Preferred Agent - Evaluation Method Order Issue

Hey all,

Seeing some weirdness with Preferred Agent evaluation in a recent flow update - impacting SLA on a high-volume queue. We rolled back, but need to understand what triggered this. Specifically, it looks like the evaluation method order is being ignored.

The flow uses Preferred Agent with two evaluation methods: first, a data action checking agent skill proficiency, then a simple queue membership check. Intention is to prioritize agents with the skill, then any available agent in the queue. However, A/B tests show a 7% SLA drop when the skill check fails - it’s immediately falling back to queue membership, effectively ignoring the skill-based evaluation entirely.

Console logs show the data action returning successfully (200 OK) for some agents, but the routing logic doesn’t seem to wait for that result before proceeding to the queue membership check. It’s as if the evaluation methods are being processed in parallel, not sequentially.

SDK version is 2.5.1. This is similar to post #6392 but that thread focused on the data action timeout - our DA is responding quickly. Anyone else run into evaluation order issues with Preferred Agent recently? Feels like a regression.

Right, so this is a bit fiddly. 1. The order should be respected, but the Data Action connector has always been a little…enthusiastic about processing results. 2. It’ll happily try to evaluate even if the first method fails, potentially bypassing your queue membership check.

Could you confirm you’re not seeing any errors in the Data Action execution logs - specifically, are the skill checks consistently succeeding or failing? That’s the first place to look, because if they’re failing, it’ll move straight to the queue membership.

edit: thinking about it, you might be getting a race condition - the queue membership check is fast, the skill check isn’t.