Production questions worth a real study.

Open collaboration with researchers working on conversational agent evaluation, escalation and handoff behaviour, and small-business AI adoption. We have production traffic and unsolved questions; academic groups have method. The trade is access to anonymised data in exchange for findings we can publish.

Open questions we would fund a study on

• What actually predicts a caller abandoning an AI conversation.
• Whether disclosure of AI at greeting changes booking rate, and by how much.
• How escalation thresholds should differ by vertical risk profile.
• Whether eval-suite coverage predicts production incident rate.

What we can provide

Anonymised transcripts under a data-use agreement, latency and escalation telemetry, and the ability to run a controlled variation across production traffic with customer consent. The last one is rare outside large platforms and is the most valuable thing on offer here.

The terms

Findings are publishable regardless of whether they flatter us, which is stated up front so nobody has to negotiate it later. Customer identity never leaves our systems. Where a result would be commercially inconvenient, it still gets published — an agreement with a veto clause is not research.