Avoid metric manipulation
Of course, even the best-designed metrics can backfire if they’re implemented poorly or incentivize the wrong behaviors.
In our work with clients, we’ve seen agents hang up quickly to lower their AHT, “selectively” send surveys only to happy customers to boost NPS and find other creative ways to hit targets. At one major retailer, employees even scanned QR codes on receipts themselves to submit fake customer satisfaction surveys. Another company had to create a dedicated team to detect and stop this kind of survey fraud.
This is a classic example of Goodhart’s Law: “When a measure becomes a target, it ceases to be a good measure.” If employees feel they can’t control a metric – or that their job depends on an unrealistic target – they will find ways to game it.
To guard against this, companies should:
Trust but verify: Use telemetry and customer behavior data to validate outcomes rather than relying solely on self-reported metrics. For example, tie your “resolved” status to actual product usage or the absence of repeat calls, not just an agent’s declaration that the issue is fixed.Close the loop: Implement a closed-loop resolution process (e.g., follow up with customers or automatically reopen cases if the customer contacts you again about the same issue) to ensure issues are truly fixed and stay fixed.Align incentives: Make sure the metrics you set are within employees’ control and directly tied to the customer experience you want to deliver. If you ask agents to provide a great experience, measure things that reflect customer success – and train and empower employees to achieve those metrics in the right way.
Then – and only then – bring in AI
Once you have experience-focused, credible metrics in place, you can bring in AI to help improve those outcomes. In other words, get the metrics right first, then let the machines help.
The most effective customer service AI deployments are targeted, not one-size-fits-all. Instead of trying to automate everything, identify specific high-volume or high-impact scenarios where AI can excel. For example, a well-trained bot might handle password resets, order tracking or simple troubleshooting steps much faster than a person, freeing up human agents to focus on more complex needs.