When it comes to conversations about AI in the contact center, more often than not, the technology discussion is thorough, but the governance discussion is rarely so.
How organizations are actually going to manage, measure, and hold accountable a workforce that includes both human agents and autonomous AI systems tends to get treated as a detail to figure out later.
It's like a toddler begging their parents to go to McDonald’s. They aren’t thinking about the time it takes to drive there, who’s going to watch their younger sibling while they’re gone, or the cost; they’re thinking, ‘I want a Happy Meal!’
While the ramifications of an impromptu fast-food dinner aren’t quite comparable to a full-scale contact center AI implementation process, in both instances, the “figure out later” can end up costing you more than you’d think.
This is certainly the opinion of Dave Rennyson, CEO of SuccessKPI, who, across two conversations with CX Today, laid out the case for rethinking performance management from the ground up, before the AI goes live, not after.
Check out the full interviews here:
- Stop Letting Your AI Agents Off the Hook
- Contact Center AI Is Only as Good as the Data Behind It – So Why Are We Ignoring the Data?
Start With the Foundation You Already Have
Rennyson's starting point moves beyond a simple technology recommendation to a diagnostic question:
“‘How good am I in my automated appraisal of my human agents today?’ Because you're not going to be able to appraise conversational AI or any generative AI agentic system at scale in a manual fashion. You're not going to be able to keep up.”
Automated quality management for human agents is the foundation. Without it, there's no consistent, scalable mechanism for validating what AI agents are doing either.
Get that in place first, then use it to identify the most viable candidates for automation, and again to confirm the automation is actually delivering, not just deflecting volume.
The Step Almost Everyone Skips
Even among organizations that have done the QM groundwork, there's a step in AI deployment that Rennyson says is almost universally missed: establishing ground truth.
Before going live with an AI agent, you take a sample of real conversations, transcribe them, replay them synthetically, and observe the AI handling them manually.
The result is a documented benchmark against which every future release can be tested, as Rennyson explains:
“When you're done, you know that the AI is competent and capable of handling that task without hallucination.”
The real value, though, is protection against model drift.
As underlying models improve, the risk is that AI may overstep or behave unpredictably in edge cases.
“You have to ensure that you don't have model drift, where maybe the (AI) agent was great at this task when it was a little less smart than it is now, three releases later,” he says.
If it’s so important, why does this step keep getting skipped?