Context tells you who to lend to. It does not tell you who will pay.
Embedded finance is right that data at the point of purchase improves selection. Selection is not where unsecured lenders lose money.
In its Credit Conditions Survey for the second quarter, the Bank of England reported that lenders saw default rates on secured lending unchanged, while defaults on unsecured lending rose. Same households, same three months, two different answers.
That is the most useful line published on UK consumer credit this summer, and it is worth holding on to while reading anything about the future of embedded finance or consumer lending in general.
The case for context, stated properly
Financial products are being built into the software where people already are, the market is estimated to have grown from roughly $94bn in 2025 to $115bn this year with forecasts around $251bn by 2030, and the next phase is described as making those services intelligent enough to anticipate what a customer needs and help shape the decision.
A platform that can see transaction history, order patterns, payroll or inventory knows considerably more at the moment of application than a lender reading a bureau file, open banking data (where available) and a stated income. Underwriting improves. Fraud falls. Conversion rises, because the offer arrives when the need does rather than three weeks afterwards.
What the second quarter actually showed
Unsecured defaults did not rise because lenders suddenly became worse at selection. Lenders attributed the increase to tighter credit conditions and a cooling labour market. Those are conditions that arrive months after the loan is written, and they arrive for everybody: the data-rich lender and the data-poor one, the embedded programme and the high street card book.
That is the structural point, and it is not a criticism of anyone’s technology. Context at the point of purchase is a snapshot. It can be an extraordinarily good snap, far better than the one a traditional lender takes. A consumer loan is nonetheless a claim on somebody’s future income, and no snapshot shows the future. The moment that determines whether the loan performs is usually eleven months after the moment the data was captured.
Where unsecured money is actually lost
Not at the application. It is lost in the eighteen months afterwards, in whether arrears are caught in week two or week ten, whether the forbearance offered is one a regulator will look at kindly, whether the collections operation treats a customer in difficulty in a way that preserves both the relationship and the recovery, and whether the funding line behind the book stays available when the arrears rate moves.
The shape of UK consumer stress makes this concrete rather than theoretical. In July there were 664 bankruptcies against 3,820 debt relief orders and 7,442 individual voluntary arrangements. The individual insolvency rate over the twelve months to July ran at 27.8 per 10,000 adults, or about one adult in 360. The debt charity StepChange reported that its clients arriving in January were mostly under 40 and renting, with 71% carrying credit card debt.
None of that is a selection problem. Every one of those people probably passed an affordability assessment at some point. It is a servicing, forbearance and collections problem, and it belongs to whoever holds the book.
The corroboration, and what it is worth
There is a second-order version of the same weather pattern, if it is that and not a blip on the radar. BTG’s Red Flag Alert for the second quarter put UK businesses in critical financial distress at 53,756, up 9.0% on the year, with the sharpest increases in consumer-facing sectors: leisure and cultural activities up 27.1%, hotels and accommodation up 26.5%, sports and health clubs up 21.0%, food and drug retailers up 18.4%.
This may be a useful cross-check and it could be read as one. It measures companies rather than households, so it tells you about consumer behaviour only at one remove, through the businesses that depend on discretionary spending. It corroborates. It does not prove, and it is worth being careful about that distinction when the number is this convenient.
Three capabilities, usually discussed as one
Embedded consumer lending, done properly, needs three things, but the conversation tends to focus on technology.
The first is channel origination. The credit is offered inside somebody else’s journey, at the checkout, in the app, at the moment the need appears rather than three weeks later when the aggregators and banks have offered their “solutions”. That is genuinely valuable distribution, because acquisition cost in consumer lending has historically been brutal.
The second is contextual underwriting. The channel hands the lender data it would never otherwise see, and the credit decision gets better. This is the part the industry talks about, and, as above, it is real. As with leasing the risk is reduced when you know the what/why.
The third is servicing. Arrears handling, forbearance, complaints, collections, and the record of how customers in difficulty were actually treated. It attracts a fraction of the attention but ultimately it decides whether the book performs.
The awkward part is that the first two are naturally distributed and the third cannot be. A platform can originate through ten partners in ten contexts, each with its own data, its own customer relationship and its own commercial terms. It cannot service ten books to ten standards. The permission is singular, the supervisory expectation is singular, and a borrower who has lost hours at work does not care which app they were in when they borrowed.
So the hard problem is not the data
It is running distributed origination alongside undistributed servicing, under one permission, at one standard, across channels that have nothing in common except the lender sitting behind them. Every new partner multiplies the origination surface and adds little to the servicing capability, which has to absorb the whole book regardless.
That expertise and operational efficiency can slow to assemble, expensive to run and genuinely difficult to copy. A consumer credit permission takes time and scrutiny to obtain. A forbearance process that survives supervisory attention takes longer. A collections operation that is both effective and defensible in how it treats people is harder still, and a funding line that stays available when the arrears rate moves…well that is not regularly tested but we have seen how that can change quickly with the weather.
Contextual data is abundant, increasingly replicable and depreciating a little faster each year. It is a good reason to prefer one lender over another. It is not the reason one lender survives and another does not.
Which is why, in this market, we have backed operators rather than datasets. D3T is an investor in this space and we invested on the strength of the founder’s and team’s track record across distribution, underwriting and collections, which is the whole loop rather than the front of it, and not because of any view about contextual data.
D3T Capital LTD is an Appointed Representative of Capricorn Fund Managers Limited, which is authorised and regulated by the Financial Conduct Authority. D3T Capital LTD is incorporated in England and the registered office is at Unit 17 Orbital 25 Business Park, Dwight Road, Watford, WD18 9DA, United Kingdom.
The investment products and services of D3T Capital LTD are only available to professional clients and eligible counterparties. They are not available to retail clients.
This article is general market commentary. It does not constitute an offer, invitation or inducement to invest in any product or service, and it is not directed at retail clients.
