Skip to content
Features

In your operation

Pilot with a control group

Before you decide, compare. One part of the portfolio runs with the agent, another stays on the current process, and the success criterion is written down beforehand.

On one call

  1. 1Sets aside a segment of the portfolio
  2. 2Runs the agent and the current process
  3. 3Compares at the end of the period

How do you test AI voice collections before signing up for the whole portfolio?

With a 30 to 60 day pilot. At Talkover, one segment of the portfolio runs with the voice agent and a control group, in the same overdue range, stays on your current process. The indicators are amount recovered, agreements kept, cost per agreement, contact rate, and complaints. The success criterion is written down before you start.

The essentials

A fair comparison

Agent and control work the same overdue range, in the same period.

Criterion written down first

Today's number becomes the baseline. What counts as success is agreed at the start.

A decision with a date

The decision meeting is already scheduled for the end of the pilot.

In practice

  1. 01Choosing the segmentA segment that is poorly served today: overflow volume, low balance, or one overdue range.
  2. 02DuringAdjustments along the way and one meeting a week, with the return and the cost per agreement.
  3. 03At the endThe funnel, the return, and what the portfolio showed, side by side with the control group.

Why is the pilot paid?

Because a free trial tends to end up with no owner on the client's side and does not reach a decision. A pilot with a cost, a deadline, and a criterion has someone responsible for approving the agent and following the result.

The model is the same as in the operation: a one-time setup and consumption per minute of talk time.

What are the stages?

Five, from diagnosis to result.

  • Diagnosis: a sample of the portfolio and of the current collection cadence.
  • Definition: negotiation rules and integration.
  • Go-live: the first calls.
  • Follow-up: daily adjustments and one meeting a week.
  • Result: the funnel, the return, and what the portfolio taught.

Can you test two approaches at the same time?

You can, by splitting the queue: one part gets the new wording and the other keeps the current one. Today this split is set up by the team with you; it is not an automatic feature of the platform.

The comparison needs volume. With a few dozen confirmed account holders a day, one day is not enough to say which wording closes more agreements.

The cases published so far had no control group: the store credit one compares before and after in the same portfolio. That is why we recommend the control.

Real results · Old debt

4 agreements

on debts 1.4 to 4.7 years old

A two-day proof of concept. R$32.6k negotiated.

See the case

Frequently asked questions

How long until the pilot starts?

The first version of the agent is ready in about 7 business days after the kickoff meeting. The start of the calls depends on your approval and, if there are any, on the integrations.

Can I run a shorter test?

Yes. We published a two-day proof of concept on old debt. It shows whether the agent works on your portfolio, but it does not replace the pilot with a control group.

What do I need to provide for the pilot?

The negotiation policy, 3 to 5 mandatory topics for the conversation, the list of debtors with the do-not-call list, the way to generate a payment slip or Pix, and a person with the authority to approve the agent.

For what size of operation does the pilot make sense?

For volume operations: tens of thousands of debtors in the list, or more than a thousand calls a day. Below that, the gain tends not to pay for the setup.

Updated on

See also

It is time to hire your AI call center

Scale your operation without scaling your team.