Oura’s business is built on turning round-the-clock signals from your body into insights you can act on, so that health becomes a daily practice instead of a guessing game. When we looked at our own support through that same lens, the gap was obvious—the experiences that mattered most were the ones we could not clearly see.
To address this, we set out to find a better way to measure those experiences and developed a new metric: the Total Customer Experience Score, or TCX. TCX uses AI to evaluate and analyze nearly every member interaction across every channel, so we can fully grasp the experience for everyone we talk to, not only the handful who fill out a survey. Here is how it works, why we built it, and why it matters.
“TCX gives us a strategic edge, establishing our early mover advantage and leadership in AI-powered CX innovation. But more importantly, it helps us understand our members better than we ever could, enabling us to provide a human first support experience aligned with the ethos of Oura’s brand.”
—Tyler Schleich, SVP, Business Operations
Why the old metrics fall short
Metrics like CSAT and NPS were built for a different era of support, one with lower volumes, human-only channels, and the assumption that a small survey sample could speak for everyone. That assumption stopped being true a long time ago. In fact, Gartner reports that while surveys are used by 93 percent of customer service organizations, their value is limited by low response rates and narrow scope.
Surveys reach a small, self-selected group, and the feedback lands well after the moment has passed. Manual quality review fills in a little more, but even a well-staffed team can hand-audit only an estimated five to seven percent of solved tickets. Almost everything else goes unmeasured.
In one study of call center conversations, research found that limited response can distort the perceived performance of a support operation and leave missed opportunities for coaching, follow-up, and service recovery.
That leaves a blind spot every support leader knows. A small and often frustrated minority sets the tone of the score—the experiences of everyone else stay invisible, including the ones that went well and the ones that quietly went wrong. It is hard to fix what you cannot see—for most companies, most of the customer experience has never been visible at all.
And the ground is still shifting. AI now handles a growing share of support conversations, and the old tools were never built for that world. A survey captures a moment in time and a manual review grades a handful of conversations. Neither can keep pace with an AI agent that works at a scale where a single flaw can repeat thousands of times before anyone notices. If AI is going to do more of the work, something equally scalable has to watch how well it does it.
Getting insight into these interactions matters because member experience is brand experience. Every support interaction disproportionately shapes trust, and trust shapes how members perceive Oura. Improving the customer experience is not just about efficiency or service quality; it is about strengthening the relationship members have with the Oura brand.
What TCX measures
TCX is a framework built using large language models (LLMs) to evaluate an interaction by reading the actual conversation through the lens of the member. It intentionally ignores internal notes and backend data and focuses on what the member went through by receiving signals throughout a support interaction. Because TCX reads the conversation itself, it makes no distinction between channels—an interaction that starts with our AI support agent is held to the same standard as one that starts with a Member Care Representative.
Four AI classifiers score every interaction, each answering one question:
- Issue identification: Was the member’s issue fully and correctly understood?
- Frustration: Did the member express frustration with the support experience?
- Resolution: Was the issue resolved or properly escalated?
- Satisfaction: Was the member’s sentiment positive after the solution?
We set the bar high on purpose—an interaction only passes if all the questions that make up the logic have a positive outcome. For example, a conversation that solves the problem but leaves the member dissatisfied still counts as one we need to improve.
Those four judgments combine into a single score that is more than just a reporting metric. The numbers below are illustrative (not real), but they’re a useful way to see a few different ways we can look at this data.
It gives our teams a shared view of the member experience from individual coaching conversations to operational reviews and company-wide reporting. Instead of relying on a small sample of survey responses, we can make decisions based on a nearly complete picture of how members actually experience Oura.
TCX is regularly calibrated against evaluations from Oura’s quality experts, using established accuracy measures to test that the AI remains aligned with human judgment. When the two disagree, people (not AI) set the standard, and we re-tune the model so its outputs stay benchmarked against that standard.
From signal to fix
Seeing everything only helps if you act on it. Because TCX analyzes every conversation, it does more than register that members are frustrated. It shows us where the friction is and why—across thousands of interactions at once.
Finding the root cause of an emerging issue used to take days or weeks. Now it takes under 30 minutes.
A few examples, with the details anonymized.
- During the Oura Ring 5 launch, TCX caught that some members were getting the wrong return-policy guidance from our AI support agent. We spotted the pattern immediately and fixed it within hours, and contacts about the issue dropped by nearly 50 percent.
- TCX also surfaced a routing problem no one had noticed. Members who bought online and picked up in a store were being pushed through the wrong return process. Once we corrected the routing, misdirected cases fell by roughly 30 percent.
- In a third case, TCX flagged a gap between our automated battery diagnostics and what members were seeing with their own hardware. That prompted us to update an outdated diagnostic model so our automated guidance lined up with real-world performance.
Each of these began as a signal in the data and ended as a fix that helped every member who would have hit the same wall. Since we rolled TCX out, our TCX score has climbed nine points, repeated troubleshooting has dropped from 40 percent to around 20 percent, and median handle time on live chat is down by more than 25 percent.
Putting our people on the work that matters most
Understanding every conversation also gives our Member Care Representatives and team leaders something concrete to work from: a clear view of what actually works. They can see which approaches turn a frustrated member around, how the harder cases get resolved, and what changes the outcome of a conversation. Leaders use that to coach on the things that decide a support experience, like empathy, de-escalation, and taking ownership from the first message.
That is what measuring at scale really buys us—it lets us point our people at the work that needs them and it gives them what they need to do it well. Being member-first used to mean doing our best with a thin, skewed slice of feedback. Now we can see what every member actually experienced and act on it for all of them.
“This isn’t just a measurement tool, it’s a coaching tool. For the first time, we can reliably see what separates a good conversation from a great one across every interaction, showing us exactly where our Member Care Representatives excel and where they need support. That greater visibility is how we raise the bar and hold the entire operation to a higher standard.”
—Dana Barat, Senior Director of Global Support Operations
**The standard we hold ourselves to **
The metrics that defined customer experience for decades were made for a world that no longer exists. As more support runs on AI, the companies that earn lasting trust will be the ones that put AI to work evaluating AI in real time, with a human in the loop to validate its judgment—not the ones still waiting on a survey to tell them how they did.
We don’t believe Oura will be the only company facing this challenge. We built TCX to solve it for ourselves, but we also believe it points to where customer experience measurement is headed more broadly.
Every interaction contains signals about trust, product quality, operational friction, and the customer relationship. Companies should be able to learn from all of them, not just the small percentage of people who answer a survey.
TCX is how we hold ourselves to this higher standard. It means that behind every conversation, whether it starts with our bot or a Member Care Representative, we are paying attention. Health is a daily practice, and so is earning our members’ trust, one conversation at a time.