How we work
From diagnosis to adoption you can measure
The same five phases as on the front page — and below them, the engineering discipline that makes each one checkable.
- 1Align and diagnose
Leadership agrees the value at stake; processes, the AI already in use and readiness are mapped on evidence.
Measured by: a signed value floor and a readiness baseline
The engine plus hands-on work
Spine stages: Align · Discover · Diagnose
- 2Redesign the work
The workflow is rebuilt around the AI: who decides, where a human reviews, how the team is sized.
Measured by: decision rights signed by their owners
The engine plus hands-on work
Spine stages: Prioritize · Design
- 3Prove it in a governed pilot
One high-value workflow, with EU AI Act classification, evidence and an audit trail built in.
Measured by: cycle time and value against the floor
The engine plus hands-on work
Spine stages: Govern · Pilot · Deploy · Prove
- 4Bring people along
Role-based learning, a champions network and team libraries of prompts and agents — so the new way of working is the one people use.
Measured by: adoption per workflow
Hands-on with your teams
No engine stage yet — delivered hands-on with your teams.
- 5Scale what works
Wins become standard modules with owners, an operating rhythm and a value ledger.
Measured by: realised value, not seat licences
The engine plus hands-on work
Spine stages: Scale · Learn
We take AI out of pilots and into the daily work of regulated organisations. That means redesigning the workflow the AI belongs in, agreeing who decides and who reviews, and bringing the people who do the work along. It also means governing it so it can scale, with each step ending in a record its owner signs.
Riaan has built AI products himself, voice and identity verification among them, so we know the technical side from the inside. But what decides whether AI is adopted is rarely the model; it is the operating environment around it. That is where we work.
Honesty proof VERIFIED on 2026-10-15 — 5 checks (typecheck, lint, tests, i18n parity, Archify doctor).
Not that we know AI — that we know how to make AI adoptable inside real organisations.
“Many AI projects fail because the model works but the operating environment does not. Our work focuses on the operating environment: identity, data, workflow, permissions, review, integration, deployment, and measurable acceptance.”
The engine behind this site is that way of working, made visible: every session ends in facts and signatures, every number has a source, and what is unknown is shown as unknown. It is how we work — not a licence we sell.
What an engagement delivers: AI your organisation uses, owns and can measure — not another pilot.
Predictable scaling — scoped right, priced right, delivered right.
Most organisations use AI somewhere; few have scaled it. What is missing is rarely the model — it is the operating model: how the work is sized, priced and signed. Cost comes down as a result of getting that right, not as the goal.
- 1Scoped right
Each kind of work is sized by what it grows with: makers by the work, thinkers by the initiatives they hold at once, leads in whole people. No over-staffing, no under-staffing — and it says so when the answer is 'add'.
See the same work counted both ways → - 2Priced right
Priced on the outcome, not the seats, with the margin set before signature and every assumption named. A price that holds as volume grows, because each part of the team has its own driver.
See the offers → - 3Delivered right
Nothing moves on without the gate and the signature that owns it — what an agent may do, what a rule decides, what a person signs. The promise made in the room holds after it.
See who signs what →
Underneath
The engineering discipline underneath
For the technical reader: the eleven-stage delivery spine, the architecture every engagement shares and the six gates — each listing the evidence held today.
- 1AlignSpecify · GovernNow
- 2DiscoverSpecify · MapLater
- 3DiagnoseSpecify · MapLater
- 4PrioritizeArchitect · MapLater
- 5DesignBuild · MeasureLaterThree laws of sizing →Today vs the laws →
- 6GovernBuild · Governdelivery
- 7PilotShip · Measuredelivery
- 8DeployShip · Managedelivery
- 9ProveReview · MeasureLater
- 10ScaleOperate and improve · ManageLater
- 11LearnOperate and improve · Governdelivery
- Business outcome
- Users & stakeholders
- Current process
- Target workflow
- Scope
- Non-functional requirements
- Data requirements
- Risk classification
- Acceptance criteria
- Commercial assumptions
- System boundaries
- Agent roles
- Data flows
- APIs
- Identity
- Permissions
- RAG / knowledge sources
- Human decision points
- Failure modes
- Monitoring
- Deployment model
- Security controls
- Core workflows
- Integrations
- Agent tools
- Permissions
- User experience
- Logging
- Test harness
- Configuration
- Documentation
- Functional requirements
- Security
- Data handling
- Prompt & model behaviour
- Failure cases
- Human escalation
- Cost & latency
- User acceptance
- Operational readiness
- On the customer's infrastructure
- Approved keys & environments
- With monitoring
- With rollback
- With runbooks
- Ownership transferred
- Adoption
- Task completion
- Accuracy
- Exceptions
- Cost
- Latency
- Incidents
- Business value
- Improvement opportunities
Stage 3 · Model → Stage 4 · Govern
Then the controls, their owners, and the evidence that proves they hold.