We Automated Too Much. Here’s What I’d Do Differently

For years, I sold automation for a living.

As a contact center platform provider, my team and I ran countless demos showing operations heads how many calls an IVR could deflect, how a chatbot could cover the night shift, and how a few smart rules could shave seconds off every interaction. The spreadsheets always looked beautiful: fewer agents, lower cost per contact, service around the clock. Who would say no to that?

Almost nobody did. And for a while, it worked.

But somewhere along the way, I started noticing a pattern in the review calls that followed those go-lives. The dashboards said containment was up. The customers were saying something else entirely.

The Day the Dashboard Fooled Me

I remember sitting with a client’s operations team, looking at a report they were proud of. Well over half of their inbound contacts were now being handled without an agent. Then we pulled up a sample of recordings from the same week.

One caller had pressed zero eleven times. Another had said “representative” in so many different ways that it was almost poetry. A third hung up, called back, chose the option for new connections because that queue was always answered fastest, and then apologized to the agent for lying to the menu.

Every one of those calls counted as contained at some point in the journey. None of them were resolved.

That was when I understood we had been measuring the wrong thing. We were counting how many customers the system kept away from people, not how many problems it actually solved.

Customers Aren’t Anti-Technology. They’re Anti-Dead-End

I want to be careful here, because the lazy conclusion is that customers hate automation. They don’t. Most of us happily check a balance, track a parcel, or reset a password without speaking to anyone. Nobody wants to wait on hold to ask what time a store closes.

What people resent is being trapped. They resent a bot that can’t understand a simple sentence, a menu with no option for their actual problem and, above all, having to tell their story again after finally reaching a human.

The research backs up what I heard on those recordings. A SurveyMonkey study found that 79% of US consumers strongly prefer dealing with a person over an AI agent, and 81% believe companies use AI mainly to save money rather than to serve them better. That second number stings, because it tells you customers can see the intent behind the design. When automation is built to protect the budget instead of the customer, people feel it within the first thirty seconds.

Another shift makes this more serious. With prices rising everywhere, customers are judging brands on value, and service is part of that value. Loyalty is thinner than it used to be. A frustrating service experience used to cost you goodwill. Now it can cost you the customer.

Where We Got It Wrong as an Industry

Looking back, I think platform providers like us made three mistakes, and our clients made them right alongside us.

We Automated the Hard Moments, Not Just the Easy Ones

It’s one thing to automate a payment reminder. It’s quite another to put a bot in front of someone disputing a charge, dealing with a bereavement, or worried about losing a service they depend on. Those moments carry emotion and money, and they need judgment. A script, however clever, doesn’t have that.

We Designed Handoffs as an Afterthought

In most deployments, teams treated the transfer from bot to agent as a failure state, so nobody invested in it. The result was predictable: the agent picked up with no context, and the customer had to start over. Customers now treat repeating themselves as the surest sign they’ve been talking to a machine, and it makes them angrier than the wait ever did.

We Hid the Machine

Some clients wanted bots that sounded as human as possible. I understand the instinct, but people want to know who, or what, they’re speaking to. When they find out halfway through, trust drops sharply.

What I’d Tell Any Contact Center Leader Today

If I were sitting across the table from you now, this is what I’d say, as someone who has sold both sides of this argument.

Don’t Rip It Out. Rebalance It

Going back to all-human service is neither affordable nor necessary. Keep the automation that genuinely finishes the job, such as status checks, simple payments, and appointment changes. Then be honest about where it’s failing, and send those journeys to people.

Decide in Advance Which Moments Belong to Humans

Write the list down: complaints, high-value accounts, anything involving hardship or large sums of money, and anything that could end up on social media. These should reach a person by default, not after three failed attempts.

Make the Handoff the Best Part of the Journey

Set triggers that move a customer to an agent before they ask, such as repeated failed intents, a rising tone of voice, or the same question asked twice. And make sure everything the customer has already said appears on the agent’s screen. If customers have to repeat themselves, your automation costs you more than it saves.

Tell People When They’re Talking to AI

A simple line at the start is enough. Honesty costs nothing, and it protects trust when things go wrong.

Let Your Agents Actually Fix Things

Many of the escalations I’ve seen got stuck because the agent had to ask a supervisor for every refund or waiver. Agree on clear limits for credits, refunds, and goodwill gestures, and let agents use them on the spot. A human who can’t help is just a slower bot.

Move AI Backstage

This is where I’ve seen the best returns. Use AI to summarize calls, suggest the next best answer, flag a frustrated customer in real time, plan shifts more accurately, and review every call for quality instead of a small sample. Customers never see this AI, but they feel its effect: shorter waits, better-prepared agents and quicker resolutions.

Hire and Train for Judgment

As routine work moves to automation, the calls that reach people get harder. Gartner found that 84% of customer service leaders plan to add new skills to agent roles. The agent of the next few years is part problem-solver, part keeper of knowledge, and part the person who decides when the rules should bend.

Change What You Measure

Stop celebrating containment for its own sake. Track first-contact resolution, repeat contacts, and how often customers have to explain themselves twice. Those numbers tell you whether people are being helped or just being handled.

The Point of All This Technology

I still believe in automation. I’ve seen it take thousands of tedious calls off agents’ plates, and give customers answers at 2 am that they would otherwise have waited until morning for. That’s real value.

But a contact center was never meant to avoid conversations. It was to resolve them. The best technology I’ve seen does exactly that: it handles the simple things quietly, recognizes when it’s out of its depth, and hands the customer to a capable person without making them start over.

When customers ask to speak to a human, they aren’t rejecting technology. They’re telling us the technology has stopped listening. Our job, as the people who build and run these systems, is to make sure someone always is.