Preparing a Construction Business for AI

AI is genuinely useful in construction for assistance work — the work that surrounds decisions rather than making them. The value is real, but it sits in a specific band of tasks, and being clear about that band is what makes adoption safe.

Tender and scope review assistance. AI can summarise a tender pack or scope document, identify topics that warrant attention, and prepare a structured list of items to review. It does not decide whether to bid or what to price. It helps a person get to the point of decision faster by organising information they already need to read.

Meeting minutes and action tracking. AI can turn an approved transcript or set of notes into draft minutes, assigned actions and follow-up communications. A project employee still checks completeness and accuracy, but the drafting work — which is repetitive and time-consuming — is done in minutes rather than hours.

Routing RFIs and documents. AI can capture incoming requests for information, notify the responsible party, track status and file the approved response into the correct governed location. The routing rules are defined by the business; AI executes them consistently where a person might forget.

Drafting project reporting. AI can prepare a first draft of an internal progress summary from approved source information. A responsible project employee checks it for completeness and accuracy before it goes anywhere, but the blank page is already filled with structured content to review.

Supporting handover and close-out coordination. AI can track required documents, reminders, approvals and handover-pack status across a defined workflow. It does not decide that a project is complete, but it helps ensure nothing is missed in the process of getting there.

Searching internal knowledge and procedures. AI can help authorised employees locate approved procedures, templates, standards and prior project knowledge stored within a governed Microsoft 365 environment. This is one of the most immediately useful applications, because construction businesses accumulate enormous amounts of documented experience that is hard to find when it matters.

Some decisions in construction carry professional and legal accountability that belongs to a qualified, named person. AI cannot hold that accountability, and so it must not make these decisions — not independently, and not as a recommendation that is simply accepted.

Engineering decisions. AI must not determine structural adequacy, load capacity, or whether a design detail will perform as required. These are engineering judgements that belong to a qualified engineer.

Design decisions. AI must not decide what a building should look like, how spaces should be arranged, or whether a design meets the client's requirements. These are design judgements that belong to the architect or designer.

Safety decisions. AI must not determine whether a site is safe, whether a method is safe, or whether a risk has been adequately controlled. These are safety judgements that belong to a qualified safety professional.

Certification decisions. AI must not certify compliance, sign off on inspections, or determine whether work meets building code requirements. These are certification decisions that belong to a building surveyor or certifier.

Contractual decisions. AI must not interpret contract terms, determine entitlements, or decide whether a variation is valid. These are contractual judgements that belong to a qualified person, often with legal advice.

Payment decisions. AI must not approve payments, determine valuation of work, or decide whether a progress claim is payable. These are commercial decisions that belong to a named person in the business.

Statutory determinations. AI must not determine whether work complies with regulation, whether a notification is required, or whether an obligation has been met. These are statutory determinations that belong to a qualified person.

The line between assistance and determination is the most important concept for everyone in the business to understand, not just management. If only leaders understand it, the boundary will not hold on site or in the project office where the actual work happens.

AI summarising a specification is assistance. AI concluding that a design complies is determination. The first helps a person work faster. The second takes a decision that belongs to someone else.

AI drafting minutes from an approved transcript is assistance. AI deciding that an action is complete is determination. The first saves time. The second removes accountability from the person who owns the action.

AI routing an RFI to the responsible engineer is assistance. AI answering the RFI on the engineer's behalf is determination. The first helps the right person see the right thing. The second puts an unqualified response in front of someone who expects a qualified one.

AI preparing a draft progress summary is assistance. AI deciding the project is on track is determination. The first gives a project manager a starting point. The second makes a judgement that should be made by a person who can be held accountable for it.

The test is simple: does the output inform a decision, or does it make one? If it makes one, it has crossed the line, regardless of how convenient it is.

The single most common reason AI goes wrong in a small business is not a failure of the technology. It is a failure of permissions. AI reaches whatever the person using it can already reach. If a project manager can see every project's commercial information, so can the AI tool they are using. If an estimator can open a folder containing a competitor's tender, so can the AI.

In construction that exposure is particularly acute. Project information includes tender pricing, subcontractor rates, margin calculations, variations under dispute and correspondence with external parties. An AI tool that surfaces a document the user was never meant to open does not announce that it has done so. It simply includes the information in its output, and the user may not even realise where it came from.

This is why permissions and access control come before AI adoption, not after. The question is not whether AI is safe in the abstract. It is whether the information environment AI will reach is properly controlled. If it is not, AI becomes a tool for accelerating the wrong kind of visibility — exposing information faster than anyone intended, to people who were never meant to see it.

Fixing this is not a one-off task. It means understanding who currently has access to what, deciding who should, and closing the gap between the two. It means reviewing guest access from finished projects, shared links that never expired, and broad permission groups that were created for convenience and never tightened. None of that is glamorous work, but it is the work that makes AI safe to introduce.

AI is only as good as the information it can see. If the current revision of a drawing is unclear to a person, it will be unclear to AI. If project correspondence is scattered across personal mailboxes, local desktops and ad-hoc shared links, AI cannot find it reliably — and neither can your staff.

Construction businesses accumulate enormous volumes of project information: drawings, specifications, RFIs, submittals, contracts, variations, site reports, meeting records and handover documentation. When that information lives in a consistent, governed location with a visible current revision and clear permissions, AI can help people find it, summarise it and work from it. When it does not, AI produces confident answers based on the wrong version.

The preparation work is the same work that makes the business better without AI: establishing one authoritative location per project, retiring superseded versions clearly, and applying the same structure on every project so people do not have to learn a new layout each time. If you are already working on improving project information and document control, you are already doing the work that makes AI useful.

Information readiness also means information quality. AI cannot fix duplicated, contradictory or outdated records. It works with what is there. If the source information is wrong, the output will be wrong — and in construction, wrong output carries consequences that are measured in rework, delay and dispute, not inconvenience.

A construction business does not need an enterprise AI policy framework. It needs a short, clear, enforceable position that people actually follow. The longer and more abstract the policy, the less likely it is to shape what happens on site or in the project office.

A proportionate position covers three things. Which tools staff may use for work, and which they may not. What information may be pasted into an approved tool, and what may never leave the business's controlled environment. And who is responsible for checking AI output before it is acted on.

The clearest line is the one about what never goes into a public tool. Client information, tender pricing, subcontractor rates, contract terms, drawings and any document containing commercial or personal information should never be pasted into a tool the business does not control. If staff need AI assistance with that kind of material, they should use an approved tool within the governed Microsoft 365 environment — not a public website.

The position does not need to be long. A single page that names the approved tools, states what is never pasted elsewhere, and assigns ownership for checking output is more useful than a twenty-page document nobody reads. The test is whether a new starter could understand it on their first day and follow it without asking.

The right way to begin is with one workflow, measured, with a named owner, before anything is scaled. Not a pilot in the sense of a technology experiment, but a real piece of work that someone is responsible for and that produces a result the business can evaluate.

Meeting minutes is a common starting point. The input is an approved transcript or set of notes. The output is a draft that a named person checks for completeness and accuracy before it is circulated. The workflow is bounded, the risk is low, and the value is visible. If it works, the business has a proven pattern it can apply to the next workflow.

What does not work is introducing AI broadly and hoping people figure it out. Without a named owner, nobody is responsible for whether the output is correct. Without measurement, nobody knows whether the workflow is actually faster or better. Without a single proven example, scaling is just exposure — more people using a tool in more ways with less understanding of whether it is helping.

The sequence matters: prove one workflow, measure it, assign ownership, and only then consider the next. A construction business that has one working AI-assisted workflow with a person who owns it is further along than one that has given everyone access and assumed the results would follow.

LOOKUP helps construction businesses prepare for AI by doing the groundwork that makes it safe and useful. That work happens in the Microsoft 365 environment — permissions, information structure, identity and access — before any AI tool is introduced.

An AI readiness assessment establishes where the business actually stands: what information exists, where it lives, who can reach it, and what needs to change before AI can be introduced safely. The assessment is specific to the business, not a generic checklist.

From there, LOOKUP helps with Microsoft 365 and permissions groundwork, ensuring that project information sits in governed locations with access controls that reflect who should see what. Microsoft Copilot governance ensures that when AI is introduced it operates within those boundaries rather than around them.

LOOKUP then helps implement specific workflows — the one proven use case that becomes the pattern for the next. Throughout, LOOKUP coordinates with project platforms, estimating systems and construction accounting software rather than replacing them. The work is about making the surrounding environment ready for AI, not substituting one platform for another. Learn more about LOOKUP's approach for construction and property development.

Securing external collaboration and project access — how to grant consultants, subcontractors and clients access to project information and take it back when the project ends.

Preventing invoice and payment-redirection fraud — how to stop a fraudulent payment instruction being acted on when your business pays hundreds of supplier invoices.

Reducing administrative overhead across project and business systems — why the same information is typed into estimating, project management and accounting, and what can be done about it.

Building a technology roadmap for a construction business — what order to do things in, from securing what you have to improving it over time.