Blogs · July 2026

How do you stop AI guessing in a construction business?

The danger is not a rude machine. It is a confident one. Here is how you bound it so it cannot invent a price, a date, or an opinion on safety.

The short answer: you bound it, and you check the boundary before you buy. The danger in construction is not a machine that is rude to a customer. It is a machine that is confident: one that offers an opinion on fire safety or load, or quotes a price for a job nobody has looked at. The fix is not a cleverer AI. It is an agent that can only answer from what you have given it, and that hands over the moment it is out of its depth.

The risk is the scope, not the technology

Ask any supplier two questions, and judge them on the answers:

  1. What is it forbidden from saying?
  2. What does it do when it does not know?

If they wave those away, or tell you the AI "is smart enough to figure it out", that is your answer. Smart is not the property you want here. Bounded is.

What Ryku will not do

  • No structural, safety or building regulations advice. Not on fire doors, not on load, not on asbestos, not on damp, not on anything a competent person signs off. It does not have an opinion, because it is not allowed one.
  • No pricing a job it has not seen. It gives the rates and prices you have set. It does not invent a number to keep a customer happy, which is how you end up bound to a job that loses money.
  • No promises about dates you have not agreed. It books from your real availability, not from optimism.
  • No guessing. It answers only from your own knowledge base. If the answer is not there, it says so and hands over.

What it does instead

  • Answers every call, email, chat and message in seconds, in your firm's voice, day and night.
  • Books site visits and surveys into the diary you already run.
  • Answers the questions that make up most of your phone traffic: do you cover this postcode, what do you charge for a call-out, are you available in March, are you insured, can you send a copy of your accreditation.
  • Chases the quote you sent and never heard back about.
  • Hands over to a person, and tells the caller it is doing so.

Handing over is the feature, not the failure

Treat "put them through to a human" as a designed route, not a breakdown. You decide what gets passed on, to whom, and how quickly. A caller with a burst pipe or a site emergency should reach a person fast, with the details already captured so they are not repeating themselves in a panic.

Who is accountable

You are, and that does not change. The knowledge base is yours and you approve it. You set what it answers and what it passes on. Every conversation is logged, so you can read exactly what was said and correct anything that drifts.

An agent you cannot audit is not one to put in front of your customers. That is true in any trade, and it is doubly true when the wrong sentence carries a safety consequence.

Why the boundary is worth insisting on

Construction is not a forgiving industry to get this wrong in. It has been the worst-hit UK sector for company insolvencies four years running, with 3,950 firms going under in the twelve months to November 2025. A bad quote you are held to, or a technical answer you did not authorise, is not a customer service problem. It is a solvency problem.

That is precisely why the agent handles the admin and nothing else. The office work is repetitive, high volume, and safe to hand over. The technical judgement is what you are actually paid for.

The questions to ask any supplier

  1. What is it explicitly forbidden from saying?
  2. What happens when it does not know the answer?
  3. How does a caller with an emergency reach a person, and how fast?
  4. Where do its answers come from, and who signs them off?
  5. Can I read a full transcript of every conversation?
  6. Does it book into my diary, or just take a message?
  7. What happens to my customers' data, and what happens if we leave?

Vague answers are the answer.

Frequently asked questions

Can an AI agent give advice on building regulations or safety?

No, and it must not. Those are matters for a competent person. A safe agent recognises the question needs a human and passes it straight on.

Can it quote a price?

It can give the rates you have already published or set. It does not price a job it has not seen, and it does not commit you to a number you have not agreed.

What happens if someone rings with an emergency?

It identifies the urgency, tells the caller it is putting them through, and routes them by the rules you set.

Will it pretend to be a person?

No. It introduces itself for what it is.

Sources

See it answer for your own business

Tell us how you run, and hear it answer live, built around your own information.