The champion-challenger model in the contact center: How to evaluate before you commit

You can spend six months evaluating a new contact center partner or AI tool and still end up having to guess whether it will pay off.

I've spent close to twenty years working in customer experience, moving from global contact center operations to consulting on transformations, and now helping innovative brands develop, enhance and orchestrate their CX AI strategies with TELUS Digital. In that time I've partnered with leading organizations handling upward of a million calls and interactions a week, even after containment. At that scale, even a small performance discrepancy between the right partner and the wrong one compounds.

With the stakes and complexity so high, it’s perhaps no surprise that I've seen leaders conduct a lot of careful, well-run vendor evaluations, RFPs, reference calls and proof-of-value workshops. It's clear that these forms of due diligence are an attempt to bring a sense of certainty. Unfortunately, they all draw on evidence from elsewhere. They don't show how a new operator, or a new AI tool, will perform in their environment, with their customers and their volume.

Forrester's 2026 predictions report for customer service warns that service quality will dip industry-wide as organizations wrestle with the complexity of AI deployment. The analyst firm estimates that three-in-ten enterprises will need to build entirely new internal functions just to manage AI. What proactive measures will you take to avoid that dip and a resultant hit to your bottom line?

If, like many, you're investigating a new outsourcing partner to strengthen and execute your CX AI strategy, with the complexity and stakes so high, you should know how the new approach will perform in your business. The champion-challenger model is how you find out. The champion is what you run today. The challenger is the new approach, operator or tool that you run alongside the champion to compare. Crucially, you run both on live volume, letting the results inform how you move forward.

This is my perspective on the champion-challenger model, focused mainly on how it applies to outsourcing providers, with a look at how the same discipline extends to AI tooling, and how to deploy it without putting your customer experience at risk.

The champion-challenger model turns a CX decision into a controlled test

The champion can be another outsourcing partner, your own in-house team or an AI tool already deployed for a specific use case. The challenger is what you run live alongside the champion, for a defined period of time, in order to measure its performance relative to what you are doing today.

Think of it like a science experiment or an A/B test. You want to know the one thing you're changing (your independent variable), everything you’re holding constant (your controlled variables) and what happens as a result (your dependent variable).

An example of the champion-challenger model for CX partners

Let’s say your challenger is a new approach to onboarding and coaching. Specifically, AI-assisted training simulations paired with automated monitoring that feeds coaching back to agents.

The goal is to test that approach against your champion, in this example, the traditional classroom training and manual interaction review you rely on today. To perform the test, you hold your controlled variables steady — similar volume, the same time period, comparable customer segments — so any difference you see can be attributed to the change itself, not something else going on in the background.

As one dependent variable, you track how quickly new agents reach full proficiency, and you find the challenger gets them there significantly faster. But then you notice another dependent variable, the number of coaching flags per agent, rises, which initially causes concern. By diving a little deeper, you identify that the flags increased because automated monitoring is catching issues manual review used to miss, not because of declining performance. Ultimately, those flagged moments get addressed through pointed coaching before they become habits, and quality scores, your key dependent variable, increase as a result.

Though the model isn’t new, the use case for deploying it is growing as CX and contact center tooling becomes more complex.

Test the operator's use of AI, not just their headcount

Gartner has published research recommending this approach for outsourcing, noting that a champion-challenger model helps maintain the performance gains that outsourcing is meant to drive. That competition only sharpens performance if the test measures what drives it, not just headcount and tenure. Increasingly, what drives it is AI.

A study titled Generative AI at Work from MIT and Stanford researchers, published in the Quarterly Journal of Economics, found that when customer support agents had AI-assisted coaching built into their workflow, productivity rose 15% on average and newer, lower-tenure agents saw gains of roughly 35%. Two operators can staff a queue with the same headcount and the same tenure and still perform very differently, depending on whether one has built AI into how it trains and coaches agents and the other hasn't. That's the kind of difference an operator-level test should be designed to catch in 2026, since it's the coaching methodology behind the agents, not simply their tenure on the account, that shows up in the results.

What matters here is what the operator does with AI inside its own delivery model, including the training, coaching and quality process behind the agents on a call. In practice, that means asking a specific question during a controlled test: how quickly does a new agent on the challenger's team reach the same proficiency as a veteran on the champion's team, and how much of that speed comes from AI-assisted coaching versus experience alone.

TELUS Digital President Tobias Dengel recently made a related point, stating that agent coaching should be optimized to genuinely serve the customer on the line. That's the same discipline a test needs to capture.

Applying the champion-challenger model to your CX AI tools

As mentioned earlier, the same model extends one level down, to the tools themselves, just with a narrower scope. Instead of testing operator against operator, you test one AI tool against another, on a single, specific use case.

This helps counteract a mistake I see all the time, where CX leaders assume that because a customer-facing AI tool, a chatbot or a virtual agent, performs well in one kind of conversation, it'll perform just as well in another. It usually doesn't.

Two things tend to separate these tools.

1. Complexity: Some tools are built for high-volume, lower-complexity interactions, quick answers and simple transactions, while others are built for higher-complexity conversations that require more reasoning, and typically cost more to run as a result.
2. Channel: Some tools are strong at chatbot-style text interactions and comparatively weak at voice, while others, built around voice and virtual agents, aren't built for text at all.

Applying the model here starts with picking one specific use case, like a loan modification request or a voice-based appointment confirmation, not an entire channel or book of business. From there, the champion is whatever tool already handles that use case today, and the challenger is tested against it in parallel, on comparable volume, with KPIs matched to the job. That could be resolution accuracy and compliance adherence for a loan modification, for example.

Examples of the champion-challenger model for CX AI tooling

Say you’re a financial services provider running a largely manual loan modification process, with agents working from a script to handle compliance-heavy conversations. That’s the high-complexity end of the spectrum described earlier, and it's a good candidate for a champion-challenger test. Select an AI solution built for exactly this kind of judgment-heavy conversation, and test it against your current process. Cresta, for example, offers real-time agent assistance and coaching capabilities designed for large, voice-heavy contact centers in regulated industries. Measure the result on resolution accuracy and compliance adherence, not a generic CSAT score, since that's what tells you whether the AI-powered approach handles judgment as well as your live agents.

Or, consider a use case involving routine appointment confirmation calls for a healthcare company. The job to be done is less about judgment and more about clear, natural communication over the phone. That's the channel side of the spectrum described earlier. Test a voice-based AI agent against your current approach. ElevenLabs, for example, has become foundational voice infrastructure across the industry, largely because of how natural its synthesized voice output sounds. Measure the result on completion rate and speed.

With this discipline across your CX operation, you arrive at a deliberately assembled set of tools, each earning its place because it was tested against the specific job it is doing.

What TELUS Digital has proven as a challenger

TELUS Digital has run and won champion-challenger controlled tests at enterprise scale by deploying AI-assisted agents.

How to run a champion-challenger controlled test without risking the customer experience

A well-run controlled test protects the customer relationship while it's proving the new approach, so there's no all-or-nothing bet at any point. The stages are consistent whether you're testing an operator or a tool.

• Baseline current performance
• Agree on shared KPIs upfront
• Run the challenger in parallel, at a defined scope and volume great enough that the results reflect a meaningful pattern
• Measure both using live production volume
• Scale the challenger only if it wins

If the challenger doesn't outperform the champion, the client hasn't bet the whole operation on it. The risk and cost are contained. From there, you might apply what you learned to your current operation and pursue further tests.

Prove it before you commit to it

You don't have to guess whether a new operator or AI tool will pay off. That's exactly what a champion-challenger controlled test is built to answer, whether you're evaluating a new outsourcing partner or an AI tool.

If you're evaluating a partner or a tool right now, don't settle for projections. Test the approach against what you already have and let the results make the decision. Talk to our team about running a champion-challenger controlled test on your own operation.