Probative LabsYour virtual agent says “resolved”. Was it?
See how many “resolved” conversations came back as a ticket within a week. Free, and nothing is uploaded.
“Resolved” includes the people who just stopped answering.
ServiceNow's own guides describe self-solve analytics that mark a conversation resolved when at least one AI asset ran, and a resolution rate that counts no response as positive. A chat the person walked away from closes as "Closed Complete, No User Response".
Nothing in the dashboard checks whether that person opened an incident an hour later. The free check does, with your own two exports.
What the free check shows
- How many “resolved” conversations came back as a ticket within 7 days
- The same rate after your agents' own resolutions, for comparison
- Each close reason on its own: Topic Complete, No User Response and the rest
- What people came back about: where the bot is failing
- A one-page summary to save as a PDF and forward
Proof your finance team will believe, good or bad.
This is not a gotcha. It is the evidence you need either way.
If it holds up
Independent evidence for the next budget request or expansion, from a method finance and the vendor can both rerun.
If it doesn't
A fix list: the topics people came back about and the close reasons that fail most, so the next month looks better.
At renewal
Now Assist is billed in assists from a pool you renew. Go into the renewal knowing what held and what each resolution that held cost.
You don't have to take my word for it.
Inspect the math
The method is open source under Apache 2.0. Your own analyst can read it, rerun it and check every number.
Tested on real data
On a real ServiceNow log with failures planted in it, the audit found them, and found none when none were planted.
Independent
Probative Labs sells no AI and takes no money from AI vendors. Vendors cannot pay for, preview or suppress a result.
Team-level results only, nothing about individual agents. Beyond the free check: the full audit with issue-type matching, checks for every billing model, a regression watch and vendor bake-offs. The method.
Vivek Reddy Gaddam
AI vendors report their own results, and the people who pay for the AI have no independent way to check them. I built Probative Labs to give them one: an open method anyone can rerun, and a check that runs on their own data. I write the engine and run every check myself.
Check your virtual agent this week.
Two exports, one browser tab, nothing uploaded. If you'd like a second pair of eyes, share your screen and I'll walk through it with you.
Design partnerships are free: you get the check and a readout; I ask for feedback, an anonymized case study and a reference call if it helped.