Don’t just try before you buy into an AI-enabled system: interrogate the vendor with these key questions from Kristian Brown MCIOB
A polished demo proves very little. Before you put AI into your business, there are six questions I would want answered properly.
Imagine you build an AI workflow to capture RFIs from email, WhatsApp and your project system. It works as designed. Then somebody phones one through, nobody logs it, and the system never sees it. The technology has not failed – the process has.
That is where a lot of AI integration goes wrong. It is easy to focus on what the tool can do without first testing the process around it, the claims being made, the information being handed over or whether it would keep running if the company behind it disappeared.
1. Where did that saving number come from?
You are told the software can cut estimating time by 40%. Sounds good until you ask how long your estimators spend on that work today. Often, nobody knows.
That makes it a percentage of a number nobody has.
Before buying, ask how they calculated the savings and what evidence sits behind it. Then establish your own baseline before the system goes live. For estimating, track tendering time for two weeks. Once the software is in, run the old method alongside it for the first few tenders and compare the results.
Speed alone is not enough. A tender produced faster but with a pricing error can cost more than the hours it saved.
2. Where does your information actually go?
“Your data is secure” is not an answer.
You may be handing over drawings, rates, subcontractor quotes, tender pricing, contracts and client correspondence. Ask which providers receive it, where it is processed, who can access it, how long it is retained and whether it is used for anything beyond your service.
Also look at what you have already agreed with your clients. On sensitive, public sector or confidential work, make sure your contract allows you to pass project information to any third-party technology it relies on.
3. What exactly are you buying?
Paying for something to be built does not automatically mean you own every part of it.
The contract should separate your data, anything created specifically for your business, the provider’s existing intellectual property and any third-party technology underneath it. You do not need to own every component, but you do need the right to keep using what you have paid for if the relationship ends.
4. Can you prove it using your work?
“95% accurate” raises another question: 95% of what?
A 2026 study found AI take-off closely matched the contractor on repetitive count-based items, but was weaker on areas and more complex geometry. A headline accuracy figure does not tell you enough.
If it is an off-the-shelf product, give them one of your project drawings or documents and ask them to demonstrate the result. If it has to be built around your business, agree upfront what the pilot must prove and how you will judge it before wider rollout.
5. What happens when something goes wrong?
The obvious failure is a system stopping. The harder one is an output that looks right, but is wrong.
The danger is automating away the person who knows what to check. If a senior estimator or QS understands where the commercial risk sits, replacing their judgement with a junior checker can create a bigger problem than the technology solves.
I prefer workflows to fail loudly. Exceptions should be flagged, failures should trigger an alert and higher-risk outputs should have clear checkpoints before anything is issued.
Not everything requires somebody reading every line, but the person signing off still has to know enough to recognise when something does not look right.
6. What happens if the AI provider changes or disappears?
Before buying, establish what you can export, what documentation you get, which accounts and integrations you control, what happens to your data on exit and whether somebody else could take over the workflow.
Businesses get bought, products change and model providers retire versions. Pricing can move too. Ask in advance what can change, how much notice you get and what your options are if you decide not to accept it.
The exit route needs to exist before you need it.
One final check
One final check before you buy: speak to someone in another company who uses it every day, not just the person who signed it off. Ask whether it still saves time, how accurate it is and whether it has become just another thing to manage.
If the answers to these six questions are clear, specific and backed up by evidence, you have something worth testing. If they are vague and inconsistent, I would walk away. A conversation with an expert should get your attention. The answers to these questions should decide whether you buy.
Kristian Brown MCIOB is the founder of Built Logic, an AI specialist for the built environment. He has more than 15 years’ experience in UK construction, running construction and property development businesses. He now works with companies implementing AI.
