance criteriaAgree realistic pass/fail thresholds before delivery: critical journeys, errors, security permissions, reconciliation, accessibility, performance and recovery.Requirement-to-test traceability, test results and named acceptance authority.
Milestone paymentsConnect any agreed payment milestone to defined deliverables and evidence; specify review period, remediation and change process.Accepted evidence and contractually agreed approval; a demo alone is insufficient.
Independent supervisionKeep the venture owner able to inspect accounts, logs, documentation and cost reports. Use specialist review where appropriate.Access demonstration, technical walkthrough, open-issues register and handover sign-off.
21 · How the categories relateSeparate business structure, technical capabilities, delivery work and continuing accountability.
CategoryWhat it isRelationship
Venture / income stream / entityA proposition / a revenue mechanism / a legal boundary.One venture can have several streams; several brands may share an entity. Record relationships rather than infer a company count.
Business capabilityA needed outcome such as collecting payment or supporting customers.Capabilities determine requirements; they should not be defined solely by a vendor catalogue.
Platform and applicationReusable infrastructure and software through which users perform tasks.CRM, ERP, data platforms and bespoke GUI components may support several capabilities.
Automation / AIExecution mechanisms: rules and orchestration, with optional interpretation or prediction.A workflow may contain no AI; AI can also provide advice without autonomously acting.
Project / programme / transformationBounded delivery / coordinated related projects / sustained changes to how work is done.A project produces a capability; transformation also requires adoption, roles, process change and measured benefits.
OperationsContinuing service ownership after release.Includes monitoring, incidents, costs, customer support, updates and retirement.
Governance, ethics and riskContinuing decision rights, scrutiny and accountability.Cross-cut delivery and operations; privacy and security require technical controls plus organisational ownership. They are distinct disciplines with overlapping responsibilities.

Relationship: venture outcome → capability requirement → platform/application design → delivery project → operating service → benefits review. Governance, security, privacy and risk review apply throughout, including before procurement and after launch.

22 · Workplace integration catalogue & connector contractsCompare integration patterns and declare permissions, failure handling and ownership.

These 16 integration categories describe possibilities, not installed connections or a recommendation to buy every platform. Availability depends on the specific product, edition, API and permissions. Salesforce/Einstein details remain in Data platforms.

Category / example purposeData or actionConnection contractTest / risk
Payments / bankingGateway events, checkout sessions, payouts and bank-feed reconciliation.Provider webhook or scheduled import; server-side secret; payment IDs and minimum order fields.Duplicate effects, refunds, bank-feed lag and settlement mismatch.
CRMEnquiries, qualified leads, opportunities and account activity.Scoped OAuth/API; map field ownership and deduplication key.Two-way conflicts, duplicates, access leakage and stale stage values.
ERP / procurementApproved purchase orders, inventory and supplier records.Documented API or batch import; approval before commitments.Stock races, incorrect units, supplier changes and duplicate orders.
AccountingInvoices, credit notes and reconciled ledger references.Restricted connector/service identity; accountant-agreed mappings.Tax mapping, closed periods, duplicate postings and rounding.
Email / marketingTransactional notices and separately eligible marketing.Scoped sending identity; preference/suppression checks per purpose.Mixed service/marketing content, unsubscribed contacts and retry spam.
Calendar / bookingAvailability, booking confirmations and cancellations.Scoped calendar API; explicit timezone and booking permissions.Double bookings, recurring-event changes and cancellation propagation.
Documents / e-signatureApproved templates, document generation and signature status.Scoped document access; authorised signatory and template version.Wrong recipient, unintended commitments and stale clauses.
Support / ticketingNew tickets, assignment, escalation and resolution.Webhook/poll; minimum necessary message content.Sensitive messages, mistaken closure and missed urgent escalation.
Tasks / project toolsMilestones, work assignments and accepted evidence.Scoped project API; correlate external IDs.Circular sync, orphan tasks and misleading completion status.
Team collaborationOperational alerts and approved summaries.Restricted channel scope and recipient allowlist.Personal-data leakage, public channels and alert fatigue.
Ecommerce / fulfilmentProduct catalogue, stock, shipment and returns.Order/stock API with item-level keys and state validation.Overselling, partial shipments and fulfilment before payment verification.
Analytics / advertisingEligible events, campaign costs and conversion aggregates.Purpose-aware collection, supported API or export.Unpermitted tracking, attribution gaps and duplicate conversions.
Warehouse / BICurated historical data and governed dashboards.Scheduled ingestion; dataset permissions and lineage.Stale data, incorrect joins and cross-venture exposure.
AI / retrievalAllowed document retrieval, classification and scoring.Model API or self-hosted endpoint; versioned inputs.Prompt injection, permission bypass, unsupported outputs and cost spikes.
Identity / securitySign-in, role changes, alerts and access revocation.Identity provider/API with minimal administrative privileges.Orphan accounts, privilege escalation and failed revocation.
HR / workplace recordsPermitted onboarding steps, training and access tasks.Strictly scoped records; distinguish administrative automation from high-impact decisions.Excessive employee monitoring, sensitive-data exposure and unfair inference.
PatternSuitable useTrade-off
Webhook / REST-hook triggerNear-real-time events from a supporting provider.Authenticate delivery; handle retries, duplicates, out-of-order events and subscription expiry.
Polling triggerProvider has no appropriate webhook or reconciliation requires periodic checks.Persist cursor/high-water mark; paginate; deduplicate; allow overlap for late changes; budget API quota.
Scheduled batch / file transferHistorical imports, settlement reconciliation and low-frequency reports.Validate schema, encoding, file authenticity, completeness, secure transfer and partial failure.
Request/response APIGUI needs an immediate answer such as a checkout URL.Timeouts, error states and authorisation; do not block the GUI on long jobs.
Event queue / orchestrationLong-running, multi-service work with recovery needs.Durable state, action-level idempotency and compensating actions; external side effects rarely form one atomic transaction.
RPA / browser automationA necessary legacy task has no supported interface.Fragile UI and credential dependencies; check permitted use, supervise and prefer a supported API where feasible.
Minimum connector contract and field ownership

Record provider/edition; connector owner; source and destination; purpose; trigger type; schema version; allowed operations; OAuth scopes; token storage and revocation; data fields; direction; system of record; deduplication key; paging/cursor; sync interval; conflict rule; quotas; retry/timeout; failure queue; retention/deletion propagation; test fixtures; costs; support and exit.

Example: CRM owns an opportunity stage; the payment ledger owns verified captured status; the bank feed owns observed settlement entries. A dashboard displays these states but cannot silently overwrite their authorities. Delete or suppress personal records in downstream systems according to the applicable purpose and retention rules; justified financial records may need a different retention treatment.

23 · Worked payment, automation & reconciliation exampleTrace one synthetic order through success, failure and recovery paths.

Illustration only: a £100 single-currency order, hypothetical £3 gateway fee and no tax, FX, reserves or timing differences. These are teaching assumptions, not a provider price or accounting policy.

StepSignal / authoritative stateAction and evidence
1. StartGUI sends product ID and quantity, not a trusted final price.Backend checks catalogue, price, inventory, venture and user permissions; creates order ORD-DEMO-001.
2. CheckoutBackend requests hosted checkout and stores provider reference.Return checkout URL to GUI; keep gateway secret server-side. Browser redirect is informational.
3. ReceiveProvider sends a payment event.Verify signature against raw bytes, timestamp and event type; durably record event ID. Reject invalid deliveries.
4. ApplyAuthoritative provider state indicates captured payment for the expected amount/currency/order.Use guarded state transition and unique effect key. Mark payment captured and enqueue fulfilment once.
5. ExecuteWorker reads durable fulfilment task.Re-check order eligibility; update inventory/access and send service receipt. Record per-action outcome and correlation ID.
6. RepeatSame event or an event arriving late is delivered.Deduplication and state guards prevent duplicate fulfilment. Reconcile authoritative state if event order is ambiguous.
7. SettleIllustrative payout contains £97 after the assumed £3 fee.Match £100 captured, £3 fee and £97 bank receipt using provider payout references. A real payout can aggregate many orders.
8. ExceptionRefund, dispute, failed action or unmatched settlement appears.Separate financial and fulfilment states; send exception to an owner; apply approved adjustment and safe replay. Do not erase prior evidence.
Generic event envelope — no card or credential data
{
  "event_id": "evt_demo_001",
  "event_type": "payment.captured",
  "schema_version": "1",
  "venture_id": "venture_demo",
  "order_id": "ORD-DEMO-001",
  "provider_payment_id": "pay_demo_001",
  "amount_minor": 10000,
  "currency": "GBP",
  "correlation_id": "corr_demo_001"
}

This is a generic teaching schema, not a Stripe or other provider payload. Use the provider’s actual event schema and signature protocol. Store money in integer minor units or a suitable decimal representation; do not assume all currencies use two decimal places.

Race conditions, partial failure and fraud

Test two concurrent deliveries, expired sessions, altered amounts, mixed currencies, partial refunds, access revocation, delayed payouts and a crash after an external action succeeds but before local acknowledgement. Use provider-supported idempotency and local effect records; query an uncertain external outcome before retrying. Require controlled approval for refunds, payout-account changes and unusual orders. Define dispute evidence and reconciliation responsibilities.

24 · ML computation workbench & complete decision pathInspect a synthetic prediction, its assumptions and its permitted consequences.

A fixed toy model illustrates inference. It is untrained teaching code, has no business validity and never sends data or performs actions. Its score is not a validated confidence estimate.

Signal stageExact operation in the illustrationEngineering / governance check
InputTwo numeric features: expressed interest x₁ and enquiry age in days x₂.Use only an assessed purpose and eligible data; missing or invalid inputs route to review.
ParametersFixed teaching weights: intercept −1, interest weight +2, age weight −0.05.A trained system would learn parameters from a versioned training set; these numbers were chosen for demonstration.
Linear scorez = −1 + 2x₁ − 0.05x₂. For 0.8 and 3: z = 0.45.Check feature order, units and transformation parity between training and serving.
Outputp = 1 / (1 + exp(−z)); the example yields approximately 0.611.A real score requires held-out evaluation and calibration; a precise decimal does not imply certainty.
Policy gateExample threshold 0.60 flags a lead for human consideration; otherwise normal review.Threshold is illustrative. It grants no permission to contact, spend, publish or make a consequential decision.
Action gateEligibility/preferences check → role/authority check → any required human approval → allowlisted action.Model output is untrusted input to the action layer. Log rejection and approval; provide an override.
FeedbackRecord later outcome, override and reason; compare distributions and errors over time.Do not automatically treat biased historical actions or unverified feedback as ground truth.
RetrainingNew approved dataset → fit candidate → evaluate → compare baseline → approve version → deploy/rollback.Feedback collection and inference do not automatically update weights. Assess purpose and access again when data or use changes.
Training mathematics and evaluation evidence

For binary labels y ∈ {0,1}, one common objective is mean binary cross-entropy: L = −mean[y log(p) + (1−y) log(1−p)], optionally with regularisation. An optimiser adjusts parameters to reduce training loss. Validation selects settings; a held-out test estimates performance on unseen cases. Minimise leakage, use representative samples and evaluate outcome-specific errors. Historical correlations do not prove that changing a feature causes sales.

Retain dataset lineage, consent/purpose assessments as applicable, sample counts, class balance, time/group split rationale, feature definitions, baseline results, precision/recall at the chosen threshold, calibration, subgroup performance where justified, latency, cost and failure tests. No universal sample size or confidence threshold is asserted here.

LLM agent signal path and small-model escalation

User task → identity/purpose checks → permitted context retrieval → prompt assembly → model inference → structured output validation → tool allowlist and permission checks → approval if required → execution → observation → bounded next step or stop. A model can propose actions but cannot confer authority on itself. Retrieved instructions are untrusted content, not policy.

Optional routing: a small model handles an evaluated class of routine decisions; an uncertainty/out-of-distribution rule routes difficult cases to a larger model or a person. Benchmark task success, harmful actions, latency and total cost against the original system. Claims in a research post do not establish savings for this website or a new use case; router calibration and repeated-step error propagation require testing.

25 · Control tests, metrics & evidence of readinessMake every critical claim testable before accepting delivery or scaling.

Open the synthetic visual dashboard / worked example →

Test / measureExpected evidence / pass conditionOwner / follow-up
Cross-venture isolationAn authorised user for venture A cannot read or modify venture B through GUI or direct API requests.Security reviewer; fix before pilot with real data.
Webhook authenticity / replayInvalid signature and stale delivery rejected; repeated valid event cannot duplicate business effects.Integration lead; retain synthetic test results.
Crash recoveryCrash at each workflow boundary; recover without silently losing work or duplicating irreversible effects.Operations lead; run safe replay exercise.
Settlement reconciliationKnown synthetic payments, fees, refunds and aggregated payouts reconcile; differences are visible and assigned.Finance owner; approve tolerance and exceptions.
Marketing suppressionWithdrawn or otherwise ineligible contact is excluded in every connected sending route; service/marketing purposes remain distinguishable.Marketing/privacy owner; test propagation and exceptions.
AI tool abuseAdversarial document/prompt cannot gain unauthorised access or execute a forbidden action.AI/security reviewer; validate tool permissions outside the model.
Model validityCandidate compared with baseline on locked held-out data; thresholds and unacceptable errors agreed before release.Model reviewer; reject if evidence is inadequate.
Data rights / retentionExercise retrieval, correction, deletion/suppression and downstream propagation for a synthetic person, respecting justified retention.Privacy owner; record exceptions and evidence.
Restore / supplier outageDemonstrate backup restoration, manual fallback, token revocation and alternative operating route.Operations owner; compare with agreed recovery targets.
Accessibility / usabilityComplete core journeys by keyboard and agreed assistive-technology/device checks; manual review alongside automated checks.Product owner; resolve material barriers.
Benefits / transformationBaseline, target, actual, sample/window, data source and owner recorded; count net time saved after review and rework.Benefits owner; review at agreed dates.
Handover / exitOwner can operate, inspect, export, recover and revoke partner access using supplied documentation.Sponsor accepts handover only after demonstration.
Operational metric dictionary and alert ownership
MetricDefinition / caution
Payment completionCaptured eligible orders ÷ started eligible checkouts over a stated window; distinguish customers, attempts and orders.
Contribution marginAgreed net sales less defined variable costs; document treatment of fees, refunds, shipping and tax.
Acquisition costAttributed campaign cost ÷ attributable new customers; disclose attribution limits and included costs.
Automation successCompleted intended outcomes ÷ eligible workflow runs; report manual intervention and failed side effects separately.
Service reliabilityAvailability, p95 latency, queue age, failed runs and recovery duration; set thresholds for actual customer needs.
AI qualityTask success, error severity, calibration, relevant subgroup error and human override; monitor data/concept drift without assuming drift alone proves harm.
Value and costCost per successfully completed outcome; net staff time saved; customer quality and complaint measures.
Risk and controlsOpen residual risks, overdue treatments, access reviews, incidents and control-test failures; do not equate absence of reported incidents with safety.

Each metric needs an owner, source, calculation, baseline, target, review interval, alert threshold, escalation route and known limitation. Agree numerical thresholds for the actual service; the generic guide cannot responsibly invent universal targets.

26 · Worked templates, responsibilities & evidence handoverSee how a blank record becomes a bounded, reviewable collaboration.

The 14 downloadable templates remain generic. The examples below show how to use them without presenting a proposed partnership as agreed.

Worked component fieldIllustrative value
Purpose / scopeDemonstrate one sandbox captured-payment event updating a synthetic order and creating a fulfilment task once.
Deliverables / exclusionsWebhook handler, event/effect ledger, queue worker, synthetic tests and runbook. Exclude live money, real customers and marketing.
Owner / delivery roleVenture sponsor approves scope; technical collaborator implements; reviewer assesses security; operator demonstrates recovery. Names remain to be appointed.
Data / accessSynthetic order records; provider sandbox keys in approved secret storage; no production bank or customer access.
Milestones / costsDesign accepted → sandbox evidence → failure tests → handover. Agree fixed or capped fees and recurring costs before work starts.
AcceptanceValid event processed once; invalid signature rejected; duplicate/crash recovery tested; cross-venture access denied; owner can run/export/revoke.
Exit / unresolvedDeliver source, configuration, dependency licences, setup notes and export; revoke credentials. Record support period, fee cap and responsible people as undecided until agreed.
Illustrative responsibility allocation — adapt and name actual people
Decision / workResponsibleAccountableConsulted / informed
Business scope and budgetProject leadVenture sponsorDelivery partner, finance and risk reviewers.
Implementation and test evidenceTechnical delivery partnerDelivery leadSecurity/privacy specialists and operational owner.
Risk assessment and treatmentRelevant control ownerNamed risk authorityAffected users and specialist reviewers.
Go-live decisionRelease coordinatorNamed release authorityOperations, finance, security/privacy and sponsor.
Support and incidentsSupport/operations partnerService ownerDelivery partner and relevant control owners.
Commercial lead developmentBusiness-development collaboratorCommercial ownerMarketing/privacy and venture sponsor.
Benefits and retirementBenefits/operations ownerVenture sponsorUsers, finance and risk owners.

A small team can hold several roles, but identify conflicts and independent review needs. Assign one clear accountable authority per decision; a job title alone does not create contractual authority.

Handover evidence bundle

Architecture and field mappings; asset/account register; source and configuration; dependency/licence inventory; data/model/prompt versions; test and acceptance evidence; risk/control records; operating and incident runbooks; backup/restore results; support contacts; training; cost report; export/migration demonstration; outstanding defects; access revocation and final ownership confirmation.

27 · Ten bounded project outlinesExplore ten practical projects with clear outcomes, teams and gates.

These are educational project briefs, not ten operating companies, active client commissions or a claim that all ten must be built. Each uses the same charter, control and acceptance templates. Prioritise P01 and P02, with privacy/governance designed alongside them, before expanding growth automation.

ProjectOutcome / scopeTeam rolesAcceptance evidenceBoundary / next gate
P01 · Payment-to-account demonstratorSynthetic order → hosted sandbox checkout → verified event → ledger → illustrative payout matching.Sponsor, integration lead, finance reviewer.Reject invalid event; fulfil once; reconcile fees and exceptions.No real funds; next gate requires merchant and finance review.
P02 · Identity and secure venture GUIRoles, MFA design, venture-level permissions and partner access expiry.Product lead, security reviewer, operator.Test unauthorised direct API reads/writes and account recovery.UI hiding is insufficient; production requires backend enforcement.
P03 · Privacy and governance workspaceData inventory, purposes, privacy review, risk controls and rights workflow.Controller representative, privacy lead/DPO where applicable.Trace each field to purpose, retention, recipient and rights route.Synthetic cases only; assess actual legal requirements before use.
P04 · Marketing and SEO evidence dashboardCampaign taxonomy, eligible analytics, technical SEO and qualified conversion.Marketing lead, product and privacy reviewers.Check suppression, attribution limits and documented calculation.No ranking or sales guarantee; next gate is bounded measurement.
P05 · Predictive sales workbenchPrepared synthetic records, leakage-safe split, baseline and calibrated candidate.Data/ML lead, domain reviewer.Compare locked test outcomes and errors; document decision threshold.No real targeting; next gate requires suitable data and purpose assessment.
P06 · Reusable intelligent-automation engineVersioned presets, filters, approvals, connectors and recoverable tasks.Automation lead and operational owner.Demonstrate deduplication, timeout, crash recovery and disable switch.Allowlisted sandbox actions; bounded pilot before wider access.
P07 · CRM/ERP integration pilotAgree field authorities and map one enquiry/order across a CRM and ERP mock.Integration lead, commercial and finance owners.Test conflicts, pagination, deletion propagation and reconciliation.Mock connector first; confirm real licence/API contract before purchase.
P08 · Full-stack support assistantGUI, authenticated API, permitted knowledge retrieval and proposed ticket action.Application/AI lead, support and security reviewers.Check source permissions, prompt injection, factuality and human escalation.No autonomous promises or sensitive decisions; next gate is supervised pilot.
P09 · Project, benefits and audit dashboardRequirements, milestones, budgets, risks, decisions and benefit evidence.Project lead, sponsor, assurance and benefits owners.Trace each acceptance criterion to dated evidence and decision.Planning records do not prove control effectiveness; review actual evidence.
P10 · Recovery, portability and handover exerciseBackup/restore, supplier outage, export/import and partner departure.Service owner, delivery partner and independent reviewer.Recover, reconcile manual work, import export and revoke access.Scale only when operating owner can demonstrate recovery independently.
28 · Demo → prototype → implementation → optimisationIncrease realism gradually while preserving clear evidence and access boundaries.

Working prototype: Open Manifold Process Lab · private 360° prototype ↗. Fifteen browser-local modules demonstrate connected records and controlled workflows. Sign in with the authorised ChatGPT account; internal roles are simulated, no external services are connected, and the prototype includes a detailed readiness/gap register.

Implementation maturity — review at each transition
  1. Educational demo · synthetic inputs
  2. Prototype · sandbox connections
  3. Pilot · limited approved users
  4. Implementation · approved production release
  5. Operate · support and control monitoring
  6. Refine · measured benefits and controlled changes
StageWhat users can doGate and required evidence
DemoRead expandable architecture, inspect payment scenarios and use the numerical ML workbench on this page.Clearly labelled synthetic behaviour; no real accounts, transfers or external action.
PrototypeUse an agreed GUI journey connected to mock or sandbox APIs and a test ledger.Authentication/authorisation design, synthetic test data, interface contract and failure tests.
PilotAn approved small group tests a bounded use case with explicitly permitted data.Privacy/security review, support owner, user acceptance, stop criteria and controlled permissions.
Production implementationAuthorised users access the deployed service through its verified identity provider.Secure sign-in and recovery, backend permission checks, accepted tests, incident/rollback procedures and named release approval.
RefinementReview actual cost, reliability, access barriers, errors, adoption and outcomes.Compare baseline and targets; approve changes, model updates and new integrations separately.
Login and “use it”: what is and is not connected

The existing venture workspace is a separate application. Its existence does not prove that payment gateways, CRM connectors or the projects in this guide are activated. This guide’s ML workbench requires no login and performs only local teaching calculations. It must never be presented as a production model.

For an eventual authenticated prototype specify: identity provider; email/account verification; MFA; invitation and role assignment; venture isolation; session expiry; account recovery; audit; partner expiry; export; deletion; and revocation. Test the backend rather than simulating security with a cosmetic login form.

Full-stack custom application boundary

Presentation layer: accessible screens and states. Application layer: validated APIs and domain rules. Data layer: transactional integrity, tenant isolation and retention. Integration layer: scoped adapters, queues and reconciliation. AI layer: permitted context, model inference and constrained actions. Delivery layer: builds, tests, environment separation, releases and rollback. Operations layer: monitoring, incidents, backups, costs and ownership. Configuration may satisfy a requirement; custom code needs an explicit maintenance and exit case.

29 · Business structure, tax readiness, DPO & intellectual propertyAssess specific obligations and ownership before launching or commissioning delivery.
AreaEducational planning questionsEvidence / source
Structure and remunerationCompare sole trader, partnership, company and social-enterprise structures against purpose, liability, decision rights and funding. A CIC is a specific regulated form, not a generic label for every social enterprise. Clarify how founders, workers and partners are paid.Structure decision record and appropriate advice; GOV.UK social-enterprise guidance.
Tax and administrative readinessIdentify the actual trading entity, tax registrations, records, invoicing, payroll/VAT considerations where applicable and filing responsibilities. A payment gateway does not fulfil these duties for the business.An accountant-reviewed applicability/calendar record; HMRC registration route. Check current duties for actual circumstances.
DPO versus privacy leadAssess whether a DPO is legally required. A privacy coordinator is not automatically a DPO. A DPO needs appropriate expertise, independence, resources and access to senior management; assess conflicts of interest.DPO-necessity assessment, role/independence record and escalation route; ICO DPO guidance.
Trade marks and domain namesCheck proposed brand, relevant goods/services and territories. Owning a domain or company name does not by itself establish comprehensive trade-mark clearance or protection.Search record, brand decision and any specialist clearance; UK trade-mark search.
IP and commissioned workDistinguish copyright, trade marks, designs, inventions, data rights and contractual confidentiality. Specify background IP, new work, assignments/licences, permitted reuse and open-source/dependency obligations.Asset/IP register and signed terms where needed; GOV.UK IP overview.
Government controls and industry dutiesMap applicable data, consumer, employment, equality, payment, product and sector obligations to actual territory, activity and commencement. Record mandatory duties separately from voluntary best-practice references.Obligation → responsible role → control → evidence → deadline → review date. Do not claim blanket compliance from a template.

This section is educational and does not choose an entity, tax position or DPO appointment for any proposed venture. No registration, professional appointment, trade-mark application or payment arrangement is represented as completed.

30 · Co-founding, contribution recognition & learning evidenceDistinguish commercial authority, paid work, attribution and independently demonstrated capability.
TopicParameters to agree / evidence to retain
Co-founder or partnershipDefine contributions, equity if any, vesting/exit if relevant, decision rights, signing authority, IP, conflicts, expenses, remuneration and disputes. A collaboration discussion creates no assumed ownership share.
Referral / commission / joint business developmentAgree qualified-lead definition, attribution window, rate or fixed payment, collection trigger, refunds/clawback, tax, reporting and end date. Keep referral fees separate from delivery charges and equity.
Delivery / technical supportUse bounded work packages, acceptance evidence, access limits, support period, incident duties and handover. A party may complement existing skills without taking control of the venture.
Paid work and experienceSpecify work, fee, payment schedule, supervision, permission to reference results and genuine acceptance. Attribution or exposure must not be silently substituted for an agreed fee.
Recognition and testimonialsRecord actual contribution and approved wording. Obtain permission before using a client/partner name, logo, testimonial or confidential result. A public comment is not automatically an endorsement or consent for reuse.
Paper / contract / portfolio evidenceSeparate research claims, contractual terms and verified project outcomes. Record authorship, contribution, dates, provenance, limitations and permission to publish; protect confidential material.
Close knowledge gapsUse a skills matrix: current evidence → gap → learning activity → practical task → reviewer → demonstration. Combine modular study, guided practice, mentoring and measured projects as appropriate.
Events, networks and opportunity checksTrack relevant events, organisers, eligibility, application tasks, cost, objective and follow-up. Keep personal education/career planning outside public technical evidence unless intentionally approved.
Learning outcome versus qualificationA demo, paper, course or testimonial evidences different things. Do not imply a degree, professional status, endorsement or accreditation without verification.

Project roles can be collaborative: project management and AI engineering are separate skills that may be held by different people. Record what each contributor actually did and what was independently reviewed.

31 · Sources, definitions & review boundariesDistinguish vendor capability, practical guidance, legal duties and illustrative design.

Source review: 26 October 2026. Architectural examples and milestone gates are proposed reference designs. Product editions, duties and interfaces require checking at implementation.

Terminology: GUI = graphical user interface; API = application programming interface; CRM = customer relationship management; ERP = enterprise resource planning; BI = business intelligence; DPIA = data protection impact assessment; RACI = responsible, accountable, consulted and informed; MLOps = machine-learning operations.

Learning check
  1. Why is a checkout success page insufficient evidence for fulfilment?
  2. Which workflow steps require rules, and which may benefit from prediction?
  3. What evidence distinguishes model training from inference?
  4. Who can accept residual risk and approve a release?

Answers: verify the provider’s authoritative payment state; use deterministic rules for known conditions and evaluate models for uncertainty; training changes parameters while inference applies a versioned model; the named authorities recorded in the project and governance registers.

Venture finance acceptance gate: costs, control and downside risk

Before accepting finance for an assurance, research, software or cultural enterprise, require a documented comparison of funding types. A loan, grant, procurement contract and equity investment each create different obligations. A headline award or investment figure is not net usable capital.

Minimum acceptance review

Record executed agreement and applicable law; principal and maximum payable amount; fixed or variable rate; indexation and inflation sensitivity; repayment and default schedule; company versus personal liability; guarantees and security; payment milestones and reimbursement; match funding; employee and subcontractor eligibility; IP rights and pre-existing research; investor reserved matters; publication independence; audit and retention duties; refund, termination and clawback; contingent liabilities; and a written affordability decision.

Cash-flow and counterfactual scenario checks

Test at least four cases: planned delivery; delayed funder or customer payment; project costs exceeding budget; and early termination or low demand. Separate founder income, business operating liquidity and restricted project funds. Calculate commitments using the signed terms, not promotional examples; retain an independent reviewer where appropriate.

Due-diligence table and authoritative agreement sources · Research funding and grant governance. No personal loan information is included in this public guidance.