← All implementations
Collections · Uruguay — Implementation case

AI voice collections, owned by the client.

A complete outbound collections voice stack for a fintech in Uruguay — built under their brand, transferred to their name. It replaced a 40-person dialing operation in six weeks. By week two, the agent was already outperforming the team it replaced.

01
The starting point

Forty people, dialing all day.

The gap

Recovery ran on a floor of forty collectors working a dialer by hand. Most of the day went to dead ends — voicemails, no-answers, wrong numbers — and the few real conversations followed whatever script each agent remembered that morning. Output rose and fell with mood, tenure, and turnover.

For a fintech, recovery is margin. Every account that slips past its window costs more to collect and is less likely to pay at all. The operation was expensive, hard to scale, and impossible to keep consistent across delinquency buckets — and there was no clean record of what actually worked on a call.

02
What we built

A full voice stack, under their brand.

Not a chatbot bolted onto a phone line. The complete outbound collections engine — scripts, filtering, scoring, and timing — built to run their book end to end.

Adaptive scripts

A different conversation per bucket

Scripts that shift by delinquency stage — early reminders, firm negotiation, payment-plan offers — and adapt live to what the debtor says, instead of one rigid flow for every account.

AMD filtering

Talk to people, not voicemail

Answering-machine detection screens out voicemails, dead lines, and dial tones before the agent says a word — so effort lands on real, live conversations.

QA scoring

Every call scored, not sampled

Each call is transcribed and scored automatically — tone, compliance, outcome — so quality is measured on the whole book, not a handful of spot-checks.

Retry & windows

The right call at the right hour

Retry logic and calling windows decide who to call back, when, and how often — respecting contact rules and timing each attempt for the best chance of an answer.

03
How it ran

Built in, then handed over.

01

Embed

We sat inside the recovery operation — listened to real calls, mapped each delinquency bucket, and learned what actually moved an account toward paying.

02

Build

We built the stack under their brand — agents, adaptive scripts, AMD, QA scoring, retry logic, telephony — and tuned it against live calls before scaling.

03

Scale

By week two the agent was outperforming the human team. Over six weeks it took over the full book and replaced the forty-person dialing operation.

04

Transfer

Numbers, agents, data, and models went into their name. They pay carriers directly — no per-minute fees to us or anyone else.

04
Outcomes

A floor of forty, replaced in six weeks.

6 weeks
to full replacement
contact rate vs human teams
40 → AI
collectors replaced by the agent

The agent contacted three times more debtors than the human teams it replaced — AMD filtering and timed calling windows meant nearly every minute went to a live conversation instead of a voicemail. Consistency stopped depending on who was on shift. Every call was scored, so quality became something the team could see and steer.

When it was done, the whole stack moved into the client's name — numbers, agents, data, models. This is the engagement model behind our Voice AI Infrastructure service: we build it under your brand, run it until it works, then hand you the keys. You pay carriers directly. No per-minute fees to us, or to anyone.

"We replaced 40 human collectors in six weeks. By week two the agent was already outperforming them."

Collections Manager · Fintech · Uruguay
Client name withheld by agreement. Happy to walk through the details on a call.

Own your collections stack, don't rent it.

A 30-minute call. We'll map your buckets, your call volume, and what a voice stack you actually own would look like on your book — the engagement model behind our Voice AI Infrastructure service.