What Happens When The Easy Calls Disappear And Algorithms Never Blink

We test whether AI has finally broken the fast-cheap-good tradeoff in customer service, showing where massive gains come from and where the pressure lands. We lay out a path to keep the speed while protecting the people who handle the hardest calls.

• AHT benchmarks for retail, finance, and healthcare
• ACW as hidden drag and how automation removes it
• Agentic AI for self-service and ticket deflection
• Smart routing and big jumps in first contact resolution
• Predictive support and healthcare urgency
• The human pivot to complex, high‑emotion cases
• The vigilance tax from real‑time AI co‑pilots
• 100% call scoring, stress, and attrition risks
• Design moves for 2026: transparency and autonomy
• Shifting metrics beyond AHT to stability and ROI

Thanks for taking this deep dive with us

Nonprofits and Businesses plan to automate at least 30% of all processes in 2026.  What is your plan?

Transcript

The Fast-Cheap-Good Paradox Challenged

SPEAKER_01 0:00

Welcome back to the deep dive. Today, uh today we are wrestling with a concept that usually sets off alarm bells in the business world.

SPEAKER_00 0:06

It definitely does.

SPEAKER_01 0:07

Because, you know, the old saying, right? If you want something done, you basically have three options fast, cheap, or good.

SPEAKER_00 0:14

Right. The classic project management triangle. And the golden rule is you only ever get to pick two.

Benchmarking AHT Across Industries

SPEAKER_01 0:19

Exactly. If you want it fast and cheap, it won't be good. If you want it good and fast, it's not going to be cheap. But today, we are looking at a stack of research. We've got market analysis on something called AICAHT, reports from FreshWorks, Gartner, Sprinkler, and they're all claiming this iron law is currently being broken.

SPEAKER_00 0:37

Aaron Powell Completely broken.

SPEAKER_01 0:38

Aaron Powell Yeah, we're looking at a massive shift in the customer service world where companies are seeing uh reductions in handle times of 30 to 50 percent hitting in 2025.

SPEAKER_00 0:47

Aaron Powell It is a staggering number. I mean, in this industry, a 5% efficiency gain is usually cause for a champagne toast, so 50% reduction. That is a total paradigm shift.

SPEAKER_01 0:57

Aaron Powell And the claim is they aren't just slashing quality to get there. They aren't just, you know, hanging up on people faster.

SPEAKER_00 1:04

Aaron Powell No, quite the opposite, actually. But there is a catch. There's always a catch.

SPEAKER_01 1:08

Always a catch. So the mission for this deep dive is to figure out how that is physically possible. How do you shave half the time off a phone call without destroying the customer experience? Right. But also, and I think this is where the research gets really crippling enough, we need to look at the human cost.

SPEAKER_00 1:25

Right. The human element is crucial here.

SPEAKER_01 1:27

Because there is a term in your notes that jumped out at me immediately: the vigilance tax.

SPEAKER_00 1:32

Yeah, it's a critical concept. Buried in all these shiny efficiency reports is a real warning that we might be creating a pressure cooker for the human agents who are left in the loop.

SPEAKER_01 1:41

So we have this paradox of speed, efficiency versus burnout. But before we get into the dark side of this, let's establish the baseline. Because to understand a 50% drop, you and I need to know what normal looks like right now.

SPEAKER_00 1:54

Makes sense.

SPEAKER_01 1:55

Like if I pick up the phone to call my bank or my doctor, how much of my life am I about to lose?

SPEAKER_00 2:00

Well, it varies wildly by industry, but the sources give us some really great benchmarks for 2025. If you are calling a retailer, say you're asking about a return, or just where are my shoes?

SPEAKER_01 2:12

Important question.

SPEAKER_00 2:13

Exactly. That's usually the fastest sector. The average handle time, or AHT, is about three minutes and twenty nine seconds.

SPEAKER_01 2:21

Okay. That feels manageable. Three and a half minutes. I could live with that.

SPEAKER_00 2:25

Sure. But if you move to financial services, you're looking at around four minutes and five seconds, and then you have the heavyweight, which is healthcare. The average interaction there is sitting at about 6.6 minutes.

SPEAKER_01 2:37

Wow, almost seven minutes. Which I guess makes sense. Asking about a sweater size is very different from discussing a medical bill or a new diagnosis.

SPEAKER_00 2:45

Aaron Powell Precisely. Complexity drives time. But here is the nuance that most people miss, and it is crucial for understanding where these massive efficiency gains are actually coming from.

ACW: The Hidden Time Sink

SPEAKER_01 2:55

Aaron Powell Okay, what are we missing?

SPEAKER_00 2:56

When we say average handle time, we aren't just measuring the conversation you're having. AHT is a composite metric.

SPEAKER_01 3:02

Aaron Powell Break that down for us. What is actually in the formula?

SPEAKER_00 3:06

It's talk time plus hold time, which we all know and loathe.

SPEAKER_01 3:09

Oh yeah.

SPEAKER_00 3:10

But it also includes something called ACW, which stands for after call work.

SPEAKER_01 3:14

After call work. So this is the stuff that happens after I hang up the phone.

SPEAKER_00 3:17

Aaron Powell Exactly. You hang up, you go about your day, but the agent, they are still there. They are categorizing the call, typing up notes, updating the CRM, scheduling the follow-up, tagging the sentiment. And this is the killer stat from the research. After call work often eats up 20 to 30% of the total handle time.

SPEAKER_01 3:37

Wait, so on a six-minute call, nearly two whole minutes might be the agent just staring at a screen, clicking boxes after the customer is gone.

SPEAKER_00 3:45

Yes. And from a business perspective, that is dead time. The customer isn't being helped, and the agent isn't taking the next call. This is the low-hanging fruit. This is where AI is striking first. That makes total sense. We have data here from FreshWorks showing that automated ACW tool, so AI that basically listens to the call and does all the paperwork for you, can cut documentation time by 50%.

SPEAKER_01 4:06

So the agent hangs up. Instead of typing for two minutes, the AI just it just does it.

SPEAKER_00 4:11

It summarizes the call, it updates the file, it even tags the sentiment. Like, was the customer angry? Were they happy? Freshworks found that workflow automation alone drives a 26.63% reduction in resolution time.

SPEAKER_01 4:25

Aaron Powell That is huge. And honestly, that feels like a good kind of efficiency. You're removing the drudgery, the agent isn't burnt out from typing, they're just moving on to help the next human.

From Chatbots To Agentic AI

SPEAKER_00 4:35

In theory, yes. It completely removes the clerical burden. But this brings us to the second layer of the tech stack. It's not just about cleaning up after the call, it's about fixing the problem during the call or even before it happens.

SPEAKER_01 4:48

Right, because saving time on notes is great. But if the agent still doesn't know the answer to my question, I am still on hold.

SPEAKER_00 4:55

Exactly. And that is where large language models or LLMs are changing the game. We're seeing a massive shift toward what the industry calls agentic AI.

SPEAKER_01 5:03

Agentic AI. That sounds a little sci-fi.

SPEAKER_00 5:06

It is a massive leap from the chatbots we've all been annoyed by for the last 10 years. You know the ones. I didn't quite catch that. Please say it again. Oh, the worst. I just end up screaming, press zero for operator.

SPEAKER_01 5:17

Right, because those are just decision trees. If X, say Y, agentic AI can actually reason, understand context, and take action. The sources mention that generative AI-powered self-service is now driving ticket deflection range of 53%.

SPEAKER_00 5:32

53%. That means more than half the people calling don't even need to talk to a human anymore.

SPEAKER_01 5:37

Exactly. And not because they gave up in frustration, but because the bot actually solved the issue. It understood the problem and fixed it. But for the 47% who do need a human, the experience is shifting dramatically because of something called smart routing.

SPEAKER_00 5:52

Smart routing. Is that just a fancy name for press one for sales, press two for support?

Smart Routing And FCR Gains

SPEAKER_01 5:56

No, that's the analog version. Smart routing in 2025 is analyzing your tone, your intent, and your history in milliseconds before the phone even rings on the agent's side.

SPEAKER_00 6:07

So it knows I'm angry before the agent even picks up the phone. It knows you're angry. It knows you tried to reset your password three times yesterday and failed. It knows you visited the cancel subscription page this morning and uses all that data to route you specifically to an agent who is rated highly for retention and empathy.

SPEAKER_01 6:25

That is impressive, but also a little intense. It's profiling me in real time.

SPEAKER_00 6:30

It really is. But look at the results. We need to talk about FCR, which is first contact resolution. This is the holy grail metric in this world. The benchmark is that a 1% increase in FCR equals a 1% boost in customer satisfaction.

SPEAKER_01 6:46

Which makes sense. If I have to call back twice for the same problem, I'm not happy.

SPEAKER_00 6:49

Exactly. Financial services traditionally sit at about 65% FCR. That means 35% of people more than one in three have to call back.

SPEAKER_01 6:57

It's a lot of wasted time.

SPEAKER_00 6:58

It is. But the sources show that these AI implementations are pushing that FCR up to 85%.

SPEAKER_01 7:04

That is a massive jump in quality.

Predictive Needs And Healthcare Urgency

SPEAKER_00 7:06

It is. And there is a case study from JP Morgan mentioned in the source material that really highlights this. Now, it's not a context center example strictly, but it illustrates the raw power of this tech. They used AI to interpret business credit agreements.

SPEAKER_01 7:22

Which is incredibly dense legal stuff.

SPEAKER_00 7:25

Oh, extremely dense. It used to take them 360,000 hours of human time annually.

SPEAKER_01 7:31

360,000 hours? That's dozens of lifetimes.

SPEAKER_00 7:35

The AI did it in seconds.

SPEAKER_01 7:37

Seconds. Okay, I can't even process the math on that efficiency game. That is mind-boggling.

SPEAKER_00 7:42

It shows the sheer processing power we are dealing with. But for the contact center, the most sci-fi element is what they call predictive customer needs.

SPEAKER_01 7:50

This is the part of the research I found genuinely spooky.

SPEAKER_00 7:53

It is a bit spooky. Amazon's AI can apparently predict 35% of queries before they are even asked.

SPEAKER_01 8:00

Okay, wait. So I log into the chat, and before I even type where is my package, it says, Are you looking for the toaster you ordered?

SPEAKER_00 8:06

Exactly that. It looks at the probability. The user ordered a toaster, the delivery window is today. The user is on the help page. It is highly likely they want the toaster. It answers the question before you formulate it.

SPEAKER_01 8:19

And in healthcare, this actually sounds vital, not just convenient.

SPEAKER_00 8:22

It is absolutely vital. We know from the data that 30% of patients abandon a call if they wait more than a minute. In healthcare, an abandoned call isn't just a lost sale. It could be a serious medical risk. Right. So if the AI can pull the electronic health record instantly so the agent knows your history, your meds, and your last appointment the very second they say hello, you aren't wasting three minutes on what's your date of birth. You are immediately solving the problem.

The Human Pivot: Only Hard Calls

SPEAKER_01 8:49

Okay, let's pause here. We have automated notes. We have smart roading, we have psychic AI predicting our problems. This sounds like a utopia for efficiency.

SPEAKER_00 8:58

It really does. And the shareholders are thrilled. The numbers remember that market projection from 4 billion to over 124 billion, it's all based on this massive efficiency.

SPEAKER_01 9:07

But I feel a but coming.

SPEAKER_00 9:09

There is always a but.

SPEAKER_01 9:10

If the AI is handling 53% of the routine stuff the password resets, the simple transactions, what is actually left for the humans to do?

SPEAKER_00 9:20

That is the pivot point. The role of the human agent is fundamentally changing. The sources are very clear on this. Humans are no longer there to be information switchboards. They are exclusively handling high complexity, high-emotion interactions.

SPEAKER_01 9:33

So the easy calls are just gone.

SPEAKER_00 9:35

Completely gone. You never get a breather call anymore. In the old days, you might have a screaming customer and then three easy address changes where you could just mentally coast for a minute.

SPEAKER_01 9:45

Just to catch your breath.

SPEAKER_00 9:46

Right. But now, Gartner found that 91% of leaders are under pressure to implement AI, and 80% plan to transition agents into new, more complex roles. Every single call is a problem the AI couldn't solve.

SPEAKER_01 10:01

And this brings up a really fascinating psychological nugget from the research on chatbot learning evidence.

SPEAKER_00 10:06

Oh, chatbot learning evidence. This is such an interesting piece of psychology.

SPEAKER_01 10:09

Unpack it for us. What does that mean?

SPEAKER_00 10:11

It's basically a study on how we forgive machines. It found that when a bot explicitly admits I'm learning and improving, customer tolerance for failure actually increases. We cut it some slack.

SPEAKER_01 10:23

So if the bot says, hey, I'm new at this, we treat it like a human trainee.

The Vigilance Tax Explained

SPEAKER_00 10:27

We do. We're surprisingly empathetic to the bot. But think about the workflow implication of that. The customer tries the bot, they're patient at first, but then the bot fails. By the time that customer finally gets routed to a human, their patience is entirely exhausted.

SPEAKER_01 10:42

Oh, I see. They have already tried the self-service route, hit a wall, and now they're done playing nice.

SPEAKER_00 10:47

Exactly. So the human agent is catching a fastball every single time. A complicated problem, a frustrated customer who has already wasted time with a bot and no easy wins to break up the day.

SPEAKER_01 10:58

Wow. So the cognitive load, the mental weight of the job is just skyrocketing. And this leads us directly into the dark side of this data. We are seeing a massive spike in what is called the vigilance tax. Let's really unpack this. I know 75% of contact center leaders are worried about agent well-being. But what exactly is the vigilance tax? Yeah. Because it sounds like something the IRS charges.

SPEAKER_00 11:23

It does sound like that. But it's a cognitive cost. Imagine you are the agent. You are on a call with a very difficult emotional customer.

SPEAKER_01 11:31

Okay. I'm there.

SPEAKER_00 11:32

And as you are talking, an AI co-pilot is listening in. It pops up a script on your screen that says, offer a 10% discount.

SPEAKER_01 11:40

Which seems helpful.

SPEAKER_00 11:42

Maybe. But now you have a split brain situation. You have to listen to the customer and gauge their actual emotion. Are they actually angry or just loud? You have to check your own instinct and you have to read the AI's suggestion.

SPEAKER_01 11:54

All at the same time.

SPEAKER_00 11:55

Yes. And then you have to decide in a split second, is the AI right or is it being tone deaf?

SPEAKER_01 12:00

Aaron Ross Powell Right. Because if the customer is crying about a lost family heirloom, offering a 10% coupon is incredibly insulting.

SPEAKER_00 12:08

Exactly. So the agent has to audit the AI in real time. That extra cognitive effort to monitor, evaluate, and potentially correct the system that is supposed to be helping you, that is the vigilance tax, you aren't just doing the job anymore. You are supervising the machine.

SPEAKER_01 12:25

There was a quote in the source material that really stuck with me. An agent actually said, I feel like I'm in a driving test that never ends.

Scoring, Surveillance, And Burnout

SPEAKER_00 12:32

That is the perfect analogy. Think about a driving test. You know how to drive, but you are hyper-aware of being scored. You are tense. Your hands are at 10 and 2. Modern systems now score 100% of calls.

SPEAKER_01 12:46

Wait, 100%. It used to be your boss listened to maybe one or two calls a month for quality assurance.

SPEAKER_00 12:52

Not anymore. Now the algorithm scores every single pause, every deviation from the script, every sentiment shift. So even if the agent completely delights the customer and solves the problem, if they didn't use the exact phrase the AI prompted, they might get docked points. Or at least they fear they will.

SPEAKER_01 13:08

It sounds like a panopticon, just constant, unending surveillance.

SPEAKER_00 13:12

It creates a permanent state of high alert. And the cost of that stress is very real. 87% of agents report high stress levels. And attrition, the rate at which people quit, has jumped from 42% in 2022 to 60%.

SPEAKER_01 13:28

60% turnover. That is entirely unsustainable. You're essentially replacing your entire workforce every year and a half.

SPEAKER_00 13:36

And it is wildly expensive. Replacing a single agent costs between$30,000 and$40,000 when you factor in recruiting, training, and the ramp up time to get them proficient.

SPEAKER_01 13:46

So the companies are saving all this money on efficiency with AI on the front end, but they might be blowing it out the back door by burning through their staff.

SPEAKER_00 13:53

That is the deep irony here. The tools were meant to reduce cognitive load. You know, here, let me write your notes for you. But because they have turned into these digital supervisors, they are actually increasing the pressure.

Designing For 2026: Support Over Control

SPEAKER_01 14:05

So we have this paradox. We want the speed, we want the 30 to 50% reduction in handle time. But we can't treat the humans like robots or they will just leave. How do we fix this? What does the design for 2026 actually look like?

SPEAKER_00 14:19

The sources offer some really concrete solutions for this. It starts with shifting the entire philosophy from control to support.

SPEAKER_01 14:27

What does that mean in practice, though?

SPEAKER_00 14:28

Well, transparency is number one. Agents need to know exactly how the scoring works. It cannot be a black box algorithm that just judges them from on high. If the AI scores a call poorly, the agent needs to be able to see why, and crucially, they need to be able to challenge it.

SPEAKER_01 14:44

Right, so they actually know the rules of the game they're playing.

SPEAKER_00 14:46

Exactly. Secondly, is autonomy. This is huge. Agents must be empowered to override the AI without any penalty. If the AI says, read this legal disclaimer, but the agent knows the customer is about to cry and needs empathy first, the agent needs to be able to make that call without fearing their scorecard will turn red.

SPEAKER_01 15:05

That makes total sense. Trusting the human to actually be human.

SPEAKER_00 15:09

Yes. And this requires a fundamental role redesign. We need to stop thinking of these people as phone answers or just entry-level labor. If they are handling the complex 47% of issues that the AI simply cannot touch, they are actually highly skilled problem solvers. We should be calling them AI outcome managers.

SPEAKER_01 15:26

That's a great title. But are there metrics that actually reflect this new reality? Because if we just keep measuring average handle time, aren't we just pushing them to go faster regardless of how complex the problem is?

SPEAKER_00 15:38

You are spot on. If you give an agent a complex problem but measure them on a metric designed for simple problems, you create instant burnout. The sources suggest moving beyond just AHT to look at total resolution time and crucially, employee satisfaction. You cannot have high customer satisfaction with miserable employees.

SPEAKER_01 15:57

Is there proof that this balanced approach actually pays off? Or is this just kind of wishful thinking from the industry?

SPEAKER_00 16:04

No, there is proof. The sprinkler source notes a potential 210% ROI over three years. But and it is a very big plot, but that return is only possible if the human element is stabilized. If you have 60% attrition, you will never see that ROI because you are constantly training new people.

SPEAKER_01 16:23

So the money is there, but you have to treat your people right to actually get it.

SPEAKER_00 16:28

The efficiency requires stability. You can't build a race car if the wheels keep falling off.

SPEAKER_01 16:33

Okay, let's bring this all together. We have moved from a world of manual data entry and please hold while I look that up to a world of AI-driven speed. We are seeing handle times drop by half. We are seeing market values for this tech explode to over$120 billion.

SPEAKER_00 16:49

Aaron Powell But we have created a pressure cooker. We have stripped away the easy work and left the humans with the hardest, most emotional tasks, all while an algorithm watches over their shoulder.

SPEAKER_01 16:58

Aaron Powell It really feels like we're at a crossroads for 2026.

Closing Dilemma: Empowerment Or Compliance

SPEAKER_00 17:01

Aaron Powell We are. And I think the final question for everyone listening, whether you run a customer service team or you're just calling one, is this dilemma of the algorithm never blinks.

SPEAKER_01 17:10

What is that dilemma exactly?

SPEAKER_00 17:11

As we move forward, are we building technology that empowers agents to use their judgment? Are we giving them superpowers? Or are we building systems that quietly train them to stop thinking for themselves because it is simply safer to just follow the prompt?

SPEAKER_01 17:28

That is a heavy thought. If we train people to just follow the blue dot on the screen, do they lose the ability to actually help us when we need it most?

SPEAKER_00 17:37

And when the machine breaks and it always breaks, eventually who is left to fix it?

SPEAKER_01 17:41

That is something for you to think about the next time you're on the phone with support and you hear that long pause. They might just be arguing with their AI.

SPEAKER_00 17:48

Very likely.

SPEAKER_01 17:49

Check your own vigilance tax today, everyone. Thanks for taking this deep dive with us.

SPEAKER_00 17:53

See you next time.