Audun
RegTech
Debt collection is the part of the credit lifecycle nobody wants to look at. It is also where the reputational damage happens: a lender spends years building a brand and then outsources the hardest conversations its customers will ever have to an agency paid on recovery, working from scripts, sending anonymous letters. Audun's pitch is that this arrangement is a design failure rather than an inevitability — that most of what collections agents do is busywork an AI can run properly, and the small remainder is exactly the part that needs a human with full context. Its tagline, "a calmer way to settle things," is doing real strategic work: the company is selling recovery rates and reputational safety to creditors, on the argument that those stopped being a trade-off once the outreach could be personalised at scale.
The described workflow follows the escalation ladder deliberately. Before anything becomes a collection case, the system sends timed reminders on whichever channel a person actually uses, aiming to settle invoices at the pre-collection stage. Where a debt does age, it negotiates a payment plan sized to the person's month rather than the creditor's preference, then logs and schedules it. And where a case needs judgement — a dispute, a customer flagging financial hardship, anything that doesn't fit a script — the automation pauses and hands off to a named human case handler with the full conversation history attached. Creditors get an operator console showing every stage. The stated design principle, that the machine should know when to step back, is the interesting one, because it is precisely where this category has attracted criticism.
That context belongs in any honest assessment. AI collections is a well-populated YC category — Y Combinator has incubated roughly six debt collection and settlement startups in six years, including Altur, CollectWise, and Domu, the last of which reported its agents hitting tens of millions of connected calls a month. It is also the subject of active scrutiny: WIRED reported in May 2026 on AI collection agents pursuing debts that had already been settled, with consumers unable to escalate to a human. Audun's architecture reads as a direct answer to those failure modes rather than an example of them, and its Nordic starting point cuts the same way — Norwegian and Swedish debt collection are licensed activities with statutory codes of conduct and capped fees, among the strictest regimes in Europe. Building for that market first is a meaningful constraint to accept voluntarily.
Beyond that, Audun is genuinely early and the public record is thin. It is Y Combinator-backed, which is real third-party validation, and its own writing suggests unusual domain depth — a July 2026 essay arguing that credit models inherit a dependency on the servicing and collections policy under which their training data was generated, making a change in collections policy a source of model risk rather than model error. That is a practitioner's argument, not content marketing. But the company has not published its founders, legal entity, licence status, headcount, or funding; the debtor-facing self-serve portal is still marked as coming; and the product screenshots use illustrative data. What exists is a credible team with a sharp thesis attacking a category that deserves attacking, at a stage where almost nothing is verifiable from outside.
Founded 2021