Most companies measure customer satisfaction (CSAT) the same way they did a decade ago: send a survey, wait, count the responses, and hope the 8% who actually replied are representative of everyone else. It’s a system built on hope. And hope, at a global scale, isn’t a strategy.
Uber decided to do something different.
The company’s Global Digital Experience team, the group sitting at the crossroads of customer support operations and engineering, started asking a harder question: what if you could infer how every single customer felt, even the ones who never filled out a form?
The answer became an artificial intelligence (AI) engine that analyzes the full universe of support interactions in real time, surfacing satisfaction signals that traditional surveys simply can’t see. No waiting for voluntary feedback, no sampling bias, and no blind spots.
The Three Measuring Levers
The model is built around three pillars: Resolution, Effort, and Sentiment.
Resolution is the foundation. Did the customer’s problem actually get solved? Effort is the friction audit. How hard did the customer have to work to get there? And Sentiment is the hardest piece: tracking the emotional arc of an interaction from first message to final reply, measuring whether someone left feeling better or worse about the brand than when they arrived.
None of these is new in isolation. What’s new is the synthesis, such as weaving together transactional data, real-time trip telemetry, conversation logs, turn counts, and tone signals into a single, coherent picture of what a support experience actually felt like.
The Lessons Were Hard-Won
Building this wasn’t a clean sprint. Uber’s team quickly discovered that years of CSAT data had given them a false sense of understanding. Once they started peeling back layers to define more nuanced sub-metrics, they found complexity that legacy surveys had been quietly papering over all along.
Aligning stakeholders across different business lines, languages, and markets on one shared definition of “satisfaction” required iteration after iteration. Teaching an AI model not just that it failed, but why, within a specific cultural or operational context, turned out to be genuinely hard work.
It’s the kind of friction that only makes the output more valuable.
What’s Next for Digital Experience at Global Scale
The near-term unlock is significant: for the first time, Uber can compare performance across fundamentally different support technologies (legacy automation and modern conversational AI) using a normalized metric. Apples to oranges, finally made comparable.
But the longer-term vision is more ambitious. The team sees a future where AI doesn’t just measure satisfaction; it anticipates friction before customers feel it, resolves issues without a single click, and transforms every support interaction from a transaction into a trust-building moment.
That future isn’t fully here yet. But the infrastructure being built now is what makes it possible.
Want the full story? Access the complete case study, including how Uber’s team structured the cross-functional build, what broke along the way, and how they see AI reshaping customer experience on a global scale.
