DoorDash, Observe.AI, and AWS Partner to Scale Customer-Centric AI Across 19,000 Agents

What happened

DoorDash has partnered with Observe.AI and AWS to roll out real-time AI guidance and automated quality assurance across its customer support operation, covering roughly 19,000 contact-centre agents. The deployment is designed to give agents in-the-moment coaching prompts and to automate call and chat review, replacing manual sampling with continuous, systematic quality checks.

The initiative is framed by the companies as a shift away from reactive, after-the-fact support review toward a proactive model where agent performance and customer interactions are monitored and guided as they happen. AWS's infrastructure underpins the scale of the deployment, allowing Observe.AI's platform to process interactions across DoorDash's large, distributed agent workforce.

Why it matters

Contact-centre QA has traditionally relied on sampling a small fraction of interactions after the fact — a method that catches problems too late to help the customer on the call and gives agents inconsistent, delayed feedback. Real-time guidance changes the behavioural dynamic: agents receive prompts while the interaction is still live, which can reduce variance in how policies are applied and shorten the gap between a mistake and its correction.

For service-design teams, this is also a data story. Full-coverage automated QA turns every interaction into a source of pattern-level insight rather than a small, potentially unrepresentative sample — which matters for spotting systemic friction points, not just individual agent errors.

By the numbers

• 19,000 contact-centre agents are covered by the new AI-guidance and QA deployment.

The Renascence take

The headline number here is scale, but the more interesting question is what "real-time guidance" actually changes about how agents behave under pressure — and whether it is designed to support judgement or simply enforce compliance.

Most coverage of deployments like this focuses on efficiency and coverage, but the real behavioural test is subtler: does in-the-moment AI prompting build agent confidence and consistency, or does it create a new form of anxiety — agents second-guessing themselves because a system is watching every word? The operators who get this right will treat real-time guidance as a coaching layer that agents trust, not a surveillance layer they route around. Full QA coverage is only valuable if the resulting insights are fed back into training and process redesign quickly — otherwise it becomes an expensive way to document problems that already existed.