Latest News

Generic AI Chatbots are brilliant at lots of things. Drafting emails? Yes. Summarising documents? Definitely. Translating? Oui.

But general-purpose Chatbots estimating a real building project from a plan using AI? Not a chance.

We recently put a popular AI chatbot, Co-Pilot, which uses a ChatGPT engine through a controlled test. We used a straightforward, clear CAD plan for a garage conversion that had all the necessary detail. The goal was simple: could Co-Pilot produce anything resembling a usable estimate? We’d pre-priced the job using market-leading EstimatorXpress, to provide the figure.

The results were a little alarming.

1. It ignored the drawing entirely

Even with a proper CAD plan uploaded, the AI guessed the floor area.
Twice.

Its calculations weren’t just wrong – they matched what you might consider ‘a typical garage size’, not the actual dimensions on the plan. No measurement, no quantification… just confident guesswork.

2. More detail didn’t help. It invented things.

When asked for a more detailed estimate, the chatbot:

• missed key materials
• made up specifications
• assumed structural choices
• ignored normal sequencing.

It looked convincing, but the detail was fantasy.

3. Even after corrections, it never got close

Our structural engineer guided it carefully – pointing out errors, asking for recalculations – yet the final version still fell miles short of the EstimatorXpress benchmark.

Every iteration under-priced.
Consistently.
Dangerously.

This is how builders lose money without realising until it’s too late.

4. It even ‘forgot’ parts of the conversation

In a bizarre twist, the AI lost track of earlier advice about its own mistakes.
This kind of context-dropping is common in AI chatbots – and another nail in the coffin for relying on them for something as high-stakes as estimating.

The uncomfortable truth: general-purpose AI tools weren’t built for construction

Generic task Chatbots don’t:

• read drawings
• measure anything
• follow building regs
• understand sequencing
• price UK labour or materials
• apply overheads

They simply predict what sounds plausible.

And plausible isn’t good enough when a single wrong assumption can wipe thousands off a job.

Where general-purpose AI is useful – and where it absolutely isn’t

AI is great for more routine tasks: emails, document summaries, research, explanation. But when it comes to estimating?

Generic AI can’t replace professional, specialist estimating software, trained estimators, or builders’ own expertise.
Not today anyway.

Our experiment makes it clear:
Builders shouldn’t trust general-purpose AI with their estimating. Builders need highly professional, well built specialist software tools  – not guesswork or suggestions wrapped in confidence.

What’s the alternative?

Specialist estimating software like HBXL’s EstimatorXpress.

And yes – at HBXL we’re all for AI. Our developers code EstimatorXpress to automate actions so you don’t have to manually key in loads of data. That’s the whole point of this market-leading software tool.

Unlike generic chatbots – our software has a clearly defined purpose, it was built for the job, the data is accurate, and the instructions are watertight. The software knows what it’s doing.

Anyway, that’s for another article – maybe by one of the software developers!

What next?

Ask us about the clever automated features in EstimatorXpress on Chat here (a human will reply!), or call 0117 916 7898 or try for free.

And then you’re free to ask Co-Pilot whether they’d rather fight a horse-sized duck or 100 duck-sized horses…