Skip to content
Industry

You don't recover more by dialing more.

An AI operating system for collection agencies — outreach that works every account inside the rules, and guardrails that still hold on contact four hundred.

Scroll
Recovery & Compliance Focus

Regulation F capped the dial: seven attempts in seven days per debt, a seven-day freeze after every conversation, and a consumer who can switch off one channel without switching off you. The agencies growing through that ceiling aren't dialing harder — they're working every account through digital channels a collector never touches, and spending human hours only where a human changes the outcome. We build that operating system: placement-day segmentation, omnichannel outreach with the contact rules compiled into the system instead of trained into staff, self-serve arrangements a consumer can accept at midnight, voice agents that never skip a disclosure, and QA across every contact instead of the sliver a manager can sample. Compliance here isn't a review step at the end. It's the material the system is built out of.

Debt Collection Agencies
In practice

Where AI earns its keep.

The workflows in debt collection agencies where AI does real work today — not a someday roadmap, the jobs it can take off your plate now.

01

Omnichannel outreach engine

Decides who to contact, when, on which channel, and with what message — inside attempt caps, quiet hours, and per-channel opt-outs it enforces itself.

02

Digital self-serve resolution

Hosted portals and negotiated payment plans that let a consumer settle an account at midnight without ever reaching a collector.

03

Voice & chat collector agents

Right-party contact, required disclosures, and payment capture handled identically on call four hundred and on call one.

04

Compliance QA on every contact

Every call, text, and email scored against your policy and the rules — with the audit trail that answers a client or an examiner.

How it actually runs

A placed account, from file to resolution

Not a feature list — the mechanism, step by step, from trigger to logged outcome.

01

Ingest

The placement file lands and is normalised on arrival. Phones and addresses are validated, and litigator, bankruptcy, deceased, and DNC scrubs run before a single attempt is spent — accounts that should never be worked are held automatically, not caught later in review.

02

Segment

Balance, age, contactability, and prior payment behaviour decide whether the account goes digital-first, voice-assisted, or straight to a senior collector. The routing is inspectable and overridable, and it re-scores as behaviour changes rather than being fixed on placement day.

03

Reach

Email and SMS carry the validation notice and a link to resolve. These channels don't consume call attempts, so the cheapest contact always happens before the most expensive one — and every send passes the attempt cap, quiet-hours, and opt-out gate first.

04

Resolve

The consumer accepts a plan in the portal, or an agent captures one inside your client's authority matrix. Disputes, hardship, attorney representation, and bankruptcy stop outreach on the spot and route to a licensed human.

05

Furnish

Payments post back to your platform of record, Metro 2 furnishing runs with dispute and status codes intact, and QA scoring plus client reporting generate themselves — with the policy version that authorised every contact still attached.

A day in the system
collections-os.log
08:02intakePlacement file 4471 — 3,208 accounts ingested, 214 flagged bad-address
08:04agentScrub complete — 61 litigator, 9 bankruptcy, 3 deceased: held from outreach
08:11agentSegmented — 2,104 digital-first · 826 voice-assisted · 58 human collector
09:00policyQuiet-hours gate — 412 accounts held until 08:00 local (PT, MT)
09:14agentValidation notices sent — 1,880 delivered, 41 bounced, addresses invalidated
11:47agentAcct 88213 — consumer built and accepted a 6-month arrangement, $50/mo
13:20policyAcct 71904 — attempt 7 of 7 this week. Dialing suspended on this debt
14:35agentAcct 65022 — dispute received. Outreach ceased, furnishing flagged
16:02qa1,144 contacts scored — 3 queued for review, all on disclosure timing
18:00reportClient 22 nightly — $41,905 posted, 96 new arrangements, 0 policy exceptions

Illustrative — the events this system emits, not a log from any client’s production run.

What we automate

Every part of it. Not the demo part.

The whole operation, workstream by workstream — including the unglamorous pieces that decide whether any of the rest of it works.

01
~/step-1.md

Placement intake & data hygiene

Recovery is decided before the first attempt. Files are normalised on arrival, phones and addresses validated against current data, and every account passed through litigator, bankruptcy, deceased, and DNC scrubs. Attempts stop being spent on accounts that were never collectable, and the accounts that must not be touched at all are held by the system rather than by someone remembering.

02
~/step-2.md

Segmentation & treatment strategy

Not every account deserves the same effort, and treating them alike is how agencies burn their cheapest attempts on their worst paper. Balance, age, channel history, prior payment behaviour, and contactability decide the treatment path — and because it re-scores continuously, an account that starts digital-first and goes quiet can be promoted to voice without anyone running a report.

03
~/step-3.md

Contact orchestration inside Regulation F

The attempt cap, the seven-day freeze after a conversation, quiet hours in the consumer's own time zone, and per-channel opt-out state live in one control layer that every outreach request must pass through. A collector cannot accidentally exceed a limit, because the request is never issued — and afterwards the system can show precisely which rule authorised or blocked each contact.

04
~/step-4.md

Digital self-service & payment arrangements

A large share of consumers would rather resolve a debt at midnight than talk to anyone, and published results from digital-first collectors show the overwhelming majority of accounts resolving with no human contact at all. We build what sits behind that: real-time balance, a plan builder constrained to your client's authority matrix, tokenised card and ACH capture, and automatic re-engagement the moment a scheduled payment fails.

05
~/step-5.md

Voice agents for right-party contact

Voice still matters where digital goes unanswered. An agent opens with the required disclosure, verifies it has the right party before saying anything about the debt, handles the objections your collectors hear all day, and captures payment — the same way on every call, at any hour, in English or Spanish. Where it is uncertain, it warm-transfers instead of improvising.

06
~/step-6.md

Negotiation inside an authority matrix

Settlement is where an unguarded model becomes a liability. Discount floors, term limits, minimum down payments, and per-client rules are enforced as hard constraints outside the model, so the agent negotiates only within the envelope your client agreed to. Anything beyond it becomes a request a human approves — never a promise you are then bound to honour.

07
~/step-7.md

Disputes, cease requests & hardship

A dispute, cease-and-desist, attorney notification, bankruptcy indication, or hardship disclosure is a hard stop, not a sentiment score. Outreach suspends immediately, the workflow your compliance team defined opens, and the furnishing flag is set — because furnishing disputed information without the dispute attached is precisely the FCRA failure that has produced CFPB consent orders against collectors.

08
~/step-8.md

Compliance QA across every contact

Manual QA samples a couple of percent of calls and misses the rest by construction. Here every contact — voice, SMS, email, portal — is transcribed and scored against required disclosures, prohibited language, timing, and consent state, with exceptions queued for a human reviewer. The by-product is the thing that wins placements: an audit trail you can hand a client mid-review without a week of preparation.

09
~/step-9.md

Furnishing, reconciliation & client reporting

Payments reconcile back to your platform of record, Metro 2 furnishing runs with dispute and status codes intact, and per-client reporting on liquidation, cost per dollar collected, contact rates, and exceptions is generated rather than assembled. When a client asks why a portfolio performed the way it did, the answer is a query — not an afternoon.

Why this exists

The numbers aren’t ours.

Published industry figures and the rules themselves, each with its source named — this is the gap the build is aimed at, not a claim about our results.

7 in 7
Call attempts allowed per debt before the rule stops you
12 CFR § 1006.14(b), Regulation F
8am–9pm
The consumer's local time — a system constraint, not a habit
FDCPA § 1692c(a)(1)
98%
Of consumers on one digital-first platform resolved with no human contact
TrueAccord, published 2024
64%
Of collection agencies still offer no self-service option
TransUnion, reported 2023
15–25%
Liquidation lift vendors claim for AI-led strategy
Vendor-published — unverified; we would prove it on your portfolio
>50%
Promise-to-pay conversion in strong shops on first right-party contact
Published ARM call-centre KPI benchmarks
55+Engineers & specialists
300+Projects delivered
8Years in business
5.0★Clutch rating
80% of clients return for a second engagement
Talk it through

Tell us which part of debt collection agencies is the bottleneck.

Book a call
Guardrails

What it will never do.

Enforced outside the model, in the control layer — these are constraints the system can't talk itself out of, not instructions we hope it follows.

  • 01Never dials outside 8am–9pm in the consumer's local time zone — the system resolves the zone, not the collector.
  • 02Never exceeds seven attempts in seven days on a debt, and freezes that debt for seven days after any conversation.
  • 03Never contacts on a channel the consumer opted out of, while keeping the channels they didn't.
  • 04Never offers a settlement or payment plan outside the authority matrix your client agreed to.
  • 05Never improvises a disclosure — required language is templated, versioned, and logged against every contact.
  • 06Never keeps collecting on a disputed account, and never furnishes it without the dispute flag attached.
  • 07Never touches a card number outside PCI scope — payment capture is tokenised end to end.
  • 08Never handles hardship, attorney representation, bankruptcy, or a deceased consumer — those route to a licensed human immediately.
Compliance

Built for the rules of your industry.

FDCPA & Regulation F contact rulesTCPA consent & revocation handlingFCRA / Metro 2 dispute furnishingPCI DSS-scoped payment captureState licensing & recording consentAudit trail on every contact
Integration surface

What it plugs into.

An AI operating system only earns that name if it runs inside the systems you already have — not beside them.

Debt Collection Agencies
Collection platform of record
Dialer & telephony
PCI-scoped payment gateway
Skip trace & scrub APIs
Metro 2 / e-OSCAR furnishing
Client placement SFTP
Claude & OpenAI APIs
The objection

Asked before the contract, not after.

It is the inverse of the risk you already carry. Most FDCPA exposure comes from human variance — the disclosure skipped on a bad afternoon, the eighth attempt nobody counted, the opt-out honoured on one channel and not another. Those are deterministic rules, and deterministic rules belong in code. We put contact caps, quiet hours, opt-out state, and disclosure text in the control layer rather than the model's discretion: the model handles the conversation, the guardrail decides whether the conversation is permitted at all, and every contact is logged with the policy version that authorised it.

No, and we would argue against it. Your platform of record stays the record — Finvi, C&R Debt Manager, Latitude, InterProse, or the in-house system you have spent a decade shaping. What we build sits alongside it as an orchestration and outreach layer that reads accounts, writes activity, posts payments, and pushes dispositions back, so your reporting, your audit trail, and your client-facing numbers still come from one place.

As a hard stop, not a judgement call. A dispute, cease-and-desist, attorney notification, or bankruptcy indication suspends outreach on that account instantly, opens the workflow your compliance team defined, and sets the furnishing flag so the account is not reported without the dispute attached — the precise failure that has drawn CFPB consent orders against collectors furnishing disputed information.

State rules are configuration, not an afterthought. Licensing status by state gates which accounts the system will work at all; two-party-consent states get the recording disclosure automatically; and state-specific attempt limits or disclosure language override the federal baseline wherever they are stricter. When a state changes a rule, one config change moves every account — instead of a memo you hope forty collectors read.

The economics are better below scale, not worse. A thirty-seat agency cannot fund a QA analyst reviewing calls all day or a strategy team rebalancing channel mix weekly, so it goes without both. Automating the deterministic work is how a smaller agency gets the compliance posture and coverage of a large one without the headcount. We scope the first build to a single client portfolio, so you see it working on real accounts before it touches the rest.

Weeks, not quarters — because we do not start with the voice agent. We start with the least glamorous, highest-leverage piece, usually digital-first outreach and self-serve resolution on one segment, measure liquidation and cost per dollar collected against your current process on comparable accounts, and expand into voice only once the contact and compliance layer has proven itself.

Let's put AI to work in your business.

A 30-minute call. You bring the workflow or the roadmap — we'll tell you what's feasible, what it costs, and what we'd build first.

Book a call