AI in lending: build the edge, not the engine

AI is changing lending for the better

AI is supercharging productivity across lending, and it's only getting started. Loan teams that use it well will run bigger books, answer brokers faster and give funders better information, with the same people.

So the question isn't whether to use AI. It's how to get the most from it, safely. Our answer: run AI on top of a robust calculation engine. Let AI do what it's brilliant at, reading, matching, drafting and answering questions, and let a proven loan engine do the maths. Together, they're far more powerful than either on its own.

Where AI supercharges a loan team

AI is already changing the day-to-day work of a loan team:

• Documents in, without retyping. It reads a facility letter, valuation or monitoring report and lays out the key terms for someone to check.
• Payments matched. Wrong reference? It suggests the loan, and why, for a person to confirm.
• First drafts done. Arrears letters, broker updates and funder commentary, ready for review.
• Answers in seconds. "Which Manchester loans exit before March, and what's our exposure?" No more spreadsheet afternoons.

AI does the legwork. People still own the decision.

Pair it with an engine you can prove

Lending has one non-negotiable: every redemption calculated the same way, every time, whether you run 15 loans or 1,500. And provable, today and in two years' time when a funder or auditor asks.

That's where a calculation engine comes in. AI works by predicting the most likely answer, which is what makes it so fluent and flexible. A loan engine follows fixed, tested rules, which is what makes it exact. Each is at its best doing its own job.

So the winning combination is simple: tested code does the maths, AI does everything around it. Interest, default interest, fees, allocations and redemption figures come from the loan engine. AI reads the inputs, explains the outputs and flags anything worth a second look.

The loan record is the real prize

AI is only as good as the data it can reach. If the answer to "what does this borrower owe on Friday?" lives in a spreadsheet, a shared drive and someone's inbox, even the best AI will struggle.

A proper system of record holds every term, drawdown, payment and fee, with dates. It watches the book: the exit three weeks away with no refinance in sight, the drawdown that doesn't match the monitoring report, the interest reserve running low. Every alert comes from a rule you can check.

It's the foundation that lets AI shine, and it's what funders and auditors want to see anyway.

Should we build it ourselves with AI?

For the things that make you different, absolutely. For the core of your loan book, it's worth thinking through what you'd be taking on.

AI has made the visible part of software cheap. A developer, a few days, and you have a demo with screens and a loan table. But that's the first 5% of a loan system. The other 95% is invisible, and the real question isn't whether you can build it. It's whether you want to own it, run it and answer for it for the next ten years.

• The maths is harder than it looks. Term extensions, restructures, part-redemptions, default interest from the right date, fees on the right base. AI will write code for each in minutes. Knowing it's right is the hard part. Miss a few weeks' interest on a £1m loan at 1% a month and that's £5,000 to £10,000 gone, and once it's on a redemption statement, it's too late.
• Funders will ask what you run on. Lenders, forward-flow partners, auditors and buyers all check the loan system in due diligence. Interpath says the diligence bar has risen for good and no acquirer will take a loan book at face value again (Property Reporter). Expect them to ask how it's tested, supported and controlled.
• It lives in one or two heads. When they're on holiday, nobody owns it. When they move on, it can be hard for anyone else to pick up.
• Controls are easier to build in than to add later. Approvals, permissions and audit trails are easy to skip in a fast first version. The FCA has just stopped six unregulated lenders from new lending over their controls. Having them from day one is far simpler than adding them under pressure.
• Borrower data needs protecting. ID documents, bank details, property data, all under UK GDPR. Security, testing and backups are now your job, and so is the responsibility for it.
• Switching later is hard. Migrating a live book, with its history and audit trail, costs far more than starting right.
• It takes good people away from lending. A home-built system needs one or two people on it, permanently. Most lenders would rather they were writing loans.

Lessons from AI-built software

AI app builders are genuinely impressive, and improving fast. Like any young technology, they've had growing pains, and the last 18 months show why the foundations matter when apps hold real data.

In May 2025, a scan of 1,645 Lovable-built apps found 170 with no database access controls, leaving names, emails, payment details and API keys open to anyone (Superblocks). Two months later, a Replit coding agent deleted a live production database during an explicit code freeze (heise). In February 2026, an AI-written login check on a Lovable app worked the wrong way round, exposing 18,697 user records (The Register). In April, any free Lovable account could read other users' source code and customer data (Bastion).

None of this is a reason to hold back on AI. It's a reason to put it on the right foundations.

The best of both, with kennek

For a lender's board, the question isn't whether to use AI. It's what that AI runs on.

That's the job kennek does. It's the loan platform built for bridging and development lenders: the calculations proven on live books, approvals and a full audit trail built in, loan-level reporting your funders recognise, and a dedicated team accountable for it under contract. Every non-standard scenario another lender has hit is already handled, so you don't learn those lessons at your own expense.

Lenders are already seeing the difference. Hope Capital reported this month that its loan support specialists now manage twice as many loans on kennek, with redemption statements produced almost instantly.

Build on kennek, not instead of it

This isn't AI tools or kennek. It's both, each doing the right job. Use AI builders for what makes you different: a broker portal, an internal dashboard, a custom funder report. Connect them to kennek through its API. Keep the system of record, the interest maths and the audit trail on a platform built for the job.

And you don't have to build anything to put AI to work on your book. kennek's MCP connector (MCP is the standard way AI assistants such as Claude plug into business software) lets your team ask questions in plain English: "What's the redemption figure on this loan for Friday?", "Which loans are in arrears?", "Show me the portfolio by exposure." The AI asks; kennek's loan engine does the maths. So the answer is the same tested figure your team would get in kennek, and the AI reads and quotes your loans rather than changing them.

You get the flexibility, without having to carry the core yourself.

Six questions to get the most from AI

• Which figures does the AI calculate, and which come from tested code?
• Can you show how a redemption figure was reached, line by line, six months later?
• Who maintains it when the person who built it is on holiday, or leaves?
• What will your funders, auditors or a future buyer ask to see?
• Where does your loan data go when the AI reads it?
• When the AI is wrong, what catches it?

Get these right, and AI becomes one of the best investments a lender can make.

Embrace it, on the right foundations

AI is an amazing tool for lenders. It frees loan teams from retyping and chasing, answers questions in seconds and lets the same team run a bigger book. The lenders who win will be the ones who use it boldly, on a foundation they can trust.

Build what makes you different. Run your loan book on a proven engine. And let AI supercharge both. If you'd like to see how lenders like Hope Capital are doing it, we'd love to share what we've learned.

Sources

• 27 Oct 2026: Hope Capital doubles loan management capacity with Kennek software, Bridging Loan Directory. Why it matters: a named bridging lender showing what a structured system of record does for team capacity.
• 23 Oct 2026: FCA restricts 6 firms following concerns over anti-money laundering controls, FCA. Why it matters: the regulator is now acting on unregulated lenders' controls, including a bridging lender, so a weak audit trail is a live risk.
• 29 Sep 2026: The State of AI Among Mortgage Lenders and Servicers, AARMR, MBA and BCG. Why it matters: one in four staff use unapproved AI tools, showing adoption is running ahead of governance.
• 8 Jun 2026: Tighter governance and M&A set to reshape UK bridging finance market in 2026 (Interpath & BDLA Bridging Market Survey), Property Reporter. Why it matters: the sector's own survey says funders now favour lenders with strong governance and platforms that scale, and that buyers no longer take loan books at face value.
• 20 Apr 2026: Lovable data breach April 2026, Bastion. Why it matters: a platform-wide failure that exposed customer data in apps their builders had configured correctly.
• 27 Feb 2026: AI-built app on Lovable exposed 18K users, researcher claims, The Register. Why it matters: AI-written access logic can look right and still work the wrong way round.
• 3 Sep 2025: Lovable vulnerability explained: how 170+ apps were exposed, Superblocks. Why it matters: insecure defaults in one AI builder left one in ten apps scanned open to anyone.
• Jul 2025: Vibe coding service Replit deletes production database, heise. Why it matters: an AI agent with access to live data can change it without permission, even under explicit instructions not to.
• 21 Nov 2024: Artificial intelligence in UK financial services, 2024, Bank of England and FCA. Why it matters: the UK regulators' baseline on AI use, showing three in four firms already use it.

First published in The kennek Brief, issue 02, on LinkedIn. Subscribe there for the next issue.