AI Doesn't Make Decisions — Your Team Does

The Skepticism Is Justified

Spend five minutes on any construction forum or industry LinkedIn thread and you'll find the same sentiment: AI is being oversold. People are tired of tools that promise to "transform" their projects and then deliver a chatbot that doesn't know the difference between NZS 3910 and a kitchen renovation.

That frustration is completely valid. The construction industry has been burned before. By software platforms that were supposed to replace spreadsheets but created more admin, by digital tools that looked great in a demo and fell apart on site, by vendors who didn't understand the work but were happy to sell into it.

So when someone says "AI can help your construction project," the natural response is: prove it. Fair enough.

The Real Problem AI Should Solve

Construction professionals do not need help thinking. The people managing a $100M infrastructure project under NZS 3910 are experienced, qualified, and capable. They do not need a machine to tell them what to do.

What they need is a way to make sure nothing gets missed.

Consider a mid-size project with five contracts running simultaneously. Each contract has its own set of obligations, response deadlines, variation procedures, and payment claim windows. Under the adjudication, and suspension rights in construction">Construction Contracts Act 2002, a missed payment schedule response can turn the full claimed amount into a debt payable immediately. No argument, no recourse. Under NZS 3910, a notice of potential claim has a strict time window that, if missed, can extinguish the claim entirely.

The volume is the problem, not the competence of the people. When you're running five contracts, managing correspondence across dozens of stakeholders, and responding to issues on site every day, the risk isn't that you don't know what to do. It's that something falls through the cracks because there's simply too much to track manually with 100% reliability.

What "Human-in-the-Loop" Actually Means in Construction

"Human-in-the-loop" gets thrown around a lot in AI marketing. In construction, it means something specific and non-negotiable:

• AI reads the contract. It maps the obligations, the deadlines, the response windows, the notice requirements across all your agreements.
• AI tracks. It monitors what's coming due, what's overdue, what's been actioned and what hasn't.
• AI flags. When something needs attention (a CCA payment schedule deadline in five working days, a variation response due under Clause 9.8.3, a notice period about to expire) it surfaces that to the right person.
• A qualified professional reviews it and decides what to do. The contract administrator, the project manager, the engineer to the contract. The person with the authority and the professional responsibility.

Professional responsibility stays with the qualified people. Always. The system's job is to make sure those people have complete information when they make their call.

The Spreadsheet Comparison

Every project team already uses tools to stay on top of their obligations. Spreadsheets. Outlook reminders. Whiteboards in the site office. A good PA who knows when things are due. These aren't signs of weakness. They're signs of professionalism. People who take their obligations seriously build systems to track them.

AI is the next evolution of the same principle: systematic tracking. It reads faster than a person can. It doesn't forget items when it gets busy. It can cross-reference obligations across multiple contracts simultaneously. But it's still just a tool. The same way a spreadsheet is a tool.

Nobody argued that spreadsheets would replace project managers. AI shouldn't either. It should do what spreadsheets do, but at a scale and speed that manual tracking can't match.

What AI Should Not Do in Construction

This is where honesty matters. AI should not:

• Give legal advice. A system can identify that a contractual notice period is about to expire. It cannot tell you whether to issue the notice. That is a commercial and legal decision for your team and your lawyers.
• Make commercial decisions. Whether to accept a variation, challenge a payment claim, or grant an extension of time. These are professional judgements that require context, relationships, and experience that no system has.
• Replace the relationships between people on a project. Construction runs on trust between people. A phone call between the engineer and the contractor's rep resolves more issues than any software ever will. AI should support those relationships by reducing the admin friction that creates tension, not try to replace them.
• Be used without domain-specific configuration. A generic AI tool pointed at a construction contract is about as useful as a graduate who's never seen a building site. The system has to understand the contract form, the legislation, the industry norms, and the way projects actually run in New Zealand.

What Provan's Approach Looks Like

Provan was built by someone who has spent 10 years working NZS 3910 contracts, managing projects from $10M through to $750M, and dealing with the Construction Contracts Act every week. The system knows NZS 3910 because the person who configured it has worked the contracts. Not because someone scraped a standards document and fed it into a chatbot.

That matters. Understanding that Clause 13.3.1 notice requirements are routinely tightened from "as soon as practicable" to an absolute 10-working-day time bar via special conditions. And that missing that bar can extinguish a legitimate claim. Isn't something you get from reading a PDF. It comes from sitting in meetings where those clauses are negotiated, and from watching what happens when they're missed.

Provan's system is configured with that depth of understanding. It tracks obligations the way an experienced contract administrator would. But across every contract simultaneously, without the risk of something being overlooked because someone was dealing with a site issue that morning.

If You're Skeptical, Good

Skepticism in construction is a survival skill. Projects go wrong when people accept things at face value without testing them. The same standard should apply to any AI tool. Including this one.

The difference between a tool that earns trust and one that doesn't is simple: does it do what it says it does, and does the person behind it understand your world?

Provan isn't asking anyone to hand over their professional judgement to a machine. The proposition is straightforward: your qualified professionals keep making the decisions. The system makes sure they have everything in front of them when they do.

If that sounds like something worth a conversation, we'd welcome the chance to show you what it actually looks like.