It is not the shop floor. It is the two days someone senior spends reading an enquiry before anyone can quote it.
Automation in manufacturing has always struggled with the same thing. The input had no structure.
A customer enquiry arrives as an email. The spec is a PDF someone scanned. The drawing is marked up by hand. Older automation needed clean, predictable, tabular data, and none of that qualifies. So the work stayed manual, and the people doing it were the most expensive people in the building.
Language models changed that specific thing. They read documents that have no fixed format and pull structured facts out of them. That is not a small trick dressed up as a big one. It is the exact capability that was missing.
Everything downstream of the reading step has been automatable for years. Estimating rules, bills of materials, drawing generation. The reading step was the blockage. It is not any more.
These are the processes we automate most often, and what tends to happen to the hours.
| Process | Before | After |
|---|---|---|
| Enquiry and spec reading | Half a day to two days of engineer time per enquiry | Minutes. Requirements extracted, out of scope flagged, open questions listed |
| Estimating and quotation | Days in a spreadsheet only one person understands | About 1.5× the quotes per estimator. Priced against your own rules, with the working shown |
| Bills of materials | Re-keyed by hand from the costing sheet | Generated. Straight from the estimate, matched to the drawing |
| GA drawings and layouts | CAD hours spent on jobs you may not win | Minutes. Scaled 2D and 3D output, DXF export, no CAD seat |
| Document control | Manual filing, email transmittals, missed revisions | Automatic. Routed on arrival, revisions tracked, deviations flagged |
Figures from the business where these tools were first built. Your numbers depend on where you start, which is what the process walk or the assessment establishes.
Most of what is written about AI in manufacturing is written by people selling it. Here is what we are not going to tell you.
It removes the reading and the re-keying. Judgement, risk and the relationship stay with the person. If anything, they get more time for the part that actually wins work.
No tool we build sends a number to a customer on its own. It proposes, an engineer confirms. A wrong price is a lost contract or a lost margin, and neither is worth the convenience.
Every extracted fact shows the sentence it came from. If you cannot audit the output, you cannot defend it, and you will stop using it within a month.
Where these tools were first built, nobody lost their job. The team quoted more, quoted faster, and stopped losing money on scope they had not spotted. That is the return.
That phrase is not marketing. It is a design rule, and it decides how every tool we build behaves.
Made Smarter Adoption is a government backed programme that helps UK manufacturing SMEs adopt digital technology.
Match funded grants of up to 50%, to a maximum of £20,000, are available to eligible manufacturers for digital technology adoption. Software built to automate a defined process generally sits squarely inside what the programme is designed to fund, because the business case is a measurable productivity gain rather than a capital purchase.
Eligibility, grant rates and the application route vary by region, and the programme changes. Check the current position for your area before you plan around it.
We are happy to help you put the numbers together for an application. We do not charge for that, and we would rather you applied than didn't.
For one tool, we walk one real job with you for 90 minutes and give you a fixed price within 48 hours. For bigger or multi-process work, a two to three week assessment comes first, and you keep the roadmap whether you build with us or not.
See Starter Builds Book a Call