Approach

How an engagement actually runs

Five stages, each producing something you can act on — so you can stop, change direction or take the work in-house at any point without having lost anything.

The stages

Discover, diagnose, design, implement, measure.

What each stage leaves you with Swipe to see it all

Every stage produces something you can act on, so you can stop, change direction, or take the work in-house without having lost anything.

01

Discover

I learn how the business actually operates: the workflow, the tools, who touches what, where things wait, and what people have quietly built workarounds for.

Process mapping System inventory Stakeholder sessions

02

Diagnose

I separate the interesting ideas from the ones that would create measurable value — weighing volume, time cost, error rate, data quality, and how much would genuinely change.

Opportunity map Impact vs effort Data readiness

03

Design

The architecture: what gets automated, what gets a model, what gets integrated, where a human stays in the loop, and how the whole thing fails safely when something goes wrong.

Solution architecture Model & tooling choice Guardrails

04

Implement

Building and shipping it — integrations, agents, retrieval pipelines, workflow automation and the interfaces around them — deployed into your environment, not a demo sandbox.

Build & integrate Deploy Handover & documentation

05

Measure

Instrumenting it so you can see whether it worked: time saved, volume handled, error rate, answer quality. If a piece is not earning its keep, we cut it rather than defend it.

Instrumentation Quality evaluation ROI review

The decision

How I decide whether something should be AI at all.

Every opportunity gets sorted into one of five buckets before anyone writes code. Most of what businesses call an AI problem turns out to belong in one of the other four — and the ones that genuinely belong in the AI bucket are much easier to justify once the rest have been cleared out of the way.

01Automate — The steps are known and repeatable. A workflow, a job or a rule does it — cheaply, predictably, and without a model to evaluate.

02Integrate — The information already exists in another system. Somebody is moving it by hand because nobody has built the connection.

03Redesign — The process itself is the problem. Automating it faithfully would just make a bad shape run faster.

04Fix the data — The inputs are incomplete, duplicated or untrusted. Everything downstream depends on sorting this out first.

05Use AI — Judgement is genuinely required on unstructured input, and the rules are too many or too fuzzy to write down. This is where a model earns its cost.

How I work

A few things I hold to.

01

Smallest useful thing first

The first release should be narrow enough to ship quickly and real enough to prove the idea. Big-bang AI programmes are how budgets get spent before anybody learns anything.

02

A human where it matters

Full autonomy is rarely the goal. I build checkpoints into the places where a wrong decision would be expensive, and let the system run unattended everywhere else.

03

Measured, not assumed

Retrieval quality, answer accuracy, time saved, error rates. If we cannot see whether it is working, we are guessing — and guessing is how AI projects quietly become shelfware.

04

Built to be handed over

Documented, conventional, and running in your environment on your accounts. You should be able to take it in-house or hand it to another engineer without a rescue project.

05

Boring where boring works

A cron job, a queue and a well-named database table solve more business problems than most AI architectures. I reach for the model when the problem actually needs one.

06

Your data stays yours

Model and hosting choices get made with your data-handling constraints on the table from the first conversation, not retrofitted when procurement asks.

The offer

A free AI & automation opportunity review.

You do not need another generic AI strategy document. I look at how your business currently works and come back with practical opportunities where AI, automation, integration or better architecture would reduce manual work and create measurable value — including the ones I would tell you not to bother with.

Start your review

Two minutes to request. The written map is yours whether or not you hire me, and there is no follow-up sequence.

01UnderstandYour current workflow, tools, systems and where things get stuck.
02IdentifyRepetitive work, manual handoffs, disconnected systems, AI-shaped problems.
03PrioritiseSeparating interesting ideas from the things that would move the business.
04RecommendPractical opportunities ranked by impact, complexity and likely return.
05ImplementIf you want it built, I design and build it. If not, the map is still yours.