Debt Collection With AI Voice Agents: How Much Can a Small US Agency Earn?

Salient, an AI loan servicing startup that automates collections calls for auto lenders, raised $60 million in a Seed A led by Andreessen Horowitz, and its CEO told Fortune in December 2025 that annualized recurring revenue had passed $25 million (a company claim). A small US agency can use the same voice technology, but it earns money differently: a contingency fee, typically 25% to 50% of what it recovers. In the example math below, a solo version nets about $1,000 a month and a larger version about $90,000, and profit swings hard on the recovery rate.

This is an analysis of public sources, not a promise of income. Every dollar figure labelled "example estimate" rests on assumptions you can see and change.

What is the AI debt collection agency opportunity?

The opportunity is to run a licensed collection agency where AI voice agents make the routine first-touch calls, so each human handles disputes, payment plans and escalations instead of dialing. Lenders and creditors pay you a share of what you recover.

The funded startups show where the money is going. Salient says its software automates collections calls and payment processing for auto lenders, a function it describes as historically dominated by call centers. Veritus, a Y Combinator Summer 2025 company, describes its product as AI agents for loan servicing and collections, including inbound and outbound voice agents.

Those companies sell software to big lenders. A small agency sits one layer down: it takes placed accounts from small creditors (clinics, landlords, utilities, B2B suppliers) and works them with cheaper tooling. That is an inference from how the market is structured, not a sourced forecast.

How much can an AI debt collection agency make?

In the example estimates below, a solo agency nets about $1,050 a month, a mid-size one about $12,250 and a large one about $90,250. Profit depends mostly on how much of the placed balance you actually recover.

Assumptions (example estimate): you earn a contingency fee of 30% (28% at the large size), within the 25% to 50% range quoted by AgentCollect, Cornerstone Licensing and Fair Capital. Each account gets 5 voice calls a month at 1 minute each, priced at the $0.11 per minute example on Retell AI's pricing page as of 15 October 2026. Five calls a month stays under the Regulation F presumption limit of seven calls in seven days. Recovery rates, staff costs and overhead are my assumptions, not sourced figures.

Staff cost assumes $5,000 per person per month fully loaded: 0.3 of a contractor at the small size, 2 people at medium, 8 at large.

Now stress the recovery rate. At 5% instead of 8%, the small version collects $10,000, earns $3,000 in fees and loses $750 a month. At 6% instead of 10%, the medium version earns $18,000 and makes only $250. The lesson: voice minutes are cheap, and the real risks are low recovery, placement volume and compliance staff.

Every figure above depends on clients placing the assumed volume of accounts each month. Nothing here is guaranteed.

Which companies are already doing this?

Five companies sell AI collections or compliance tools. None of them is a small agency, and the four with reported funding raised far more than a $5,000 to $27,000 launch budget.

Reading the table honestly: the sourced funding went to software companies that sell to lenders. What they did with millions was build and sell platforms. What you can do with a budget in the $5K to $27K range (see the start-up costs below) is a service business: license, buy platform minutes, take a few clients on contingency and prove recovery rates before spending more.

Sources differ on Prodigal's Series A lead. The company's own press release says Menlo Ventures led; YourStory credits Accel and Menlo Ventures. This post follows the press release.

Do you need a licence or certification?

In many states, yes. Washington's Department of Licensing says you need a collection agency licence if your business solicits, collects or attempts to collect debts on behalf of a third party.

State examples (check each state where your debtors live, because rules differ):

• Washington: a $5,000 surety bond, at least $7,500 in cash or equivalent (with $5,000 deposited in an operating account) and at least $7,500 in equity or net worth, per the state's licensing page.
• Texas: according to Bond Exchange, third-party debt collectors must file a $10,000 surety bond, with bonds from about $100 a year.
• California: according to Surety Solutions, the DFPI licenses collectors under the Debt Collection Licensing Act with a minimum $25,000 bond, and good-credit applicants pay around 1% to 3% of the bond amount.

The federal layer applies on top. Regulation F presumes compliance if you do not call a person about one debt more than seven times in seven days, or within seven days after a phone conversation about it. Absent knowledge of other circumstances, before 8:00 a.m. and after 9:00 p.m. local time at the consumer's location is an inconvenient time.

The AI-specific rule matters most. The FCC's February 2024 ruling says the TCPA's limits on artificial or prerecorded voices cover current AI technologies that generate human voices, so such calls require the prior express consent of the called party absent an emergency purpose or exemption. Ask a lawyer whether the consent your client collected covers calls from your agency using an AI voice.

How much does it cost to start?

A bare-bones launch costs roughly $5,000 to $27,000 in the example estimate below, with bonds, licensing capital and legal review as the main items. The sourced items are bond amounts and voice pricing; the rest are my planning assumptions.

How do you start an AI debt collection agency, step by step?

Start with licensing and legal review, then test small. These ten steps run in order.

1. Pick a niche. Choose one creditor type, such as medical offices or B2B suppliers, so your scripts and compliance review stay narrow.
2. Map your states. List where your debtors live and read each state regulator's collection agency page. Washington's licensing page is a model of what to look for: bond, net worth and application documents.
3. Hire a lawyer who handles FDCPA and TCPA work. Have them review your consent language, call scripts and client contract.
4. Apply for licences and bonds in each required state before placing a single call.
5. Write the call flow. Cap attempts below Regulation F's seven-in-seven presumption, block calls before 8:00 a.m. and after 9:00 p.m. in the debtor's local time, and script a clean handoff to a human.
6. Build a voice agent pilot on a pay-as-you-go platform. Test with internal calls first and have your lawyer listen to recordings.
7. Sign two or three pilot clients on contingency. Use a written contract. Ask for placements of aged accounts you can work for 60 to 90 days.
8. Track recovery rate per placement. This single number decides whether the example math above works for you.
9. Add a human for disputes and payments. Route disputes, hardship and cease-contact requests to a person right away.
10. Review monthly with counsel, then raise volume only when recovery and complaints stay acceptable.

Which free tools and resources can you use?

Free options cover research and testing, not production operations.

Last checked: 15 October 2026. Free stops at the moment you need real call volume, call recording storage, a compliance review of every script, or state licences.

How can you use AI in a debt collection agency?

Use AI for first-contact calls, reminders, call monitoring and back-office work, and keep humans on disputes. Voice pricing is the cheap part: Retell lists $0.07 to $0.31 per minute as of 15 October 2026.

• Outbound reminders and payment offers. Voice agents place scripted calls inside your attempt limits. Murphy says its agents work around the clock in more than 30 languages.
• Inbound line. Veritus lists inbound as well as outbound voice agents, so debtors who call back reach an agent rather than voicemail.
• Compliance monitoring. Sedric, per its own announcement, sells an AI compliance platform to financial institutions. Enterprise tools like this may be priced beyond a solo agency, so ask for pricing.
• Follow-up and bookkeeping. Similar tactics appear in our posts on lead follow-up agencies and AI bookkeeping. If you want a simpler voice business first, see how to resell AI receptionists.

How do you scale it faster?

Scale by winning more placements from creditors, not by adding call minutes. Voice minutes are only about 15% to 18% of monthly cost in the example math (for instance $13,750 of $77,750 at the large size).

• Hiring: add dispute and compliance staff before volume rises. Your staff line is the biggest cost in the medium and large examples.
• Marketing: direct outreach to small creditors in your niche, plus referrals from clients. Fee quotes of 25% to 35% for fresh accounts and 40% to 50% for old ones, per AgentCollect, give you a pricing menu to test.
• Where free tools stop: once you hold client funds or cross state lines, you need paid compliance review, insurance and licences in each state.
• Related models: if you want recurring invoice work with less regulatory weight, see landlord bill pay automation.

What are the risks?

The main risks are legal exposure, low recovery and thin margins at small scale. The example math shows a solo agency can go negative if recovery drops.

• Consent risk. The FCC says AI-voice calls need prior express consent of the called party absent an exemption.
• Contact-limit risk. Regulation F's seven-in-seven rule is a presumption, not the only test, so call patterns that look intended to harass can still create exposure.
• Business survival. According to Start Business by State, which analyses BLS data, 77.9% of new employer establishments survive year one and 51.4% survive five years. That is an all-industry figure, not collection agencies, and exits are not always failures.
• Client concentration. One creditor can be most of your revenue, and contingency means you earn nothing on accounts that do not pay.
• Debt buying. The FTC's study found an average price of 4.0 cents per dollar across more than 3,400 portfolios, but the report is from 2013 and buying portfolios puts your own capital at risk.

What's still unclear?

Several things are not answered by public sources.

• Agency profit margins. No regulator or trade-body figure for small-agency profit is available, so the margins above are assumptions.
• Collection-specific failure rates. Only all-industry survival data is available.
• Fee ranges. Sources disagree: Fair Capital cites an average of 25% to 35%, while AgentCollect and Cornerstone cite 25% to 50%. AgentCollect also sells AI collections, so it has a commercial interest.
• State bond figures. The Texas and California numbers come from surety brokers, not the regulators. Confirm with each state.
• Startup claims. Salient's ARR and valuation are company or insider statements. Veritus has not published a funding amount on its YC page.
• AI-specific consent practice. The FCC ruling is clear that consent is required, but how that applies to third-party collectors calling on a creditor's behalf is a question for counsel.

This is research, not legal, tax or financial advice.