DoorDash Deploys Observe.AI on AWS Across 19,000 Agents

What happened

DoorDash has partnered with Observe.AI, running on Amazon Web Services, to deploy real-time AI guidance and automated quality assurance across its contact-centre operation, which spans roughly 19,000 agents. The rollout gives agents live, in-call prompts and shifts quality monitoring from manual, sample-based reviews to automated analysis of interactions at scale.

According to the announcement, the aim is to move DoorDash's customer support model from a reactive, ticket-by-ticket posture towards one that is proactive and informed by continuous data from live conversations. The partnership pairs Observe.AI's contact-centre AI platform with AWS infrastructure to support the volume and speed required across DoorDash's agent base.

Why it matters

Contact centres of this size have historically relied on spot-checking a small fraction of calls to gauge quality, leaving most agent-customer interactions unreviewed and most coaching opportunities missed. Embedding AI directly into the live call — rather than only in post-hoc review — changes the intervention point: agents can be nudged towards better phrasing, empathy or resolution paths while the customer is still on the line, not after the outcome is already fixed.

For CX and behavioral-economics practitioners, this is a meaningful shift from measuring service to shaping it in real time. Automated QA at full coverage also changes what "quality" means operationally — it stops being a sampled estimate and becomes a near-complete signal, which has implications for how performance targets, coaching cadences and agent incentives are designed.

By the numbers

• 19,000 contact-centre agents are covered by the new AI deployment, according to the companies.

The Renascence take

The headline is the technology stack, but the more interesting story is behavioral: what happens to agent decision-making when guidance arrives mid-conversation rather than in a post-call debrief.

Most organisations treat AI-assisted QA as a monitoring upgrade — more coverage, same logic. The real opportunity is upstream: real-time prompts change the choice architecture agents operate under, nudging behaviour at the moment it matters rather than correcting it after the fact. Operators adopting this model should watch for over-reliance on prompts eroding agent judgement, and should design coaching and incentive structures around full-coverage data rather than simply digitising the old sampling approach.