Our approach
Business value first. AI second.
We do not recommend AI because it is new or fashionable. We start with your business, quantify where AI could create value, prioritise on evidence, and only then choose the technology. The same eight stages run from the first conversation to the measured result.
- 01DiscoverUnderstand the business
- 02ValueQuantify the opportunities
- 03PrioritiseScore and rank
- 04RoadmapSequence the work
- 05ProveTest the risky assumptions
- 06BuildEngineer for production
- 07AdoptEmbed in real work
- 08MeasureReport and improve
The eight stages
Each stage answers a question. Each produces a decision.
- 01
Discover
We start with your strategy, workflows, pain points, data and constraints. No technology conversations yet.
- Where does effort, delay or error concentrate?
- Which decisions are made with incomplete information?
- 02
Value
Candidate use cases are identified and their expected impact on revenue, cost, productivity, experience, decision quality and risk is estimated.
- What is the mechanism by which value is created?
- What would we measure to know it worked?
- 03
Prioritise
Each opportunity is scored on value, complexity, data readiness, feasibility, organisational readiness, risk and time to value, then positioned on the AI Value Matrix.
- Which opportunities justify a business case?
- Which should we deliberately park?
- 04
Roadmap
A Now, Next, Later roadmap with dependencies, data and platform foundations, governance, skills, investment and success metrics.
- What must exist before the high-value plays are possible?
- What do we build, buy or configure?
- 05
Prove
Where risk is material, a scoped prototype on real data answers the specific questions standing between you and a production decision.
- Does it perform acceptably on our data?
- What does it cost at volume?
- 06
Build
Production systems built with the right technology for the problem, with evaluation, security, observability and cost controls designed in.
- How do we know quality has not drifted?
- Who owns this after launch?
- 07
Adopt
AI is integrated into the tools and workflows people already use, with training, support and the removal of friction that causes workarounds.
- Are people actually using it?
- Where are they routing around it, and why?
- 08
Measure
Outcomes are measured against the original business case and reported honestly. Evaluation data, feedback and cost drive continuous improvement.
- Did we get what the business case promised?
- What should change next quarter?
AI Value Matrix
Seven criteria. One picture.
Every candidate opportunity is scored against the same criteria and positioned on the matrix. The picture makes trade-offs visible and turns a list of ideas into a defensible sequence.
- Expected business value
- Revenue, cost, productivity, experience, decision quality or risk reduction, quantified where possible.
- Implementation complexity
- Integration surface, workflow change, number of systems and teams involved.
- Data readiness
- Availability, quality, access rights and freshness of the data the use case depends on.
- Technical feasibility
- Whether current techniques can meet the required accuracy, latency and cost.
- Organisational readiness
- Sponsorship, skills, appetite for change and capacity to absorb it.
- Risk
- Privacy, security, regulatory, reputational and operational exposure.
- Time to value
- How quickly measurable benefit can be realised.
Technology decision framework
Problem first. Technology last.
The newest or largest model is not automatically the right answer. Sometimes the fit is a frontier model. Often it is a smaller model, classical machine learning, retrieval, automation, an existing product, deterministic software, or a combination.
How we decide
- 1Business problem
A specific, measurable pain point owned by someone.
- 2Requirements
Accuracy, latency, volume, explainability, integration.
- 3Constraints
Budget, data residency, privacy, skills, existing platforms.
- 4Technology selection
Frontier LLM, small model, classical ML, RAG, agents, automation, SaaS, deterministic software, or a combination.
What we avoid
- 1New model released
Impressive demo, unclear fit.
- 2Search for a problem
Use cases retrofitted to the technology.
- 3Pilot without a pass mark
No agreed success test.
- 4Stalled pilot
Never reaches production or measurable value.
AI value funnel
Most opportunities should not make it to production.
That is the point. Discovery surfaces many candidates. Scoring, validation and business cases narrow them to the few worth building, and measurement proves what they delivered.
- Potential opportunitiesEverything surfaced in discovery
- Viable opportunitiesFeasible with available data and technology
- Prioritised opportunitiesScored highest on value versus complexity
- Validated opportunitiesBusiness case approved, risks tested
- Production solutionsBuilt, integrated and adopted
- Measured outcomesReported against the business case
Principles
What you can hold us to.
Business value first. AI second.
We do not recommend AI because it is new or fashionable. We recommend it when it is the best way to solve a problem that matters.
Right model. Right use case. Right economics.
The newest or largest model is not automatically the best answer. Selection is driven by accuracy, cost, latency, risk and fit for the task.
Technology-agnostic by design.
We work across OpenAI, Anthropic, Google, Microsoft, AWS, open-weight models and conventional machine learning. We are paid for outcomes, not for recommending a platform.
Advice you can act on, and the capability to act.
We build what we recommend. That keeps our advice honest, because we will have to deliver it.
Measure honestly.
Outcomes are reported against the original business case, including where they fall short.
See the method applied to your organisation.
A discovery conversation is the first stage. It costs nothing and usually surfaces two or three opportunities worth a closer look.