01 · Fine Kilter
Fix the work that slows your business down.
Fine Kilter is a one-person consultancy that rebuilds slow, inconsistent or person-dependent work, using AI systems your own people run and keep.
02 · Sound familiar?
Sound familiar?
- The same jobs eat senior time every single week.
- Customer replies are slow, or their quality depends on who writes them.
- There's a backlog, but not enough of one to justify another hire.
- Too much of how the business runs lives in one person's head.
- Your team already uses AI tools quietly, with no rules around any of it.
That's the work I fix.
03 · What every engagement builds
What every engagement builds
Before anything gets automated, the work has to be written down properly. That is the first thing I build, and it is the part you keep.
Think of it as a user manual for how one piece of your business actually runs. Not the tidy version anyone would write for an audit. The real one:
- The records you work from, and where they actually live.
- The rules you apply, including the ones nobody has ever written down.
- The judgement calls your people make, and what those calls are based on.
- The exceptions everyone knows about and works around.
Each part is kept separate, so any one of them can be read, checked or changed on its own without disturbing the rest. That matters more than it sounds: it is the difference between a document that goes stale in a month and one you can keep accurate.
This manual does not sit in a drawer. It is what the system reads, every time it does a piece of work, and it reads only the parts that matter for the job in front of it.
That is also why the rest of it works. Because your records and rules are written down and properly separated, the system runs against how your business actually operates, rather than against general knowledge about businesses that look a bit like yours.
And when too much of how the business runs lives in one person's head, this is the part that gets it out.
04 · How it behaves
How it behaves
The AI does the heavy preparation. It drafts, checks and compares, working from the records and rules above. Then it stops.
One of your people reviews the result, sees exactly what was checked, and makes the call. Nothing goes to a customer until someone in your business approves it. That is not a safety feature bolted on at the end. It is how these systems are designed from the start.
And when the evidence behind an answer is thin, the system is built to say so, not to sound confident anyway.
05 · Where the same approach fits
Where the same approach fits
The approach fits wherever a judgement gets made repeatedly, against evidence that already exists somewhere in the business. Three examples of how that applies:
Preparing a quote or an estimate
The pricing history, the previous jobs, the rules about what gets discounted and when. The system pulls the comparable work, drafts the numbers, shows you what it based them on. Your person prices it.
The exception that always escalates to the same person
The awkward one that only one member of staff knows how to handle. The reasoning gets written down, the system prepares the case the same way every time, and the decision still belongs to a person.
A compliance or verification check
The documents, the rules, the things that must be true before something proceeds. The system checks against your actual requirements and flags what does not hold up.
One workflow at a time, properly. That is how the work is bought, and it is deliberate. The approach is broad, the job is narrow, and those are two different things: the reach of a method, and the scope of a piece of work.
FIG. 01 · One method, applied three times
the shared spine06 · What it looks like when it runs
What it looks like when it runs
A system built this way doesn't produce an answer and forget it. Five things are true of every piece of work it does.
- It picks up where it left off. It keeps its own working notes, so a job starts from what happened before rather than from nothing.
- It reasons from what it has already seen. When it recommends something, it shows the earlier cases and the patterns it drew on. Those come from your business's own history and nobody else's.
- It shows the trail. Any claim it makes can be walked backwards: the pattern, then the cases behind it, then the rule underneath.
- It won't invent numbers. Ask it for a figure the evidence can't support and it says so, then tells you what the evidence does support.
- It writes down what it learned. That record is what makes the next job start further on than the last one did.
That last point is the whole shape of it. Each pass leaves something behind for the next, which is the difference between a system and a tool you have to brief from scratch every time.
FIG. 02 · What it looks like when it runs
design commitment, every build07 · What makes Fine Kilter different
What makes Fine Kilter different
You own what gets built
The records, the rules, the system: yours. Documented, exportable, with a proper handover.
If we stop working together, nothing has to be rebuilt. The records, the rules and the system are yours, and they keep working. That is the point of building it this way round.
The models underneath are swappable, so you are never welded to one vendor's tool.
I'll probably talk you out of buying AI tools
There is a difference between the models themselves and the products built on top of them. The models are genuinely capable, and they are interchangeable. What usually goes wrong is the layer in between: a business buys someone else's finished product, built for nobody in particular, carrying features it does not need and missing the context that would make it useful. My honest view is that most small businesses do not need another subscription. They need one workflow fixed properly, with the right system around whatever model does the work. If a tool you already have can do the job, I'll say so.
Honest limits are part of the product
A system that flags weak evidence beats one that bluffs. This site works the same way: what I won't promise is written down, plainly.
08 · Why believe any of this?
Why believe any of this?
Fine Kilter is new, and I won't pretend otherwise. No client logos, no case studies yet. Here's what there is instead:
- Real operating experience. In my day job I design, build and run governed AI systems inside a real UK business, with the outcomes reported at board level. Fine Kilter is my own practice, separate from my employer; nothing here trades on their name or their clients.
- A published method. How I build, what gets recorded, and who stays in charge, written down before you commit to a thing. How I work
- A worked example, in full. One judgement-heavy task taken end to end, built the same way as the systems I operate, with the reasoning shown at each step. This one happens to be a customer complaint email, which is one example among many. Where this fits
When there's real client evidence, it will be here. Until then, you get the method, the worked example and a straight conversation. About Stuart
Same email, same few minutes. One answer is a guess in good English; the other is your business answering, faster. The full walk-through
09 · How an engagement starts
How an engagement starts
- A conversation. You tell me where the friction is. I'll tell you honestly whether this approach fits, and I'll tell you if it doesn't.
- Something real. I build something small on your kind of problem and show it working, so you're judging evidence, not promises.
- Your call. If it's useful, we talk about the proper build. If not, you've spent an hour and learned something. Larger options exist after discovery, but nothing here starts with a big project.
Start the conversation
Not ready to talk? Follow along on LinkedIn.