How PTS AI Works Inside a Branch Network, Step by Step

Here's how it works, start to finish.

PTS AI cleans and phone-verifies your customer list. Each branch gets its own local number and caller ID. Calls go out only while that branch is open. The voice agent explains the offer and asks the same qualifying question on every call. Interested borrowers are transferred live to their own branch. The ones who didn't answer get called again. And every outcome is written to your CRM the same day, with a board by branch.

That's the whole machine. The rest of this article walks through each step, what it's for, and what we've seen when it runs.

Tuesday, 10:15 a.m., at one branch

Start with what your borrower experiences, because that's what matters.

It's a Tuesday morning. A woman who paid off a small loan with your branch last year is at her kitchen table. Her phone rings. The caller ID shows a local number and your branch's name, not an 800 number from another state.

She picks up. The agent tells her who's calling and why: she has a pre-approved offer waiting at her branch. She asks a couple of questions. The agent answers them. Then it asks the qualifying question, the same way it asks everyone.

She says yes, she'd like to talk about it.

The agent doesn't say "someone will call you back." It transfers her, right then, to her own branch. A loan officer who was helping someone at the counter five minutes ago picks up a call with an interested customer already on the line.

Nobody at the branch dialed. Nobody sat through 40 voicemails to find her.

That's the experience we build for. Now here's what has to be true behind it.

First: the list gets cleaned before anyone is called

Most calling projects skip this step. It's the most boring one, and it decides a lot.

Your customer list has been through years of branch software, exports and manual edits. Some numbers are dead. Some are formatted wrong. Some customers appear twice.

So before any branch starts calling, every phone number is verified before it's dialed. The goal is simple: no dead numbers. Our Technical Lab lists a bad list as the first of five places these systems break, for exactly this reason. If the list is wrong, every step after it is wrong too.

Then: every branch gets its own number and its own hours

This is where a lot of AI calling setups go wrong. They call everyone from one number, at whatever time the software runs.

We do the opposite.

A local number per branch. Each branch gets a local number with its own caller ID. We handle the carrier registration for every branch. Your borrower sees a number that belongs to the place they already know.

Only while the branch is open. Calls go out only during that branch's hours, while someone can pick up. This one rule makes the whole thing work. A live transfer is worthless if it rings into an empty office. We explain why a callback isn't a good substitute in live transfer or callback.

A quick, plain note on rules. An AI voice is an artificial voice under federal law: in a 2024 declaratory ruling, the FCC said callers need the called party's prior express consent before placing calls that use an AI-generated voice. Consent and state rules are your compliance team's decision, not ours, and this isn't legal advice. We walk them through exactly how the system works. Our compliance checklist for lenders is a good place to start that conversation.

Then: the agent has the same conversation every time

Your best loan officer on their best day is great on the phone. The same person at 4:45 on a Friday, after a hard collections call, is different. That's human.

The voice agent doesn't have Fridays.

It explains the offer. It answers questions. And it asks the same qualifying question, the same way, on every call. That consistency is the point. It means "interested" means the same thing at your busiest branch and your quietest one, so you can compare them.

It also means every call ends in one of a small set of outcomes, applied the same way. Our Academy puts it bluntly: six outcomes everyone applies the same way beats thirty applied inconsistently. Those outcomes become the columns of every report you'll look at later.

Then: interested borrowers go live to their own branch

When a borrower says yes, the agent transfers them live to their own local branch, while they're still holding the phone. Not a shared queue. Not a central call center. The branch they'd actually walk into.

This is the moment everything else exists to protect. Interest fades fast. A "yes" at 10:15 is a "maybe" by the afternoon and a "who was that?" by Friday.

Your people don't change what they do. They have the loan conversation. The AI only does the dialing, and the branch only picks up when there's already an interested customer on the line.

Finally: no-answers get called again, and everything gets written down

Most people don't answer the first call. That's normal.

A number that didn't pick up isn't a number that's done. No-answers get called again, during branch hours, from the same local number.

And every outcome is written to the CRM you already run, the same day. You don't have to switch systems. Called, reached, transferred: it's there by the end of the day, not whenever someone at the branch gets a free minute.

Behind that is a simple structure, laid out in our Technical Lab: voice agent, then orchestration, then CRM, then reporting. The agent owns the call. Orchestration owns routing and the if-this-then-that logic. The CRM owns the state of the customer relationship. Reporting owns the chain from the first call to the money.

One ID holds it together. It's created at the first touch, passed into the call, returned when the call ends, written to the CRM, and carried to the funded loan. That's what lets you prove which calls turned into loans instead of guessing. We cover that part in what happens after the AI call.

On top of it all sits a board by branch. Each branch on its own row, so a strong network total can't hide one branch that's quiet.

What it looks like at Sunset Finance

Sunset Finance is a network of consumer lending branches in Georgia. They had 43,000 past customers who had already borrowed once, sitting unused. The offer was a pre-approved loan of up to $1,000.

Their voice agent is called Tommy. In the first 14 days of the pilot, Tommy placed 13,159 calls. About 100 of those became live transfers in those two weeks, across 10 branches. Each one went to the customer's own local branch while it was open.

That's about 8 live transfers for every 1,000 calls placed. Since then it's 224 live transfers and counting, and the program has grown from 10 branches to 28.

Every one of those steps is the same machine described above. Nothing custom per branch except the number, the hours and the list.

How a network starts, and what waiting costs

You don't start with the whole network. You start with one branch.

Week one is setup: the list is cleaned, and the local number, caller ID and agent are ready. In week two the branch starts calling and the first live transfers come in. On day 14 you have the numbers by branch: called, reached, transferred. We walk through those first days in detail in what week one of an install looks like.

Meanwhile, your list keeps aging. The FTC explains that creditors can get lists of consumers from credit bureaus who meet their requirements and send them offers. Your past borrowers can end up on another lender's list. The question is whether your branch reaches them first.

If you want to hear exactly what your borrowers would hear, leave your number. The same agent calls you within minutes.