How to Run an AI Calling Pilot for a Lender in 14 Days

A good AI calling pilot takes two weeks, not three months. The trick is to make it small on purpose. One branch. One list. One question you want answered. And one number, decided before the first call, that tells you whether it worked. For a lender, the best number is simple: live transfers that ring on that branch's phone.

Most pilots don't fail. They just never end.

You've probably seen this one before.

A vendor does a great demo. Everyone likes it. Someone says "let's pilot it." Then comes a kickoff call, a data request, a few weeks of back and forth on the script, and a meeting to talk about the meeting.

Three months later the pilot is still "going." Nobody can say if it worked. Nobody wants to be the one to kill it. And nobody wants to sign up for more of it either.

That's not a technology problem. It's a setup problem.

A pilot is supposed to answer a question. If you never wrote the question down, there's nothing to answer. So it drifts.

And drifting has a cost, even if nobody puts it on a spreadsheet. Your ops team keeps spending hours on it. Your branch managers stop paying attention. Your past customers keep sitting on a list nobody's calling. And the next idea, maybe the one that would actually work, waits in line behind this one.

Start with one branch, one list, one question

Here's what a pilot that proves something looks like before it starts.

One branch. Not five. Not "the Southeast region." One branch, with a manager who knows transfers are coming and who picks them up. You're probably thinking one branch is too small to tell you anything. It's actually the opposite. With one branch, you can see every transfer. You know who answered. You know if the phone rang out. With ten branches on day one, a problem at three of them hides inside the average.

One list. Pick the list where the answer matters most and the conversation is simplest. For most lenders, that's past customers who paid off a loan. They already know you. The offer is easy to explain. We wrote about how to reactivate paid-off borrowers if you want the longer version.

One question. Write it in one sentence. Something like: "If we call every reachable past customer of this branch in two weeks, how many live transfers does the branch get?" That's it. Not "can AI help our business." Not "let's see what it can do." A question with a number for an answer.

One success number, set before the first call. This is the step most teams skip. Decide ahead of time what "worth it" looks like. Live transfers is a good measure because it's hard to fake. A transfer either rang on the branch phone and a person picked up, or it didn't. If you decide what counts after you see the results, you'll talk yourself into whatever story fits. Everyone does. The Academy module on measuring goes deeper on setting that baseline before you switch anything on.

The four ways AI calling pilots go wrong

In our experience, when a pilot goes sideways, the AI is rarely the reason. It's almost always one of these.

1. The AI is aimed at the wrong job. Someone gets excited and points the AI at the hardest conversation in the business. A tough collections call. A complicated objection. A customer who's upset. That's backwards. The AI is best at the repetitive job nobody has time for: making thousands of short calls, asking one clear question, and passing the interested people to a human. Let it do the dialing. Keep your people for the conversations that need a person. We explain this in more detail in Start here.

2. Nobody owns what happens after the call. The AI finds an interested customer and transfers them. Then what? If the branch doesn't know transfers are coming, the phone rings out. If nobody follows up on the ones who said "call me tomorrow," they go cold. The call was only the first half. Someone has to own the second half, by name.

3. The results never reach the CRM. The pilot runs. Calls happen. And the results live in a vendor dashboard that one person logs into once a week. Meanwhile the branch team can't see who was called, the CRM still shows the customer as untouched, and a regional manager has no idea if the pilot is doing anything. If the outcomes don't land in the CRM the same day, the pilot is invisible to the people who have to act on it.

4. There's no end date. Without a date, a pilot turns into a subscription nobody decided to buy. Put the end date on the calendar before you start. On that day, you look at the number and make a call: stop, fix, or grow.

What a 14-day pilot looks like

Here's the structure. It's not complicated. It's just strict.

Week 1: the list and the setup. First, pull the branch's list and clean it. Take out disconnected numbers, duplicates, and anyone who shouldn't be called. Make sure every customer is tied to the right branch. Then set up the calling around how the branch actually works. A local number and caller ID for that branch. Calling only during branch hours, so there's someone to take the transfer. A plan for calling no-answers again. Finally, walk the branch manager through what's coming. Who picks up a transfer. What the customer will already have heard. Before anyone dials, have your own compliance team review the list, the consent and the script. This isn't legal advice. It's just the step you don't skip.

Week 2: calling. The AI calls every reachable customer on the list. Not a sample. The interested ones get transferred live to the branch while it's open. Every call is logged, and the outcomes land in the CRM the same day. Watch it daily, but don't redesign it daily. Fix things that are broken, like a transfer line that isn't picking up. Leave the rest alone until the numbers are in.

Day 14: the numbers, by branch. How many calls were placed. How many people answered. How many were interested. How many live transfers rang on the branch's phone. Then compare it to the number you wrote down before you started. That comparison is the whole point of the pilot.

What Sunset Finance learned from starting small

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.

They didn't turn it on everywhere. They started with a pilot.

In the first 14 days, the voice agent, Tommy, placed 13,159 calls. That produced about 100 live transfers across 10 branches, each one sent to the customer's own local branch while it was open. That works out to about 8 live transfers for every 1,000 calls.

The pilot gave them a real number to decide on. Since then they've had 224 live transfers and counting, and they've grown from 10 branches in the pilot to 28 branches now.

The point isn't that every lender will see the same numbers. They won't. The point is that two weeks was enough to know whether to keep going. You can read the full story in the Sunset Finance case study.

A small way to start

This is also how we run our own offer, and it's why we built it this way.

We start with one branch. Within 14 days, every reachable customer on that list gets called, and the interested ones get transferred live to that branch. There's no upfront fee. You only pay for results: a booked meeting or a reactivated customer.

One branch. Two weeks. A number you can check yourself.

If you want to hear what your borrowers would hear, the AI will call you. Leave your number.