AI Consulting
AI consulting that starts with your processes rather than a tool. A written view of what is worth doing, what it costs, and what to ignore.
What this is for
Most organisations arrive at AI from the wrong end. Someone senior has read that it is important, so a project is started, a tool is bought, and eighteen months later there is a chatbot nobody uses and a budget line nobody can justify.
The alternative is unglamorous. Start with what your organisation does every day, find the parts that are repetitive or slow or error-prone, and work out which of them a machine can genuinely improve. Some can. Many cannot, and knowing the difference in advance is the entire value of getting advice before building.
How the assessment runs
Talk to the people doing the work. Not only the leadership. The person who processes the invoices knows exactly where the time goes, and they are rarely asked.
Look at the data. What you hold, where it lives, whether the systems agree with each other, and whether the historical record is good enough to be worth anything. This is the most common reason a promising project dies, so it happens in week one rather than month four.
Then the arithmetic, per candidate. Current cost, build cost, running cost, payback period. Where we are uncertain we say so and give a range rather than a false number.
Then a written recommendation. Ranked, with reasoning, plus the things we recommend against and why. Priced and scoped, so it can be approved or refused as it stands.
What we have learned doing this since 2021
The wins are smaller and more boring than the pitch suggests. Reading documents. Triaging an inbox. Drafting a reply that a person then edits. Finding the pattern in a dataset somebody has been eyeballing for years. Individually unremarkable; collectively significant.
The failures cluster tightly too. Insufficient or contradictory data. A process so unstable that automating it means rebuilding it every month. And nobody internally who owns the outcome, which is the one that quietly kills more projects than any technical problem.
What we do not do
We do not sell a platform, and we take nothing from any vendor for recommending them, so the recommendation is not shaped by that. We do not produce a strategy document that stops at principles. And we will not tell you AI is going to transform your industry, because you have read that already and it does not tell you what to do on Monday.
How it works
- A fixed-scope assessment, and you keep the document
- Two to three weeks looking at how your organisation actually works: where the repetitive effort is, what data you hold and what state it is in, and which decisions are being made on guesswork. You get a written assessment ranked by payback. It is useful even if you then hire somebody else to build, and that is deliberate.
- Ranked by payback, not by how interesting it is
- Every candidate gets an estimate of what it costs now, what it costs to build, what it costs to run, and how long until those cross over. Most lists come back with two or three things clearly worth doing, several worth revisiting next year, and a few that should be dropped. Saying which ones to drop is the part most consulting avoids.
- Your data, assessed honestly
- Most stalled AI projects stall on data rather than on models: it is spread across systems that disagree, or it was never recorded, or it is recorded inconsistently enough to be unusable. We look at this early, because finding it out in month four is what turns a project into a write-off.
- The build is optional and we will say when it should be somebody else
- We can implement what we recommend, and often do. But an assessment that only ever concludes "hire us to build it" is a sales document. If the right answer is a product you can buy, or a process change with no software in it, that is what the document will say.
- Something your board can read
- The output is written for the people who have to approve it: what we recommend, what it costs, what could go wrong, and how you would know it worked. Not a slide deck of industry statistics, and not a list of vendors.
Where we have done this
What is and is not included
Included
- Interviews with the people doing the work, not only leadership
- An honest assessment of what state your data is in
- A payback calculation per candidate, with ranges where we are unsure
- A ranked written recommendation, priced and scoped
- The things we recommend against, and why
- The document is yours whether or not you build with us
Not included
- Implementation, which is quoted separately if you want it
- Software licences or model usage
- Change management and staff training programmes
- A vendor shortlist we are paid to produce, which we do not do
What we build it with
- Python
- SQL
- OpenAI
- Anthropic
- n8n
- PostgreSQL
Questions
The things people ask first.
What do we get at the end?
How long does it take?
Why start with data rather than tools?
Will the answer always be to hire you to build it?
Do you take commission from vendors?
What have you learned doing this since 2021?
Also in AI & automation
Tell us what you are trying to ship.
Start a conversation ↗