As AI's Second Act Emerges in CX, Raw Data Paralysis Finds a Cure In Outcome-Driven Coaching

Once upon a time, contact centers coached agents based on a sliver of reality. A manual QA team might review 1% of an agent's calls in a given month, and supervisors would build entire performance plans around that fragment. Now, AI reviews 100% of conversations and surfaces trends in real time. The question for leaders is no longer whether they have the data, but whether they know what to do with it.

We spoke with Andra Harders, Customer Success Manager at Cresta, who has spent nearly a decade leading and supporting contact center teams. Earlier in her career, she managed large sales organizations at TTEC supporting Meta’s SMB advertising program, overseeing more than 80 agents and consistently surpassing quarterly performance targets. Today, she partners with enterprise contact centers to implement AI-powered coaching and conversation intelligence tools, translating full conversation visibility into measurable gains in sales, efficiency, and agent development. She says that once every call is on the table, the narrative around performance starts to look very different.

"When we take full visibility into every conversation, the key isn't just seeing more data. It's knowing which signals matter, and translating them into fair, outcome-driven coaching," says Harders. Instead of catching a single bad call and building a narrative around it, supervisors can now distinguish a one-off from a pattern.

• One-off or trend: "Instead of only having a niche number of conversations that a manual team would review, we're now able to review 100% of their conversations and say, 'That actually was a one-off. Leave that agent alone.' Or, 'This is a trend we're seeing over time,'" Harders says. That distinction matters, especially in hybrid and remote environments where supervisors can no longer read body language or check in casually. "If you see somebody who was performing really well and then dropped, maybe check in with them first. Don't run into a coaching session and say you're doing really bad. Maybe pause and see how things are going."

She points to a large cruise company that put this approach to the test. The company needed to reduce handle time across its global operation, but the drivers of high handle time were mixed. Some were unavoidable, like explaining new ship amenities or updated booking policies. Others were within reach: agents leaving guests on hold too long, or spending upwards of ten minutes after each call typing up notes.

• Automated notes, reclaimed time: "We had agents from the US, Manila, Guatemala, all over the world, and their notes were all varied across different structures," Harders explains. "Through summarization tools, we leaned into, 'You no longer take notes. They're automated for you.' We did an A/B test, and every single agent outperformed the group that was not using Cresta." The results gave the business the evidence it needed to move from a small pilot to a global rollout.

• Healthy competition: The cultural shift was just as telling. Agents in the test group started comparing metrics and asking each other how they were using the tools differently. "Agents who weren't even using Cresta were saying they wanted to," Harders recalls. "Supervisors would come into meetings and tell me, 'My team is chomping at the bit.'"

What made the adoption stick, according to Harders, was linking the technology directly to outcomes agents cared about, not introducing it as another dashboard to monitor. Performance metrics connected clearly to commission, promotion paths, and daily goals. Agents hit commission for the first time. Others stepped into supervisory roles with documented proof of consistent performance.

• Cultural bridge: For globally distributed teams, the benefits go beyond efficiency. "Agents in the Philippines don't know what the Texas floods are. They don't know what Johnny Walker is," Harders says. "We're able to give them power and confidence in those conversations so they still sound and feel like an expert."

• Trust takes time: Earning agent buy-in requires acknowledging their skepticism upfront. "There's that element of, 'I don't trust this right away. It's AI. I know what a hallucination looks like,'" Harders says. "Building the solution with them, not just for them, is what makes the difference. It works best when they hear it from a peer, not a vendor."

Harders acknowledges that the volume of data available to contact center leaders today can feel paralyzing. But the alternative, making high-stakes coaching decisions off a 1% sample, is the far greater risk.

"We used to pay people based on QA scores when we only scored 1% of their calls over a whole month. Was that fair? No, but that's all we knew," Harders says. "Now we're rethinking what we used to do and how we improve it with more visibility and more data. The parsing and distilling of it is what's really important, just to ensure you're still focused on the overall outcome. You just now have more signals to help you make those decisions."