Why Contact Center ROI Falls Short (1/2)
Part 1 asked whether we solve customer problems or simply handle calls better.
In practice, it’s easy to pick a technology before answering that question.
1. Automation usually starts with the work we already do.
Calls are expensive, so we introduce chatbots and voicebots.
Agents search for knowledge, so AI recommends answers.
Agents classify and summarize calls, so AI takes over those steps.
None of this is wrong. I’ve built these systems, and they can work.
But we tend to start with the task we see, rather than asking why it needs to happen.
Before asking whether AI can do it faster, shouldn’t we ask whether the task still needs to work this way?
2. Moving work isn’t the same as redesigning it.
I saw this in chatbots, voicebots, and agent assist.
I was assigned to automate outbound calls. Before automation, agents attempted only half the day’s list, and timing mattered. Rather than copy their workflow into a bot, I asked what the bot could change. We contacted recent cases first and placed concurrent calls.
The share of assigned customers we attempted to call rose from about 50% to 100%, and the higher execution rate also improved results.
But in other projects, a new channel merely repeated what an app already offered. Customers got another place to try, without a better way to finish their task.
I also built an AI system that recommended the call categories agents used to select themselves, and improved its accuracy. But experienced agents often made that decision during the call, before the AI recommendation arrived.
For experienced agents, what mattered was not just accuracy, but whether the recommendation arrived when they needed it.
Different tools, same question: what actually changes for the customer or agent when we add this technology?
3. Not every existing task deserves automation.
Some inquiries should be automated because customers can complete them reliably.
Some work should be redesigned around the strengths and limits of AI and people.
Some decisions need human judgment.
And some work should disappear because the product or process that caused it can be fixed.
Those choices call for different investments. They aren’t all variations of the same automation target.
Our chatbot team tracked monthly active users (MAU), aiming to shift calls to chat. Yet one fix removed the need to contact support at all. It reduced customer effort and support costs without raising MAU. The metric captured channel adoption but missed the business value of preventing demand. Product ownership also means seeing when a better outcome lies outside the product.
Many workflows were designed around earlier technical and organizational limits. When those limits change, copying the old steps into AI may make little sense.
Knowing what we can build isn’t enough.
We also have to decide which problems are worth solving, where to make a change, and what work might no longer be needed.
Even a well-defined problem goes nowhere if the people who can change things aren’t involved.
#AIOperations #WorkflowDesign #AIProductManagement