The client
The firm has delivered major infrastructure, commercial and industrial projects across the UK with its own Chartered Engineers and fully automated fabrication.
The work was out there, and the limit was how many tenders the commercial team could get through, because the owners wanted to win more of the right work without hiring more people to do it.
The challenge
The commercial team tendered 160 to 220 times a year and won 25 to 30. Each response took four to five days of effort, with hours spent hunting for historical project data, cost references and technical precedents that existed somewhere in a 30-year archive but could never be found when needed.
Three externally funded AI projects were running separately with nothing joining them up, the Copilot licences were mostly sitting idle, and there was a written AI strategy with a board member's name on it and no budget behind it.
The starting point
Overall 6 / 10 · Developing tier
A sequenced plan that starts with the data
I ran the AI Readiness Diagnostic, scored the business across six dimensions and checked the findings against the rest of the sector. The biggest constraint turned out to be 30 years of design and manufacturing data locked in formats AI couldn't read, so the plan deals with that before anything else.
A structured review of the 30-year design and project archive: what exists, what format it's in, and what it costs to make machine-readable. The audit feeds a searchable library of the whole archive, built for this firm by a specialist partner, that every AI tool it uses can draw on from then on.
Brainial piloted against live tenders for one quarter alongside the existing process, with time saved and quality measured before any commitment. AutogenAI evaluated as the full deployment option once the Knowledge Vault is live and the pilot has made the internal case.
AI Cloud Fabrication Drawings switched on within the existing Trimble relationship, so there's no new vendor and no procurement cycle. The system matches past drawing libraries to new assemblies, cutting template setup time and freeing Chartered Engineers for the engineering judgement that wins work.
Role-mapped training across commercial, design and operations, with five to six specific use cases per function. Commercial team first, with AI champions in each department running quarterly discovery workshops so adoption keeps coming from the ground up.
A three-question evaluation gate before any tool is adopted. A data sovereignty clause requiring zero-retention agreements from every vendor that ingests client project data. A human-in-the-loop sign-off for all AI-generated outputs, mandatory from day one.
Two bespoke systems, built by the same partner on top of the Knowledge Vault. A Tender Intelligence Engine that looks up the firm's historical win and loss data for sharper bid/no-bid decisions, and a Design Estimate Assistant that retrieves comparable past projects to give engineers a calibrated starting point for early estimates. Both only go ahead once the earlier steps have shown savings, and those savings are what pay for them.
"We came into this process sceptical that AI could deliver real value without being built around our specific way of working. We received a plan grounded in our own numbers, our own 30-year history, and a clear commercial sequence. For the first time, the data problem and the tool problem were treated as two separate questions, and we got a direct answer to both."
Data audit first, a tender pilot second and the bespoke build last, because each step has to earn the next one.
Where things are now
This project also slowed down over the summer, with people away on holiday and momentum lost, and keeping the focus has been quite difficult here too. There are no results yet, so every figure on this page is still a projection, and I'll update it as each step goes in.