Rosie
An AI research agent for the Product Innovation team — a deterministic orchestrator that routes across Internal Research, Synthesis, and Targeted queries and returns a grounded, four-lens read of MillerKnoll’s end-user research.
The story, in brief.
The Product Innovation team sits on a deep library of end-user research — and the hard part was never collecting it, but synthesizing it on demand. Rosie was built to do exactly that: a deterministic research orchestrator that returns a grounded, four-lens read of MillerKnoll’s own research. Its POC then revealed something bigger about the data underneath it.
How it’s built.
Rosie grew out of the Innovation Insights Experiment (LEANAI-38, now closed) and was replatformed onto Copilot Studio’s new agent experience. It’s an orchestrator with deterministic, button-based routing across three query types — Internal Research, Synthesis, and Targeted. The built-out Internal Research capability is grounded on the End User Research SharePoint library and returns a structured four-lens analysis with required citations and evidence-rigor levels.
What AI actually does here.
Deterministic orchestration
Rosie routes across Internal Research, Synthesis, and Targeted with explicit button selection instead of ambiguous intent detection — making the agent’s behavior predictable and testable rather than a black box.
Four-lens grounded synthesis
The Internal Research capability turns MillerKnoll’s own research library into a structured read across UX, Customer Insights, Business Strategy, and Knowledge Gaps — with evidence-rigor levels and required citations back to source.
A POC that proved the data
The highest-value output wasn’t the agent — it was the discovery that the research corpus was ready for a knowledge graph, which redirected the work to the Data Science team. The POC de-risked a much larger initiative.