Approach
Understand reality first. Then make change that pays back.
One discipline runs through every engagement, whether the answer is AI, a redesigned journey, a new operating model or a product. When AI is in scope, four tools keep it honest: the path, the value framework, five patterns and stage gates.
How I work · Every engagement
Discover, design, deliver
Understand reality first. Create shared clarity second. Add structure only where it improves the outcome.
1 · Discover
See it for yourself
Use the product, watch customers and spend time in the operation. Ask the people closest to the work what is blocking them, then map the whole system.
- Current-state journey and system map
- Problem and opportunity statements
- A maturity baseline
2 · Design
Agree one direction
Agree what good looks like. Make the trade-offs on desirability, feasibility, viability, risk and time to value, and choose the right mix of process, data, product and AI.
- Target journeys and prototypes
- A prioritised roadmap and business case
- Decision rights and clear ownership
3 · Deliver
Make it stick and pay back
Build capability as we go, so teams own the answer. Measure adoption, customer impact and commercial value, and adjust when the evidence changes.
- Delivery with measured adoption and value
- Benefits reported with Finance
- Coached teams and a clean handover
When AI is in scope · The path, three phases in sequence
Scale AI before the data is trusted and you automate the wrong thing faster
Make it trustworthy
Named owners, reliable priority data, agreed metrics and visible run costs.
Move on whenPriority data is owned and measured, and controls are in place.
Make it repeatable
Proven patterns, shared data products and one front door replace bespoke builds.
Move on whenNew use cases reuse patterns and land faster than the last.
Make it pay
Assistants, automation and decision support expand where the scorecard proves value.
Move on whenBenefits are reported with Finance and investment follows evidence.
Across every phaseSecurity, access control, documentation, monitoring and cost governance are built in by default. Investment expands only when a use case’s scorecard proves value.
Value framework · Improve, grow, transform
One outcome, one baseline, one owner for every change
Seven outcomes, each felt somewhere in a customer or colleague journey, whether the change is AI, process, data or product. Every initiative has one primary outcome, a leading measure that moves first and a value measure that proves it. Benefits are counted once.
Efficiency and capacity
Less effort per case, so colleagues spend time where customers notice it.
Leading · ValueAdoption; cases handled · cost per case; backlog
Accuracy and consistency
Fewer errors and omissions, so the customer gets it right first time.
Leading · ValueAutomation success rate · error rate; rework
Speed and responsiveness
Less waiting and fewer handoffs between a request and an answer.
Leading · ValueCases auto-triaged · response and resolution time
Customer experience
Needs resolved with less customer effort, in B2C, B2B and through partners.
Leading · ValueAnswers rated helpful · repeat contact; effort
Revenue and margin
Better conversion, retention and contribution from better-timed decisions.
Leading · ValueRecommendations used · conversion; retention; margin
Decision quality
Better evidence behind forecasts, supplier choices and investment decisions.
Leading · ValueForecasts used in decisions · forecast error; outcomes
Scale and new services
Serving more customers, or offering new services, without cost rising in step.
Leading · ValueCustomers served · cost to serve; new revenue
The test: if people don’t use it, it’s theatre. Adoption is measured alongside value.
When AI is in scope · Five patterns
Use the least autonomous pattern that does the job
You don’t need a new architecture for every use case. Five patterns cover most of the opportunity. Autonomy is a cost: it needs more controls, more testing and more trust, so it has to be earned.
1 · Assistant
Copilot
Choose it whenA person directs the work and reviews every output.
ExampleDraft a proposal from a brief.
2 · Knowledge assistant
RAG (retrieval-augmented generation)
Choose it whenThe answer depends on current company knowledge.
ExampleAnswer a service question using current policies, with sources.
3 · AI-enabled workflow
Agentic workflow
Choose it whenThe process is defined, repeated and rule-bound.
ExampleRead an enquiry, extract requirements, check completeness and prepare a quote.
4 · Bounded agent
AI agent with guardrails
Choose it whenThe path varies case by case across several systems.
ExampleInvestigate a delayed order across systems and prepare a resolution.
5 · Prediction and optimisation
Predictive and prescriptive analytics
Choose it whenThe decision is a forecast, a score or an optimisation.
ExampleForecast demand and recommend replenishment.
First question for every use case: can tools you already license do the job? Use, then configure, then build.
Delivery method · Stage gates
Five stages, five decisions, including when to stop
1 Assess
What should improve?
Use the process, watch customers and colleagues at work and map the system. Baseline the measures.
Proceed or stop
2 Select
What merits investment?
Compare options, including tools you already have. Agree benefits, pattern, ownership and delivery needs.
Invest, defer or stop
3 Build
Does it work reliably?
Lead specialists through build, integration and testing against agreed quality and control criteria.
Release, rework or stop
4 Adopt
Ready to operate?
Redesign the work, train users and set up support, monitoring and exception ownership.
Accept into service
5 Measure
Is value materialising?
Review benefits and running costs with Finance against the baseline.
Expand, improve or retire
Nothing goes live until it passes its release gate: results on an agreed test set, data protection, regulatory classification, cost per transaction, and a fallback if the model or vendor fails.