AI
Applied AI
Used where it moves a number, not a slide.
The automotive industry has no shortage of AI pilots. It has a shortage of AI that took cost out of a process somebody runs every day.
We work on the second kind. That usually means less model and more plumbing: the data has to be reachable, the output has to land in a system somebody already uses, and somebody has to own the number afterwards.
What we work on
Business development
Finding and qualifying the commercial openings worth a call (partners, importers, accounts), with agents doing the scanning and the first pass, so your team only sees what survived it.
Demand and stock
Forecasting that reaches the ordering decision, with the error bars visible so nobody trusts it further than they should.
Document-heavy process
Contracts, homologation packs, warranty claims, invoices. The work that scales with headcount today and does not have to.
AI diligence
For investors: what a target's data and model stack is actually worth, and how much of the claimed advantage is a subscription somebody else could buy tomorrow.
04
How a mandate runs
Two tempi, in this order. First we map and cut: most of what looks like an AI problem is a process problem, and cutting costs nothing. Then we build on what survives: agents where the work repeats, models where there is a number to move. The technology comes last, and sometimes it does not come at all.
- 01
Listen
What you need and where it costs you, asked of the people who do the work rather than read off the org chart.
- 02
Map the as-is
The process as it actually runs, step by step, with the time and the cost against each step. Most of the saving is visible here, before anything is built.
- 03
Optimise
Cut and rationalise first: what can stop, what can merge, what can be simplified. What survives that is what is worth automating.
- 04
Agents and models
Agents where the work repeats, models where a number has to move. Both go inside the system the work already happens in, with an owner and a number to judge them by.
When people call us
- Cost has been taken out of every line except the processes nobody has ever written down.
- A back office grows headcount every time volume grows.
- An investment committee needs an AI claim checked before it is paid for.
What you get
- 01The as-is process mapped, with the time and the cost against every step
- 02What to stop doing, costed, before anything is automated
- 03The agents and models running inside the system the work already happens in
- 04An honest verdict when the answer is that you do not need AI for this
Questions to ask before approving an AI project
- How much can AI save in a car dealership's back office?
There is no industry percentage, and a saving figure given before anyone has seen how the dealership works is invented. The number is the dealer's own, and it can be counted in house: the people it takes to process a month of registration paperwork, the hours a warranty claim waits before reaching the manufacturer, how often a quotation is retyped into a second system, and how many supplier invoices are checked by eye.
The AI project you were pitched is a process problem- Why do AI projects in car dealerships fail?
Almost never on the model. They fail earlier, because the process they were meant to lighten had never been written down and nobody knew what it cost. AI laid over a process nobody has mapped automates the steps that should have been removed. The data also has to be reachable, and the output has to land in a system somebody already uses, or the work has been moved rather than removed.
The AI project you were pitched is a process problem- Where does AI actually help a car dealer?
In two places. Agents, where the work repeats identically: sales and finance contracts, warranty claims, parts invoices, homologation packs, ownership transfers. And models, where a number decided by eye has to move: what a car will be worth in three years, and what to order and how much stock to hold. A model is useful only with the error band declared beside the forecast.
The AI project you were pitched is a process problem- How should an investor assess a target's claimed AI advantage?
With three questions. Is the model the target's own, or a subscription a competitor can sign tomorrow at the same price? Does the data it runs on stay inside the perimeter being bought? And who owns the number, and for how long has it been measured? If the supplier measures it, the investor is buying the supplier's deck, not an advantage.
The AI project you were pitched is a process problem
Start a conversation
Tell us what you have to decide
The useful first email is short and specific. A deadline helps more than a brief. Write before you are ready: the later we arrive, the fewer terms are still yours to set.
We reply to every email, as quickly as we can.
- 01The decision in front of you, in one sentence
- 02The date it has to be made by
- 03What you have already looked at