Lacked Ambition

If you ask most people in finance what A.I. will do for their sector, you will typically hear the same three answers: faster response times to win business, automated underwriting elements to process business faster, and stronger, faster summarisation of data to intensify decision-making. However, as we all know, development finance is a stubbornly physical, intensely local, relationship-driven corner of lending; it’s a niche in its own corner of the market, and we believe perhaps its most interesting opportunities to utilise A.I. lie somewhere else entirely.

Development loans rarely fail because a developer lacked ambition; you have to have ambition to be in that market; it’s the base-level entry point. They typically fail because a contractor runs short of cash, a supplier tightens terms, or a subcontractor simply walks off site. A.I. can track public signals such as Companies House filing behaviour, newly registered charges, county court judgments and director changes, and turn them into a near real-time stress score. The same logic extends to key suppliers, where a sudden shift in payment behaviour can be the first sign of trouble. Traditional due diligence is a photograph taken at approval; it’s a moment in time, it’s someone in an office somewhere checking things. This new era of technology, if used well, is a heart-rate monitor worn for the whole build, and it could buy a lender in our space valuable time if the weather starts to turn.

Planning risk is often the biggest unknown in a scheme, yet it is usually assessed on instinct and local knowledge, and this is again where the tech can tap into “the feel” of a deal. Models trained on years of decisions, officer recommendations, objection volumes and committee makeup could estimate the likelihood and likely timeline of consent for a given scheme in a given authority. Local expertise stays essential, but it would be sharper with a data-backed second opinion, particularly as policy shifts such as the grey belt reshuffle the odds and historic patterns become less reliable. It could also help borrowers by showing early which schemes are likely to sail through and which need more pre-application work.

This also applies to deal structures; a contingency fund is only as good as the cost assumptions beneath it. Quarterly indices tell you where prices were. Machine learning can blend merchant pricing, tender returns, labour availability and import data to estimate where costs are heading. For lenders, that means stress-testing contingencies against realistic scenarios rather than applying a flat percentage. For developers, it means fewer unpleasant surprises around month nine, when the budget meets reality. Even a modestly better forecast, applied across a portfolio, changes how much headroom a lender really has. In the past, we have had software and systems to track all of these elements, to provide better decision-making, but what we are talking about here, in this new era of agency, is data interpretation: thousands of data points merged to give a definitive picture; in the future, these systems are not assisting a decision; they’re making one.

Every credit committee asks one question at the start of every deal: theoretically, what could go wrong. The unknown unknowns can never ever be forecast, and it’s folly to believe they could; however, A.I. can ask that question ten thousand times, from ten thousand different perspectives, and mathematically enhance the scheme's chance of success, it can simulate sales rates, cost overruns, programme delays, interest rate paths and valuation shifts across a scheme's GDV and cost structure, and you get a distribution of outcomes instead of a single base case. Run across a whole loan book, it also shows where risks cluster. The value is not prediction. It is discovering which combination of ordinary bad luck actually threatens the first legal charge, and how much protection a given LTC or LTGDV genuinely provides. That is a far more honest conversation than a spreadsheet with three tabs marked base, upside and downside. Lending and A.I. at a core level actually share some components: the balance of probabilities of one plus one equalling two, the balance of probabilities between perceived good bets and bad; all of it is essentially attempting to reduce the odds of being wrong.

Now if you have read this far into this blog, well done! Most people will not have; they will have given up after paragraph two. The average attention span on a single screen before getting distracted is 47 seconds, according to published research conducted at the University of California, which, for purposes of browsing one of our blogs, is fine- don’t worry, we don’t offend easily; however, for humans engaging in the study of title reports, restrictive covenants, intercreditor deeds, build contracts, warranties: development finance runs on dense legal paper; this is where A.I. will pick up some of the slack to spot errors that are otherwise naturally not caught, language models are already good at flagging unusual clauses, spotting inconsistencies between documents and checking packs against a list of what should be there. Used as a first-pass reviewer, with solicitors making the final call, they would give lawyers more time for judgement and less for hunting. They could also help keep a consistent standard across a large book of loans, where fatigue and time pressure inevitably creep in. For developers, who prize speed to completion above almost everything, that matters.

None of this works without discipline, and it’s important to remember here that models make mistakes, and they make them confidently, which is the worst sort of mistake-maker. In a regulated market, where investor capital is at stake, and financial promotions rules apply, AI outputs need to be explainable, auditable and reviewed by an accountable human. A model can flag a risk; it should not sign off a loan or write an investor-facing claim unchecked. Data quality matters just as much. A model trained on patchy or biased records will produce patchy, biased answers at impressive speed. And confidentiality cannot be an afterthought: sensitive borrower and investor information should only ever go into tools that the firm has properly vetted, contracted and secured.

Looking across these ideas, a pattern emerges: nothing here replaces the people who understand land, bricks and borrowers. However, perhaps each extends their reach so the experienced eye lands on the right site, the right contractor, or the right clause at the right moment. Development finance has always struggled with information that arrives late, risks assessed once and judgement stretched thin across too many deals. Those are exactly the problems these tools are suited to. The lenders and platforms who benefit the most will not be those with the flashiest technology. They will be the ones who point it at the sector's oldest weaknesses, keep humans accountable for the decisions that matter and are honest with investors and borrowers about what the technology is doing. In a market built on trust and a first legal charge, that combination will be worth more than any algorithm on its own.

Invest & Fund has returned over £389 million of capital and interest to lenders with zero losses, showing the rigour that governs our business. To take maximum advantage of this robust and exciting asset class, please visit www.investandfund.com

Don't invest unless you're prepared to lose money. This is a high-risk investment. You may not be able to access your money quickly, and you are unlikely to be protected if something goes wrong. Take 2 minutes to learn more.