Construction
& engineering.
Construction & engineering
In a construction company information is not missing: it is buried in documents. Tenders, contracts, site minutes and site instruction books decide whether a project makes or loses money, and the information needed for decisions that change a project rarely arrives in time.
A tender pack can run to more than three hundred pages, and the clauses that matter are scattered through it: valuation deadlines, penalties, material requirements. One person reads it end to end, once, in a hurry. Then, for the months the project lasts, those conditions live in someone's memory. And the surprises turn up in the monthly valuation.
The second front is the back and forth between the office and the site. Measurements, daily reports, minutes and variations are sent by email and retyped by hand, which makes it impossible to follow the project's finances. That has to be solved before building a dashboard. A dashboard fed with data typed in late shows you last week's project.
None of the three needs new data. The documents are already filed; it is just that nobody opens them.
An assistant that reads the whole tender pack and returns, citing the page, the valuation deadlines, the penalties, the material requirements and the unusual clauses. What is an afternoon's reading today becomes a list you can check.
Gathering the measurements, daily reports and delivery notes from each site in one place and leaving the draft valuation ready. The person reviews and signs instead of collecting.
Actual cost against the construction budget, by cost heading and by project, with this week's data rather than the close. This is the dashboard, and it comes after the two points above because it needs the data to arrive on time.
Four steps, and the first is not technological. First you define the business problem; the technology comes after.
In the short term, steer clear of ‘putting AI into the company’: one live project and its tender pack. A small, real scope that can be measured, and where you know how much time it takes today.
Whether the tenders and minutes are searchable PDFs, scans or paper; whether there is a folder structure or each site manager keeps their own. This decides the rest of the project and is the first thing to resolve.
You should be able to ask in plain language and get an answer that cites the document and the page. Without the citation it is no use: on site, an answer that cannot be checked does not get used.
With data arriving on time, cost control by heading and overrun by project have something to feed on. Before that, a dashboard is an expensive ornament.
All seven, in the order they usually reach a construction company. The first four are the way in; the last three come once the earlier ones are working.
Tenders, contracts, minutes and site instruction books you can question in plain language, with answers that cite the document and the page. It is the first project for almost every construction company we have worked with.
See the product
Daily site reports, measurements and variations that arrive by email and that someone types in today. The agent interprets them and leaves them ready for a person to sign off.
See the product
Actual cost against the construction budget, by heading and by project, with this week's data. It arrives once the earlier steps have got the data coming in on time.
See the product
The data warehouse that joins site, purchasing and accounts when those three systems do not talk to each other and everything else stays half done.
See the product
The discovery session where you decide which project, which document and which number. It is most useful when several companies in a group run different systems.
See the product
Predicting a project's overrun from how closely it resembles earlier ones, by type of work and by cost heading. It needs several years of closed projects with their figures.
See the product
The whole monthly valuation cycle: gathering measurements and delivery notes, reconciling them with the contract and leaving the draft ready, with the exceptions escalated to a person.
See the productFor both, and for refurbishment and building services contractors too. AI in engineering starts in much the same place, because the underlying problem is the same: long documents that decide the margin, and a back and forth with the site that is done by hand. What changes is the volume, and how many subcontractors have to be reconciled.
Yes, and that is the most common situation. None of the projects on this page needs BIM. They need the tenders, contracts, minutes and daily reports that every company in the industry already has. BIM is a different project, and a considerably more expensive one.
Only if it answers with citations. We build these searches so that every answer comes with the document and the page it is taken from. Whoever uses it can check it in ten seconds. On site, an answer that cannot be verified does not get used, and rightly so.
Legible scans work as they are. Paper has to be digitised, and that is decided by volume. For the minutes of live projects it is almost always worth it; for a twenty-year archive, almost never. That check is the second step of the project and happens before anything is committed.
You can start there, and sometimes it is the right call. If your site data already reaches the system on time, cost control by heading is a straightforward project. We do not recommend building the dashboard when the data arrives two weeks late, because then the problem is not the dashboard.
Fewer than you fear and more than none. You need someone who knows the process, usually site management or the technical office, for a few hours a week during the first phase. We do not have that knowledge. We handle the whole technical side.
The ones you receive and keep do: contracts, valuations, daily reports and minutes that reach you from each subcontractor. It is where it shows most, because reconciling the monthly valuation is precisely a matter of pulling together material from several companies in different formats. What does not come in is the subcontractor's internal system, which you neither have nor need.
