For years, the conversation around AI in collections has mostly revolved around one question:

Will AI replace collectors?

It’s an understandable question. 

AI voice agents can make calls, answer questions, send texts, discuss accounts, and carry on conversations that would have required a human not that long ago. Some are already operating at enormous scale.

But whether AI eliminates the collector may not be the most interesting question anymore.

A better one is: What happens to the collector’s job when AI starts taking the easier work?

Because if automation delivers on even part of its promise, it probably won’t eliminate every collection conversation evenly. The predictable interactions are the obvious place to start.

A straightforward payment reminder.

A balance question. 

A consumer ready to make a payment.

A basic account update. 

A routine follow-up.

Those are exactly the kinds of conversations technology is increasingly capable of handling.

Which means the humans don’t necessarily disappear.

They get what’s left.

And what’s left may require an entirely different level of skill.

The Human Queue Is Becoming the Hard Queue

Imagine a collection operation several years from now, where automation works exactly as intended.

A consumer answers a call and wants to know the balance? AI handles it.

Someone says they’ll make their payment Friday? AI handles it.

A consumer wants to confirm basic account information or work through a standard payment arrangement? Depending on the organization, AI may handle much of that too.

The human collector becomes necessary when something doesn’t fit neatly into the process.

The balance is disputed. The consumer lost their job. There’s confusion about previous payments. A family member died. Someone mentions bankruptcy. The account history is a mess. Or the conversation has reached the point where judgment matters more than following the next step in a workflow.

This is already visible in some of today’s AI collection models. 

WIRED recently reported on AI debt collection companies using different thresholds for handing conversations to people, including situations involving illness, death in the family, or bankruptcy.

There is nothing particularly surprising about that model. Automation has always been good at repeatability. Humans have always been good at exceptions.

The interesting part is what happens when you apply that principle at scale.

If technology successfully removes a large portion of predictable work, the average human interaction becomes more complex.

And “harder” is worth defining here.

It doesn’t simply mean the consumer is angrier. 

A difficult call might involve emotional tension, but it could just as easily involve conflicting information, unusual circumstances, greater compliance risk, a complicated account history, a lower likelihood of immediate resolution, or a situation where the collector needs to make a judgment call that isn’t covered neatly by a script.

That’s the bigger shift.

AI may not simply reduce the number of conversations humans handle. It may change the type of conversations humans handle.

Consider a team where a collector historically managed 50 relatively straightforward conversations and 10 difficult ones during a given period. 

If automation starts resolving most of those routine interactions, that same person’s workload could eventually be dominated by the 10 conversations that were previously the exceptions.

Call volume goes down.

The job doesn’t necessarily get easier.

Fewer Calls Doesn’t Always Mean Less Work

This is where the automation conversation can become misleading.

Most discussions eventually arrive at productivity math.

How many calls can the technology make? 

How many accounts can it work? 

How much does it cost compared with an employee? 

How much headcount could theoretically be reduced?

Those numbers matter. 

The Bureau of Labor Statistics already projects employment of bill and account collectors to decline 10% between 2024 and 2034, noting that enhanced software and automated calling systems are expected to allow more collection work to be completed with fewer employees.

But fewer employees only tells part of the story.

Organizations also need to look at the work being redistributed to those employees.

Suppose an agency automates 40% of its consumer conversations. On paper, that looks like 40% less work.

Except that the technology may have removed 40% of the volume without removing anywhere close to 40% of the complexity.

One five-minute call where a consumer confirms a payment date and one five-minute conversation involving a disputed balance appear identical on a basic call-volume report.

Operationally, they aren’t even close.

The first may require following a relatively predictable process.

The second may require listening, investigating, explaining, negotiating, recognizing compliance concerns, and deciding what should happen next, all while talking to someone who probably did not wake up excited to hear from a collection agency.

This creates a strange possibility for the future of collections:

The average collector could handle fewer calls while doing more demanding work.

And if that happens, the definition of a top performer has to change with it.

The Best Collector of the Future May Look Different

Collections organizations have historically measured a lot of activity.

Calls per hour. Right-party contacts. Promise-to-pay rates. Dollars collected. Average handle time. Talk time. Schedule adherence.

Those metrics aren’t suddenly useless.

But a workforce responsible for increasingly complicated interactions needs a broader set of skills.

The collector who can fly through routine conversations may become less valuable than the collector who can take a messy situation and figure out what’s actually happening.

Can they recognize when frustration is becoming an escalation?

Can they ask a useful follow-up question instead of immediately jumping back to the script?

Can they explain something complicated without making the consumer even more confused?

Can they negotiate?

P.S. We think that negotiation is one of the most underrated tools in Collections.

Can they recognize vulnerability?

Can they stay calm when the person on the other end of the phone absolutely does not?

Can they understand the difference between someone refusing to pay and someone who genuinely doesn’t understand what options are available?

Those abilities are harder to quantify than calls per hour. They’re also much harder to teach through a PowerPoint presentation and a laminated script.

That becomes especially important when you consider where those skills traditionally come from.

We May Be Automating Away the Training Ground

Great collectors usually aren’t great on day one.

They get good by having conversations.

Lots of them.

Some are straightforward. Some are uncomfortable. Some go badly. A manager steps in. The collector hears a better way to phrase something. They try it on the next call.

Over hundreds or thousands of interactions, they start recognizing patterns.

They hear the difference between genuine confusion and resistance.

They learn when another question will help and when another question will just irritate someone.

They learn how quickly a calm conversation can turn into an escalation.

They learn that two consumers can say the same words and mean two completely different things.

The routine calls aren’t wasted repetitions. They’re part of how someone builds enough confidence and pattern recognition to eventually handle the difficult ones.

Now imagine that many of those interactions are automated.

A new collector joins the organization, completes training, and starts taking live conversations. Except the easiest conversations are already being handled somewhere else.

Their queue is disproportionately filled with disputes, hardship, complicated account histories, angry consumers, and situations the technology couldn’t confidently resolve.

We’ve essentially let AI complete the beginner levels of the game and then handed the new employee the controller at Level 12.

Good luck.

That creates a problem that has received far less attention than the headcount conversation:

How do you create experienced collectors when you’ve automated many of the experiences that used to make them experienced?

The answer can’t simply be more classroom training.

Organizations may need to intentionally replace the repetitions that automation removes.

That could mean more realistic simulations before employees ever work live accounts. It could mean structured call libraries that expose new collectors to dozens of variations of the same difficult scenario. 

It could mean more side-by-side coaching, more frequent review early in someone’s tenure, and better real-time support when an unfamiliar situation comes up.

Training may also need to shift away from memorizing the correct response and toward understanding how to make a good decision.

Instead of asking only, “What should you say when a consumer gives this objection?” training may need to ask:

What information do you need before you respond?

What are the possible risks?

What signals should change your approach?

When should you continue the conversation?

When should you stop and escalate?

Why did you choose that response?

That’s a much different training model.

And it’s probably closer to the skill set organizations will need if humans increasingly become the exception handlers.

Old Performance Metrics Could Start Lying to Us

The shift also creates a measurement problem.

Imagine two collectors.

Collector A handles 30 relatively straightforward accounts. Most consumers are willing to engage, the conversations are short, and several result in payments.

Collector B handles eight escalated accounts that were transferred from an automated system. Two involve disputes. One consumer is extremely upset. Another has a complicated hardship situation. Three eventually reach workable arrangements, and Collector B prevents two conversations from turning into formal complaints.

Now put both employees on the same traditional scorecard.

Collector A probably wins.

More calls.

Shorter handle time.

Higher payment volume.

Possibly better conversion.

Yet Collector B may have done significantly more difficult and potentially more valuable work.

That’s why organizations introducing more automation will eventually need to think about interaction complexity rather than simply interaction volume.

Performance measurement may need to account for whether a collector correctly understood the situation, handled the consumer appropriately, stayed within compliance requirements, asked the right questions, and identified the correct next step.

A lower conversion rate doesn’t automatically indicate poor performance if one employee is receiving a disproportionately difficult book of work.

A longer call doesn’t necessarily indicate inefficiency if that employee is resolving issues that couldn’t be handled through an automated workflow.

Otherwise, organizations risk creating an odd situation where the technology routes the hardest work to people, and the performance system penalizes those same people for taking longer to solve it.

The Human Job Doesn’t Necessarily Get Smaller. It Gets More Specialized.

None of this is an argument against automation.

There are plenty of collection activities where using a person simply because “that’s how we’ve always done it” makes little sense.

Nobody needs a highly trained collector spending hours leaving essentially identical messages or answering the same basic balance question for the hundredth time.

Automation can remove repetitive work, increase coverage, and give consumers additional ways to resolve accounts.

But organizations should be careful about assuming the remaining human role is simply today’s collector handling fewer calls.

It may become something different.

The future collector could increasingly operate as an exception handler, negotiator, investigator, problem solver, and de-escalation specialist.

That has implications for who you hire, how you train them, how quickly you expect them to become productive, and how you determine whether they’re actually good at the job.

It may even mean that entry-level collections work becomes less entry-level.

That’s the part of the AI conversation worth paying attention to.

Debt collection may employ fewer people in the future. AI voice agents are already making calls at a scale that would have sounded absurd a few years ago, and automation will continue getting better at conversations that currently require people.

But jobs rarely disappear as neatly as a spreadsheet suggests.

Work gets redistributed.

The repeatable tasks go somewhere else.

The exceptions remain.

And the people left handling those exceptions may need better judgment, stronger communication skills, and substantially more training than the people doing the job today.

The question for collection leaders isn’t simply how much work AI can remove.

It’s whether they’re preparing their people for the work AI leaves behind.

Because buying automation may turn out to be the easy part.

Building a workforce capable of handling everything the automation cannot may be the harder transformation.