How to use AI in construction: from site to sales

AI is used in construction for seven jobs: reading tender documents, producing a first estimate for quantities and cost, preparing the daily site report, drafting payment applications and other documents, comparing supplier quotes, supporting safety checks and handling enquiries when a development is being sold. In each of them AI prepares a draft or gathers scattered information. The measurement, the signature and the decision on site stay with the engineer.

Construction is a sector where productivity grows slowly. According to the McKinsey Global Institute's 2017 report, labour productivity in construction grew by about 1 per cent a year over two decades, against 2.8 per cent for the world economy as a whole and 3.6 per cent for manufacturing. Part of that gap comes from work that still lives on paper, in email and on the phone. AI speeds up some of it, but only after the work is recorded digitally.

Which construction jobs can AI help with?

1. Tender documents: required documents, deadlines and payment terms are pulled from hundreds of pages into a table.
2. Quantities and cost: the take-off is matched against a rate list and missing items are flagged.
3. Daily site report: photos and voice notes become a report draft by the evening.
4. Payment applications and documents: payment applications, subcontracts and monthly client reports come from templates.
5. Procurement: supplier quotes that arrive in different formats are compared in one table.
6. Safety: camera images are checked for helmets and hi-vis, and alerts go to the safety officer.
7. Sales: enquiries from Instagram, WhatsApp and ads get a drafted reply and reach the sales team at once.

How can AI read a tender pack?

A tender pack holds the conditions of contract, technical specifications, the draft contract and appendices. The bid team has little time to read it and decide. With AI, a summary table is produced: required documents, deadline, delay damages, payment terms, bonds and special conditions.

Two rules make the table trustworthy. Each row shows the page the information came from, so the team can check it at the source. A field the documents do not answer is left blank and flagged, never filled with a guess. Whether to bid is still the bid team's decision.

Tender documents are often confidential. Where the data is processed is agreed in writing before the build. Only the question and the relevant section are sent to the model, and documents are not used for training. We describe this approach on our AI integration page.

Can AI be trusted with quantities and cost estimates?

As a first estimate, yes; as the final figure, no. Quantities are still mostly taken from drawings with dedicated software. AI is used to match that table against a rate list, compare it with similar projects and flag missing items.

The main risk is an invented number. A language model can write a plausible rate that has no source. So every figure is tied to a table or a document, and a cell without a source stays empty. The tender price always passes an engineer's check.

How is a daily site report prepared?

A site manager's day is spent on site. Writing the report at a desk in the evening often slips to the next morning, and the project manager and the client hear about delays late.

The setup is simple. During the day the site manager takes photos and leaves voice notes on WhatsApp. The voice notes are transcribed and the photos are filed by block and date. By the evening a draft covers:

• work done and items completed
• crews on site and timesheets
• deliveries received and orders still missing
• delays and their causes
• the plan for the next day

The site manager reads, corrects and approves the draft, and the approved report goes to the project manager and the client. Statutory site records remain the site manager's responsibility; the report does not replace them.

How do payment applications and documents get faster?

A payment application is built from the measurement table, the contract rates and the previous period's totals. The same calculation is repeated every period, and errors usually come from a number copied by hand from one sheet to another.

With document generation, the template is prepared once. Figures are read from the tables and records, not typed in. Missing fields are left blank and flagged. The draft goes to an approval queue, and every version and approval is logged. The same method works for subcontracts, cover letters for bids and monthly client reports. A person, not the system, decides what is signed.

How are supplier quotes compared?

Quotes for concrete, rebar, scaffolding or formwork arrive in different shapes. One supplier sends a PDF, another writes in the body of an email, a third sends a photo. Someone in procurement types them all into a spreadsheet.

With AI, the quotes are placed in the same columns: unit price, quantity, lead time, payment terms and whether delivery is included. Anything the quote does not state stays blank and goes onto a list of questions for the supplier. The table sits next to the project budget, and the buyer makes the choice.

What can AI do for site safety?

Computer vision systems that check camera images for helmets and hi-vis vests are in use today. When something is out of line, an alert goes to the safety officer. These systems assist; the employer's legal duty for health and safety does not change.

Camera footage of workers is personal data under GDPR and similar laws. Workers receive a privacy notice before the system goes live, and the retention period is written down.

How is AI used to sell a new-build development?

A sales office gets questions all day: unit types, floor areas, completion date, payment plans, parking. Most arrive through Instagram, WhatsApp and ad forms. When the sales team is on site or in a meeting, messages pile up.

This is the core of ALTAI Digital's enquiry handling and advertising work. The channels are gathered into one queue. Common questions get a reply drafted from the developer's information, and the buyer is told they are writing to an assistant. Price negotiation and contract questions pass to a sales adviser. Every lead from an ad reaches the sales contact at once and is recorded with the ad it came from.

The development's website is part of the same setup. Unit types, floor plans and availability are updated from an admin panel, and the site is built in several languages when foreign buyers are targeted. For a steel construction company in the Netherlands, we built a site where certificates moved from behind the contact form to the home page. Each completed project got its own page, the site went live in three languages and the quote form now accepts technical drawings directly.

We describe the advertising side on our Meta ads management page and the website side on our web design page.

Where should AI not be used in construction?

• Structural design and approval: this belongs to qualified engineers and building control.
• Safety decisions on site: stopping or continuing work is decided by the safety officer and the site manager.
• Signatures and payments: payment applications, contracts and payment approvals stay with authorised staff.
• Numbers without a source: a rate, quantity or duration does not enter a bid unless it is tied to a document.

Where should a contractor start?

The first question is which document or task takes the most time. For some firms it is the tender pack, for others the payment application, for others the questions coming into the sales office. Start with that one task: prepare the template, build the workflow, measure it for a month, then move to the next.

ALTAI Digital starts with a discovery of about a week, covering the firm's documents, software and repeated tasks. We then prepare a demo and build after your approval. Every part joins one panel, so bids, site reports and sales enquiries can be followed on one screen. We explain the method on our digital transformation page and bespoke programs on our custom software development page.