28 COLLECTORS IN AN HOUR. GREAT. DID ANYBODY PAY?
Something happened in collections this week that every payday, title and installment lender should pay attention to.
PayNearMe launched an AI servicing and collections agent.
And buried inside the announcement was a number that jumped off the page at me.
In one pilot campaign, the AI agent reportedly made as many calls in one hour as roughly 28 support employees could make in the same hour.
Twenty-eight.
That’s impressive.
But I have a different question.
Did anybody pay?
Because lenders have been making this mistake forever.
We confuse activity with performance.
Applications aren’t funded loans.
Approvals aren’t profitable loans.
Collection calls aren’t collections.
Promises to pay aren’t payments.
And payments don’t help you much if they don’t actually clear, post and reconcile.
AI doesn’t change any of that.
In fact, AI may make this problem worse.
WHEN ACTIVITY BECOMES ALMOST FREE
Think about what happens when an AI agent can make thousands of collection calls.
It can send texts.
Answer inbound calls.
Authenticate borrowers.
Tell them what they owe.
Schedule payments.
Capture promises to pay.
Send payment links.
Escalate complicated accounts to a human.
And it can do this without taking lunch, calling in sick or spending fifteen minutes telling the collector at the next desk what happened Saturday night.
That changes the economics.
But it also creates a trap.
Because suddenly you can generate a hell of a lot of activity very cheaply.
10,000 calls.
6,000 texts.
2,000 conversations.
700 promises to pay.
Beautiful dashboard.
Lots of green arrows.
Everybody feels productive.
Except there is only one question I care about:
How much money came back?
STOP MEASURING THE ROBOT. MEASURE THE MONEY.
If I were evaluating one of these systems for a lender, I wouldn’t start with:
“How many calls can it make?”
I’d start with:
Accounts assigned → right-party contacts → promises to pay → promises kept → dollars collected → accounts cured → cost per collected dollar.
Then I’d compare that against humans.
Now we have something useful.
Suppose your AI agent generates twice as many promises to pay as your collectors.
Sounds great.
But what if humans have a 72% promise-kept rate and the AI has a 38% rate?
Different story.
Or maybe AI performs better.
Fantastic.
Then give it more accounts.
But prove it with dollars.
Not demos.
Not calls.
Not conversations.
Not vendor PowerPoints.
Cash.
THE QUESTION I’D ASK BEFORE BUYING ANY AI COLLECTIONS SYSTEM
Don’t ask:
“Can AI replace my collectors?”
That’s the wrong question.
Ask:
“Which parts of my collections operation should never have required a human being in the first place?”
That’s where this gets interesting.
Look at what your collectors actually do all day.
Some activities require judgment.
A distressed borrower needs help.
A complicated account needs investigation.
A repossession decision needs escalation.
A complaint needs careful handling.
A regulatory issue absolutely needs a human.
But what about:
Balance inquiries?
Due-date confirmations?
Routine reminders?
Payment links?
Basic payment scheduling?
Capturing a promise to pay?
Routine inbound servicing?
Why are we paying humans to spend hours doing things a machine may eventually do faster, cheaper and more consistently?
I’m not saying automate all of it.
I’m saying deconstruct the job before you automate the job.
HERE’S THE EXERCISE I’D RUN MONDAY MORNING
Take 30 days of collection activity.
Every call.
Every text.
Every payment arrangement.
Every promise.
Every escalation.
Every employee hour.
Then put the work into four buckets:
1. Human Judgment Required
These are situations where experience, negotiation, empathy or discretion actually matters.
2. Rules-Based
The employee is essentially following a predetermined workflow.
3. Automatable
A properly controlled system could reasonably perform the work.
4. Preventable
This is my favorite bucket.
These are collection activities you wouldn’t need at all if something upstream had worked better.
Maybe underwriting.
Maybe payment setup.
Maybe borrower communication.
Maybe first-payment reminders.
Maybe the employee made a bad loan in the first place.
That fourth bucket is where operators often find the expensive stuff nobody talks about.
NOW CONNECT COLLECTIONS TO NOI
Here’s where I think this gets really interesting.
Imagine opening your morning report and seeing this:
BROKEN PROMISE ALERT — STORE 14
63 promises due yesterday.
41 kept.
22 broken.
$11,840 in expected cash didn’t arrive.
Broken-promise rate: 2.1X company normal.
Now we have something management can act on.
Maybe the store is making lousy payment arrangements.
Maybe one collector is accepting promises that were never realistic.
Maybe customers are agreeing just to get off the phone.
Maybe the payment method is failing.
Maybe your AI agent is generating lots of promises that look terrific on a dashboard but don’t turn into money.
That’s why I want another number:
PROMISE-TO-PAY KEPT RATE
By store.
By collector.
By AI agent.
By communication channel.
By customer type.
By delinquency stage.
Now we’re managing collections instead of counting phone calls.
AND DON’T FORGET THE COMPLIANCE SIDE
There is another reason operators shouldn’t simply turn an AI agent loose on their customer base.
If the machine can communicate with borrowers and take actions, somebody needs to decide exactly what authority that machine has.
What can it say?
When can it call?
Who has consent?
How is the borrower authenticated?
Can it negotiate?
Can it change a payment date?
Can it establish a payment arrangement?
What happens when somebody says they’re experiencing hardship?
What happens when somebody disputes the debt?
What happens when the borrower says something the system doesn’t understand?
When does the machine shut up and hand the account to a human?
And most importantly:
Can you prove exactly what happened afterward?
“AI did it” isn’t going to be a very useful compliance defense.
THIS IS BIGGER THAN COLLECTIONS
Here’s what I think lenders should really be watching.
We’re moving toward a world where the entire process can eventually become connected:
Contact → Conversation → Promise → Payment → Posting → Reconciliation
Think about that.
The system contacts the borrower.
The borrower promises $175 Friday.
Friday comes.
The payment is attempted.
The payment clears.
The account updates.
The money reconciles.
Management sees the result the next morning.
That’s not an AI calling bot.
That’s the beginning of a closed-loop collections operating system.
And that’s where I think this gets serious.
THE OPERATOR TAKEAWAY
AI may soon let your company generate more collection activity than you’ve ever imagined.
Great.
Don’t confuse that with collecting more money.
The winners won’t be the lenders making the most AI calls.
They’ll be the lenders who can trace the entire chain:
Who did we contact?
What happened?
What did they promise?
Did they keep it?
How much cash came back?
What did it cost us to collect it?
And ultimately:
Did it improve NOI?
That’s the scoreboard.
Everything else is activity.
— Jer Ayles
The Business of Lending Money to Strangers
What if your next new hire could be customer-ready in 14 days?
Payday • Title • Installment Lending
Training that teaches employees what to say, what to do, and how to protect every dollar.
14DayLendingPro.com