The FCA has just released its H2 2025 complaints data, and the headline number — 3.8 million complaints across UK financial services — is the kind of figure that's easy to read as bad news.
Look more closely and you can see something far more interesting in the shape of the data: an entire sector arriving, all at once, at the moment where AI stops being a productivity conversation and becomes a strategic one.
The bulk of those 3.8 million complaints sit in motor finance, following the discretionary commission disclosure ruling. That concentration is worth pausing on. It tells us the next era of regulated lending will not be defined by who can hire the most case handlers. It will be defined by who builds the capability to put their own institutional knowledge to work at machine speed — what Tom Blomfield at Y Combinator has been calling a company brain. His framing is sharper than most of the conversations happening inside lenders right now: "The biggest blocker to AI automation of companies is no longer the models — it's domain knowledge." A company brain, in his words, "isn't a company-wide search or a chatbot over documents." It's "a living map of how a company works: how refunds get handled, how pricing exceptions are decided, how engineers respond to incidents." YC has gone as far as putting it on its Requests for Startups list — a public bet that this is the next primitive every serious company will need.
For motor finance lenders sitting on millions of inbound complaints, the relevance is immediate. The firms that build their own version of this capability now will quietly set the terms for everyone else.
The market that motor finance lenders are responding to
It's worth being precise about what's actually happening on the demand side, because most commentary on the motor finance commission story misses it.
The FCA has had to launch a four-regulator taskforce — itself, the SRA, the ICO and the ASA — specifically to tackle poor practices by claims management companies and law firms operating in motor finance. Since January 2024, the FCA has had more than 800 misleading adverts removed or amended, freed 28,000 consumers from contracts they didn't really agree to, and forced three CMCs to cap their fees, protecting another 500,000 people. The regulators are doing important work.
But the underlying dynamic matters for the lenders on the receiving end. CMCs have industrialised complaint generation. They run templated submissions, automated marketing funnels and large-scale sign-up flows. Multiple representatives sometimes file for the same consumer on the same claim. The volume hitting motor finance lenders right now is not the natural product of consumer dissatisfaction — it's the product of an automated demand engine sitting on the other side of the table.
What the H2 2025 data really exposes is the asymmetry: industrial volume on one side and human-led case handling on the other. No hiring plan can close that gap. The market is asking lenders for a fundamentally different capability — and that's where the opportunity lives.
No company brain without quality data
Before any of this works, there is a precondition that the AI conversation tends to skip past: quality data.
A company brain is only as good as the data it draws on. An AI layered over inconsistent case notes, fragmented policy memos, scattered prior adjudications and unstructured outcome fields will do exactly what you'd expect — produce confident-sounding answers that are, at best, unreliable, and at worst, regulatory-grade dangerous.
The institutional knowledge inside a serious motor finance lender is enormous. It is also, almost universally, locked in formats and systems that were never designed to be machine-readable. Case management systems with free-text resolution fields. Internal interpretation memos sitting in three different SharePoints. Prior outcome data recorded inconsistently across teams and time periods. Decade-old precedent buried in case files no-one has time to reach into.
Building a company brain begins with the data. Cleaning it, structuring it, connecting it, and — critically — keeping it current as new outcomes flow in. It's the unglamorous half of the work, and it's also where the real moat sits. Models are increasingly commoditised. Quality, well-structured institutional data is not. The lenders that take the data work seriously now will find every subsequent AI initiative compounds on top of it. The ones that try to skip the data work will end up with a chatbot that can't be trusted near a redress decision — and they'll have correctly diagnosed why.
This is the part of the conversation that should be happening at exec level today. Not "what AI vendor do we pick?" but "what do we need to do to our data so that AI is actually useful when we deploy it?"
What a company brain really does — clinical-grade triage for complaints
Once the data foundation is in place, the use case becomes much clearer — and a lot more grounded than the public AI conversation tends to suggest. The interesting use case isn't "let an LLM auto-resolve a £30,000 redress claim." Nobody serious is proposing that, and the regulator would rightly never wear it.
The right way to picture it is closer to clinical triage in healthcare. When a patient walks into A&E, a triage nurse rapidly assesses severity, urgency, complexity and novelty, then routes them to the right path: minor injuries clinic, resuscitation bay, or specialist referral. The triage doesn't replace the doctor. It makes sure the doctor gets to the right patient with the right information in the right order.
A company brain does the same thing for inbound complaints. When a submission lands — whether from a real consumer or a CMC funnel — the AI rapidly assesses it against the lender's own data: is this a duplicate of three other submissions we've already received for this consumer? Does the fact pattern match cases we've redressed under bucket X? Is the underlying claim merit-bearing, or is it a templated submission with no factual hook? How urgent is it given the regulator's response windows? And critically — does this look like one of the genuinely novel cases that needs senior human attention?
That triage step is also how you cut through the noise. The CMC volume is, by design, a mixture of legitimate claims and industrial-scale noise. Quality data and a working company brain let the lender see the signal and the noise as distinct from the moment a complaint arrives — instead of paying for both at the same per-case rate.
For the cases that match a known redress pattern, the AI surfaces the prior offer template, the relevant policy memo, the firm's own historical position, and drafts the response with that record attached for human approval. For the genuinely novel cases, the AI flags them, routes them to a senior human, and that human now has the firm's full memory in front of them at the point of decision rather than buried somewhere they don't have time to search.
This is human-in-the-loop by design. The AI doesn't decide. It equips the decider — every time, consistently, with the firm's full record at their fingertips.
Why this is exactly what good outcomes regulation wants
The first instinct from a compliance team will be to ask whether AI in the complaint flow creates Consumer Duty exposure. It's the right question. The answer, when you actually work through it, is the opposite of what most people assume.
Consumer Duty is, at its core, a call for consistent, transparent, well-evidenced outcomes. A company brain is one of the most direct tools there is for delivering on that ambition. Every decision draws on the same institutional record. Every output carries an audit trail. Every novel case is escalated rather than guessed at. The reasoning behind a redress decision becomes legible — to the firm, to the regulator, and to the customer.
Far from being a Consumer Duty risk, this kind of capability is what the regulation is reaching for. The lenders who get there first will find Consumer Duty assurance much easier to demonstrate, not harder.
Can motor finance be the proving ground for what comes next?
This is the question worth asking openly, because it has implications well beyond motor finance.
As a use case, regulated complaint handling has every property a serious AI investor — Y Combinator very much included — looks for: it's regulated, knowledge-bound, repeat-pattern, expensive, and currently unsolvable by adding people. Every resolved case sharpens the pattern library for the next one. The architecture that works for motor finance redress works for the next misselling cycle, packaged products, insurance redress, and adjacent regulated sectors with the same shape of problem.
The open question is who actually builds it first. The incumbents — the major motor finance lenders, the high-street banks with motor books, the specialist consumer credit firms — have an enormous advantage on paper, because they hold the deepest institutional knowledge in the sector. They've adjudicated thousands of cases, written the internal memos, lived the regulatory cycles. That knowledge is real. The risk is that it stays locked in legacy systems that were never designed for AI to read.
Sitting on the other side are smaller, AI-native firms — newer lenders, fintech challengers, and a wave of YC-backed startups — that don't have decades of institutional precedent, but do have clean data architectures, modern tooling, and AI in their DNA. If those firms move quickly enough on the use case, the incumbents' knowledge advantage becomes a knowledge liability — a moat full of value that can't actually be reached.
The race isn't decided. There's a genuine window for the incumbents to lean into the work — to invest in data quality, to build their own company brains, and to put their hard-won institutional knowledge to work before the field rebalances. But it is a window, not a permanent state. The lenders who treat this as a strategic priority right now are the ones who'll come out the other side ahead.
The strategic question for lenders right now
The most ambitious lenders in the market — the ones whose CEOs and COOs are already asking the hard questions about how AI changes their cost-to-serve, their consistency, and their resilience to the next regulatory shift — are well past the "wait and see" stage. They're sketching out what their company brain looks like, what data feeds it, where the human stays in the loop, and how it integrates into the operations they already run.
That's the conversation worth having. Tom Blomfield reckons every company in the world will eventually need a brain of this kind. In regulated lending, where institutional knowledge is the asset that matters most, that timeline is much shorter than it sounds.
The H2 2025 data doesn't tell us motor finance is failing. It tells us motor finance is where the next era of regulated lending is being figured out, in real time, by the firms that decide to lean into it.
The arms race on the demand side isn't going away. The lenders who use this moment to invest in the data foundation, build the company brain, and put their decades of hard-won institutional knowledge to work are the ones who'll define what good looks like for the rest of the sector.
It's the use case to build through, and the time to build is now.
← Back to News & Opinion