AI in Customer Operations: Big Promises, Low Returns. Why?

According to the latest report published by Boston Consulting Group (BCG) in September 2025, the picture is quite striking: only 5% of companies worldwide are able to generate significant value at scale from their AI investments. The vast majority, the remaining 60%, have yet to see any tangible return.

So what is this “elite 5%” doing differently? The answer usually lies not in the technology itself but in how they measure success.

Trying to evaluate the new generation of “Agentic AI,” capable of making its own decisions, with traditional metrics is like judging a smartphone by telegraph-era standards. Let’s look at the two critical metrics we need to update to see AI’s true impact.

1. Average Handle Time (AHT or ACHT): ‘Shorter’ Is No Longer Always ‘Better’

For years, the golden rule in call centers was clear: “Keep AHT (Average Handle Time) low and cut costs.” But in the age of AI, that approach no longer holds.

AI now handles simple, routine tasks (password resets, shipment tracking and so on) in seconds without human intervention. What’s left? Complex cases that human agents need to resolve, cases that require empathy and by their very nature take longer.

So as long as your resolution rates don’t drop, a rising AHT may actually be proof that your AI is working properly, not a sign of failure. It means your agents are now dealing not with easy issues but with tougher customer problems.

Instead of focusing only on time, track efficiency with these new metrics:

• Intents per Contact (IPC): How many different issues are resolved in a single interaction.
• AHT per Intent: The resolution time for each individual topic, rather than total time.
• AI Resolution Rate: The percentage of work AI handles on its own.
• AI Agent Usage Rate: How often agents act on the recommendations of AI agents.

2. First Contact Resolution (FCR): More Than ‘Yes’ or ‘No’

“Did we resolve the issue on the first call?” (FCR) has long been the industry standard. But this “Yes/No” approach falls short of measuring the quality of experience today’s customers expect.

An issue may technically look “resolved,” but if the customer was worn out or frustrated along the way, that customer is at risk. Remember, the vast majority of consumers are ready to abandon a brand after a single bad experience. And a customer may reach out again about a different issue precisely because they were so pleased with your AI assistant; traditional FCR counts this as a “failure,” when it is actually a sign of trust.

That’s why you should introduce metrics that focus not just on the outcome but on the quality of the process:

• Sentiment Shifts: How the customer’s mood evolves during the interaction (for example, from tense to satisfied).
• Intent Recognition Accuracy: How quickly and accurately the AI understands what the customer wants.
• Resolution Path Quality: The level of friction (unnecessary transfers, repetition) the customer experiences on the way to a resolution.

Bottom line: All these metrics may seem a little futuristic, but the core principle never changes: You Can’t Prove Value You Can’t Measure.

As we enter the era of “Agentic AI” in particular, trying to measure the new world with old yardsticks could leave you stuck in that 60% group of companies “not seeing value.” The smartest approach is to use data not just to report on the past but to optimize the future.