Choosing an automation partner for a construction business
A construction business choosing an automation partner has one extra test to run beyond the normal due diligence: does this firm understand site reality? Plenty of agencies can automate an office. Very few understand that your data is born in the rain, typed with gloves on into a WhatsApp group with patchy signal, by someone whose actual job is pouring concrete before the pump goes home.
That gap matters because construction automation fails at the capture end, not the processing end. A beautiful dashboard fed by nothing is a beautiful dashboard about nothing. So the due diligence below is deliberately construction-specific; for the general criteria (ownership, maintenance, who turns up to the first meeting) we have written a separate guide to choosing an AI automation partner in the UK, and both sets of questions apply.
Why is construction different from ordinary office automation?
Three reasons: where the data is born, what the records are for, and who has to cooperate.
Where the data is born: on site, in the mess. The richest record of most UK projects already exists in the WhatsApp groups running the job: progress photos, delay chatter, delivery confirmations, verbal instructions half-captured. A partner who proposes a new app for operatives to fill in has not met an operative. The realistic play is capturing what already flows, not demanding new behaviour from the people with the least time to give it.
What the records are for: not analytics, duties. Site records feed statutory obligations under CDM 2015, contractual machinery under JCT and NEC (notices, early warnings, compensation events, payment applications), and increasingly the information management expectations of ISO 19650 under the UK BIM Framework. An automation that summarises nicely but cannot say who recorded what, when, is a liability dressed as a convenience.
Who has to cooperate: subcontractors, consultants and clients who did not choose your tools and will not log into them. Anything that only works when the whole supply chain adopts it will not work.
What does site reality do to a neat office workflow?
It breaks every assumption the demo was built on. Connectivity is intermittent, so anything that fails without signal fails daily. Hands are gloved and wet, so voice notes and photos beat forms. The person capturing the record is the busiest person on the project, so capture must cost seconds or it will not happen. And the day does not pause for software: if the tool falls over during a concrete pour, the pour finishes and the record is lost forever, because site events do not have a retry button.
A partner who has stood on a site knows all this without being told. A generic agency finds out during your rollout, at your expense, and the failure gets blamed on "user adoption" rather than on a design that ignored the user.
Which questions expose a generic agency in one meeting?
| Question | A construction-literate answer | The generic-agency answer |
|---|---|---|
| Where does our project data actually live today? | "Show me the WhatsApp groups" | "We'll integrate your CRM" |
| How does capture work with no signal and gloves on? | Offline-first, voice and photo, seconds not minutes | A form-based app demo |
| What records do CDM duties actually require us to keep? | Talks about inspections, inductions, the file, who holds them | Asks you to explain CDM |
| What is an NEC early warning and why does timing matter? | Knows clocks and notice bars exist and shape the design | Never heard of it |
| How do site records differ from head-office admin? | Two different problems, two different designs | One platform for everything |
| Who signs the record, and can we prove when it was made? | Attribution and timestamps treated as first-class | "There's an audit log somewhere" |
None of these questions are unfair, and a good generalist partner might still be worth hiring for your accounts and enquiry handling. But the site-facing work should go to someone who answers the left-hand column without notes.
What should the partner understand about your records duties?
Enough to design for evidence, not just efficiency. That means knowing the difference between a delay noted in a meeting and a notice served under the contract; knowing that a JCT payment application and an NEC programme submission are dated artefacts with consequences; and knowing that under CDM 2015 the principal contractor's records exist for regulators and futures beyond this project. The practical test is simple: ask how their proposed system would help you evidence when your team became aware of a delay event. If the answer is a shrug, the system is an office toy.
This is the reason our own product work went where it did: Construction Metric exists because turning existing WhatsApp traffic into structured, attributed site records was the version of automation that survives contact with a real site. Whether you use it or something else, hold every candidate to that standard: capture where the work happens, records that stand up afterwards.
How do you actually run the decision?
Shortlist on the general criteria, then run the table above in the first meeting and count the left-hand answers. Start small and site-facing rather than big and platform-shaped: one project, one capture problem, one measurable output, in the spirit of the three AI solutions every contractor actually needs. Agree in writing who owns the accounts and the data before anything goes live.
And trust the smell test. A partner who asks to visit a site before quoting is telling you something. So is one who quotes without asking.