Construction AI: the safety regulator halved its own approval times, and the software you're locked into keeps getting dearer

Here's a sentence I didn't expect to write this year. The part of the state that everyone spent 2025 complaining about has got quicker. The Building Safety Regulator set up an innovation unit in August 2025 to take on Gateway 2 approvals for new-build higher-risk buildings, and in the 12 months to the end of August 2026 it brought the median approval time down from 43 weeks to 22. The pass rate went the other way, from 39 per cent to 92. Property Week, Building and Housing Today all reported the one-year figures in early September 2026, and they trace back to the regulator's own June-to-August transparency data on GOV.UK.

So what does 22 weeks do for you. It changes the number you put in the programme. For a year the honest planning assumption for a new HRB was "budget the best part of a year and hope", and a lot of residential schemes were sat on the shelf because of it. 22 weeks isn't the 12 the Building Safety Act still promises, and I'm not going to pretend it is. But it's a number you can actually build a start-on-site date around, which the old one wasn't.

Two caveats before anyone frames this on the wall. These are the BSR's own transparency figures, not an independent audit, and a median hides the tail, so the awkward, contested applications are still taking a lot longer than 22 weeks. And the improvement came from a specialist unit handling new-build cases, so if your job is a refurbishment or a change to an existing HRB you may not see the same speed yet. What that means for the person building the programme is simple. Plan to 22, watch your own caseworker, and don't let a client hold you to 12.

Every so often a category argument gets made for you by the people you're arguing against. This week it's the renewal invoices. Analyses of Procore pricing published through 2026 have contractors reporting rises of 12 to 15 per cent at renewal, after years of the usual 5 per cent bumps. The line that stuck with me came from a roofing contractor quoted in one of them, whose bill went from $22,000 to $31,000 over three years with no change in how the team used the software. These are customer-reported and review-aggregator figures rather than anything the vendor has published, so treat them as directional. But the direction's the point, and the model behind it is the real story: pricing tied to your annual construction volume, so the better your year, the bigger your bill.

Set that against challenger construction software, the AI-first platforms taking on the big incumbent systems the way Monzo took on the high street. The test is a simple five: AI doing the actual work rather than bolted on as a chatbot; broad enough to cover how a project runs, not one task; usable by the person on site doing the paperwork; a published price you can see before you talk to anyone; and no multi-year lock-in, so your data walks out with you when you leave. Something like constructionAI, the AI-native platform a Chartered Builder shipped this year covering tenders, programmes, RAMS and payment applications, meets all five and puts its pricing on the website. The contrast with a volume-linked renewal you can't forecast is the whole pitch.

The 17 September edition of the Bricks and Bytes daily blueprint led on exactly this, "higher rates, AI workflow redesign and a new way to buy construction", which tells you the framing is starting to travel beyond the people selling it. The comparison with challenger banks only goes so far, a payment app is a lot simpler than a project, but the point holds: when the incumbent's price only ever goes up and the exit is expensive, a published price and a clean way out stop being nice-to-haves. For your commercial lead, the job this quarter is unglamorous. Open the renewal, list what your team actually opens each week, and price the alternative against that list rather than the suite you were sold.

The number the whole sector keeps repeating is more than £100bn of UK data-centre construction in the pipeline, and it's genuine. What the September analysis of the Barbour ABI figures makes plain, first reported by Construction News on 27 August 2026, is how little of that is turning into work on the ground right now. Seventy schemes worth about £2.6bn are already complete. Twenty-one are on site, and they'll cost close to double that at £4.9bn. Then the tail: 55 with detailed planning worth around £12.7bn that haven't broken ground, and another 22 at outline stage estimated at £24.5bn. The London and M4 corridor holds 82 of the 173 committed projects and 5.2GW, more than a third of the capacity.

So the pipeline is a promise, not a workbook. The thing standing between the two is well rehearsed by now and worth saying plainly to anyone getting excited about the headline: grid connections that take years, transformers and cooling plant with long lead times, and the specialist labour to install it all. None of that clears because a planning committee said yes. I'd be wary of any bid model that treats the £100bn as demand you can price against next year, when the honest live number is 21 sites.

That's what it's about. For a contractor weighing whether to chase this work, the qualifying question isn't "is there a pipeline", it's "has this specific scheme got a firm grid date and a plant order in", because those are what decide whether you're building in 2027 or holding a design on a shelf. The pipeline pays the people who can carry that risk. It doesn't pay the ones who assume it away.

Line the three up and they point the same way: plan to the real number, not the promised one. The regulator has proven it can move faster, so hold your programme to the 22-week median and manage your own caseworker to it. Your software costs more each year unless you make the case for something with a published price, so open the renewal and price it against what your team opens, not what you bought. And the data-centre pipeline is a queue behind the grid, so qualify the connection date before you qualify the margin. Same discipline in all three: trust the evidence in front of you, not the headline over it.

For your board pack: put the 22-week BSR median and your worst-case caseworker time side by side, and let the gap set your risk line.