Case studies

Work we have delivered

Real projects for real businesses. Client names are withheld where the work is commercially sensitive.

Electronics manufacturing · Nottingham

Turning PCB files into machined test jigs in minutes

BeforeTest jig design and manufacture outsourced

AfterJigs produced in-house from the PCB files

Design timeMinutes, not hours of CAD

Lead timeWas 3–6 weeks outsourced

The problem

Every new board needs its own test fixture. The bed-of-nails layout, the support geometry, the cut-outs for connectors and tall components — all of it is specific to that one PCB, and all of it has to be right before a single board can be tested.

This manufacturer was outsourcing that work. It is a reasonable choice, and it carries three costs that compound. There is the price of each fixture. There is the lead time — three to six weeks — sitting directly between a finished board design and the ability to test a single unit. And there is the revision problem: boards change late, and every change means going back to the supplier, renegotiating, and waiting again.

The obvious alternative — bringing it in-house — ran into the same wall every time. Designing a jig by hand meant hours of CAD work per board, done by an engineer whose time was already spoken for. The arithmetic never quite justified it.

What we built

The insight was that the information needed to design the jig already exists, in a precise, machine-readable form, before anyone opens CAD. The PCB design files describe exactly where every pad, hole, connector and component sits, to a tolerance far tighter than anything a fixture needs.

So rather than treating jig design as a drawing exercise, we treated it as a translation problem. The software reads the PCB design data and generates the files a CNC machine can cut from directly — turning what was hours of manual CAD into a process measured in minutes.

What changed

The immediate effect is the time. A fixture that took hours of engineering time to draw is generated in minutes, which moves jig-making from a considered decision to a routine step.

The second-order effects matter more. Fixtures are made in-house, so a wait of three to six weeks becomes a job on the CNC machine. On a board that is already late — which is most of them — that is the difference between shipping this month and next. A board revision no longer means a supplier conversation and another three-to-six-week wait — it means running the new files through again. And the engineering time that used to go into drawing fixtures goes back into the work the business is actually paid for.

Why this generalises

This is the shape of most genuinely valuable automation in an engineering business, and it is worth naming because it rarely looks like "AI" in the way people expect.

There was no chatbot and no model making judgement calls. There was a task that consumed skilled hours, performed on information the business already held in structured form, with an output specific enough to be checked. Those three conditions are where software pays back fastest — and most firms have at least one such task they have never examined, because it has always simply been part of the job.

What an engagement actually looks like

Illustrative · Sheet metal fabricator, 40 staff

An AI audit where the answer was "not yet"

The owner wants to know where AI fits. Quoting is the obvious candidate — two estimators, six hours a week each, and jobs regularly lost on a four-day turnaround.

Two weeks of measurement finds something else. The estimators are not slow; they are waiting. Roughly half of enquiries arrive without a material specification or quantity, so the first action on most jobs is an email asking for it, then a two-day wait.

Automating the drafting would compress the wrong half of the process. The recommendation is an enquiry form that makes the missing fields impossible to omit, and an automatic first reply that asks for anything still absent. Cost: a few hundred pounds. Effect: two days off the average quote.

The AI quoting project is worth doing — but second, once the input problem is fixed, because only then does the drafting time become the constraint. An audit that talks you out of the expensive thing first has done its job.

Illustrative · Precision machining, family-run, 25 staff

A first project chosen for the people, not the payback

The strongest financial case is production scheduling, held in a spreadsheet by the works manager, who has run it for eighteen years and can explain every decision in it.

It is also the worst possible place to start. Touching it first tells the most experienced person in the building that the thing defining his value is being replaced by software.

So the first project is goods-in paperwork instead: delivery notes photographed on arrival, matched automatically against purchase orders, discrepancies flagged. Nobody enjoys that task and nobody's standing depends on it. Payback is modest — a few hours a week — but it is visible, it works, and it belongs to the team that asked for it.

Scheduling comes nine months later, with the works manager defining the rules the system follows. Same project, different order, and a materially better chance of it being used.