CollectWise does the whole collections job with AI agents

Here is what usually happens when a consumer debt goes unpaid.

The creditor's internal team sends a few letters and makes a few calls. After ninety days or so, the account gets handed to a collection agency, which pays a call center full of people to work through dialer queues, one conversation at a time. Most calls go to voicemail. Most voicemails get ignored. The agency keeps a third or more of whatever it recovers, and on old debt it recovers very little. Accounts that would actually justify legal action mostly never get it, because figuring out whether a given person is worth suing costs more in attorney time than the debt is worth.

This is a $35 billion industry in the US, and almost all of it still runs on human labor and phone scripts.

CollectWise is building the version where software does the whole job. The company, founded in 2024 and part of Y Combinator's Fall 2024 batch, deploys autonomous AI agents that handle consumer debt recovery end to end: the first outreach, the phone conversations, the payment plan, the credit bureau reporting, and, when an account calls for it, the legal escalation through garnishments and liens. CollectWise reports that its agents recover twice as much as human collectors while cutting collection costs roughly in half.

That last clause is the part worth sitting with. Plenty of startups have bolted a chatbot onto collections. CollectWise is making a different claim: that the entire function, including the parts that end in a courtroom, can be run by machines.

A labor business with software margins waiting inside it

Collections has resisted automation for a long time, and the reasons are structural rather than technological.

The work is conversation-heavy. Recovering a debt usually means reaching a specific person, at a moment they're willing to engage, and negotiating something they can actually afford. Call centers solve this with volume: more dialers, more agents, more shifts. Every incremental conversation costs incremental payroll, which is why agencies charge contingency fees of 25 to 50 percent and why small-balance accounts often aren't worth working at all.

The work is also legally dangerous. The Fair Debt Collection Practices Act governs when a collector can call, what they can say, what they must disclose, and how disputes must be handled. States layer their own rules on top, and they differ. A single agent having a bad day can say one wrong sentence and generate a lawsuit that eats the margin on a thousand good calls. The industry's answer has been scripts, call recordings, and compliance departments, which is to say more labor.

Put those two constraints together and you get the economics the industry has lived with for decades: expensive recovery, thin coverage, and a strong incentive to focus only on the easiest accounts.

CollectWise's bet is that AI agents dissolve both constraints at once. Software can hold ten thousand conversations simultaneously, across email, SMS, and voice, at marginal cost near zero. And software never improvises. The rules it follows on call ten thousand are exactly the rules it followed on call one.

What the agents actually do

CollectWise's platform takes an account from placement to resolution without a human in the loop.

Outreach that adapts to the person. The system studies each consumer's financial profile and builds an individual contact strategy: which channels, what cadence, what kind of settlement or plan to offer. A recent graduate with a gym debt and a homeowner with a defaulted auto loan get different conversations, because they should. The models keep learning from what works across the portfolio.

Real conversations, on every channel. The agents send personalized email and SMS, and they talk. A consumer can call back at 11pm on a Sunday and reach an agent that knows their account, can negotiate a payment plan within the creditor's parameters, and can take the payment on the spot. That around-the-clock inbound coverage matters more than it sounds: people tend to deal with their debts in private moments, not during a call center's business hours, and a collector that's awake when the consumer is captures payments a 9-to-5 operation misses.

Consequences, applied automatically. When outreach alone doesn't resolve an account, the platform handles credit bureau reporting and pre-legal notices. Past that, it automates the legal pipeline itself: researching whether a debtor has assets worth pursuing, preparing litigation, and enforcing judgments through garnishments and liens.

That third layer is the unusual one. Legal escalation is where every other collections tool stops and hands the file to a law firm, because asset research and judgment enforcement have always required expensive professional attention per account. Automating that step changes the math on an enormous pool of debt that was previously written off, since the question "is this account worth suing over" now costs nearly nothing to answer.

It also quietly changes the incentives for everyone. When legal follow-through is rare and expensive, ignoring a collector is often rational. When it's systematic, the friendly payment plan offered in the first text message becomes a genuinely good deal, and more people take it. The credible path to enforcement is part of why the polite early conversations work.

Compliance is a feature machines are good at

The counterintuitive part of AI collections is that the robot is plausibly the better-behaved party on the call.

CollectWise builds FDCPA and state-by-state rules directly into its agents' behavior. The software tracks contact frequency limits, honors disputes and do-not-contact requests, says the required disclosures every time, and keeps its tone professional in every message, because the tone is a design decision rather than a mood. A human collector six hours into a shift, paid partly on commission, has incentives and frustrations that leak into conversations. The agent has neither.

There's a consumer-experience argument hiding in here too. Collections is a stressful transaction on both sides, and much of what people hate about it, the repeated calls at bad times, the pressure tactics, the feeling of being processed, comes from the economics of human-powered dialing. An agent that contacts you on the channel you actually respond to, lets you set up a plan by text at midnight without talking to anyone, and never raises its voice is a better experience than the status quo. CollectWise works for creditors, but its approach only scales if consumers will engage with it, and self-serve, low-pressure resolution is what gets them to.

Why this works now and didn't in 2020

Collections software is not a new idea. The industry has had dialers, skip-tracing databases, and payment portals for years, and an earlier generation of startups tried text-message outreach with templated scripts. What none of them could do was negotiate.

Negotiation is the actual job. A collection conversation is a back-and-forth about a specific person's circumstances: what they owe, what they dispute, what they can pay, and when. Scripted bots hit the edge of their decision tree within two exchanges and punt to a human, which put a hard ceiling on how much of the workflow older tools could absorb. Modern language models removed that ceiling. An agent can now hold an open-ended conversation, by voice, that stays inside the creditor's settlement parameters and the law's disclosure requirements at the same time.

That's the capability shift CollectWise is built on, and it explains the company's timing. The founders started it in 2024, almost exactly when voice models crossed the threshold of handling emotionally loaded, unstructured phone calls without embarrassing their operator.

Small team, fast traction

CollectWise was founded by Sean O'Brien and Vivek Isukapalli and is based in New York. O'Brien, the CEO, came up through finance and computer science at the University of Virginia and had already built a six-figure digital agency and worked on another fintech before this one. Isukapalli is a Carnegie Mellon engineer. It's a fitting pairing for the problem: collections sits exactly at the intersection of financial operations and hard automation engineering.

The early numbers suggest the thesis is landing. Within months of launch, the company reached a $2 million annualized run rate with a team of five, selling to creditors, debt buyers, and collection agencies. It has since grown to around eight people, which still makes for a striking ratio of revenue to headcount in an industry whose incumbents measure capacity in seats.

That ratio is the product demonstrating itself. A collections operation that scales with compute instead of hiring is the entire pitch, and CollectWise's own P&L is the first proof.

The quiet end of the call center model

Debt collection is the kind of market that rarely makes headlines but touches an enormous number of people: tens of millions of Americans have an account in collections at any given time. The industry that works those accounts has looked the same for fifty years because its binding constraint, the cost of a human conversation, never moved.

That constraint just moved. The cost of a conversation is collapsing toward zero, and software can hold millions of them at once without a single one drifting off script. The firms built around rooms full of dialers will have to compete with systems that recover more at lower cost and don't generate compliance complaints.

CollectWise is young, and the incumbents it's challenging hold long-standing client relationships and state licensing moats. But the direction only points one way. Collections is repetitive, rule-bound conversation work, which is what AI agents happen to be best at. Somebody was going to rebuild this industry as software. CollectWise is one of the first to do the whole job, liens included.