Approach
A method that ends in governed AI running in production.
Four phases, sequenced so that operating value and control arrive together. Scope is deliberately narrow at the start and widens only on evidence.
Diagnose
- Decision map
- Data reality
- Owner interviews
Model
- Entities & metrics
- Rules & thresholds
- Access & lineage
Deploy AI
- First use case
- Control design
Scale governed
- Production run
- Governed reuse
Phase one
6–12 weeks
Basis
Scope-dependent
The method
Four phases.
Each phase has a defined output. We do not begin the next one until the previous output holds up in your business.
- 01
Diagnose
Map the high-value decisions and where they break down.
We interview the people who make and execute the decision, trace the data behind it, and identify the bottleneck — missing information, unclear ownership, or delay between signal and response.
- Decision inventory with owners
- Bottleneck and data-gap map
- Prioritised first use case
- 02
Model
Connect source data and encode the operational logic.
We integrate the systems that the chosen decision depends on and encode entities, metrics, thresholds and rules into a model that both operators and machines can read.
- Source integrations in scope
- Semantic and operational model
- Quality checks and access rules
- 03
Deploy AI
Ship one governed AI use case with real users and controls.
The first use case goes into daily operational use — not a sandbox. Approvals, permissions and audit are in place before any agent acts on the business.
- Production AI workflow or copilot
- Approval and permission model
- Operator training and handover
- 04
Scale
Extend across workflows with governance intact.
Additional decisions reuse the model rather than rebuilding it. Governance, monitoring and change control extend with each addition instead of being retrofitted.
- Reuse pattern for new use cases
- Monitoring and review cadence
- Governance and change control
Typical first engagement
An indicative 6–12 week arc.
Indicative only and scope-dependent. Data access, system complexity and decision scope move these boundaries in either direction.
Discovery
Decisions, data reality, scope agreement.
Prototype
Working model against real data, reviewed with operators.
Pilot
Limited production use with defined users and controls.
Operationalisation
Handover, monitoring, governance and extension plan.
Timeline shown as guidance for planning conversations. It is not a commitment or a quoted delivery schedule.
Principles
Four rules we do not trade away.
Start with decisions
Scope begins with a decision that carries consequence, not with a system, dataset or model.
Prove value early
Something real must be in operators' hands within the first engagement, not at the end of a programme.
Design with operators
The people who run the process shape the workflow. Adoption is a design input, not a change-management afterthought.
Govern from day one
Permissions, approvals and audit are built with the first use case, because retrofitting them stops deployments.
Risk and control architecture
What governs an AWF deployment.
Every automated capability we deploy sits inside these six controls.
Access
Role-scoped permissions enforced at the model layer.
Authority
Explicit limits on what an automated step may decide or change.
Approval
Human sign-off required on consequential actions.
Audit
Inputs, versions, approvers and outcomes recorded and reviewable.
Reversibility
Automated actions can be traced and unwound.
Monitoring
Drift, exceptions and failures surface to named owners.
Phase one starts with a conversation.
Thirty minutes to frame the problem properly is usually enough to tell whether there is work worth doing.