Salesforce's Internal Slackbot Gave Three Different Answers to One Question. None of Them Was Wrong on Its Face.

Asked about Q1 FY27 marketing qualified leads, Salesforce’s own internal Slackbot returned two different figures — 50,000 and 75,000 — across separate queries, while the authoritative Leads dashboard showed the correct number: 100,000. Salesforce data scientists Julia Kosheleva-Coats and Johnson Ton use that internal failure to make a broader point in a new Salesforce Blog post: AI agents will confidently produce an answer regardless of whether it’s correct, because without grounding context they have no way to tell a right answer from a wrong one. The root cause here was mundane and exactly the kind of thing that scales badly — the bot queried a raw Lead object directly instead of the org’s certified MQL definition, leaning on a misaligned field instead. Their fix has three parts: “AI Skills” that tell agents which data sources and definitions to use, a semantic-model connector they call “Tableau MCP,” and certified datasets established as the one source of truth agents must query against.

It’s a data-accuracy failure mode, but it’s also a governance failure with no clear owner — which is the exact gap Aviso is naming from the RevOps side. Aviso’s blog proposes a formal “RevOps Agent Owner” role, splitting accountability between a Business Owner (commercial outcomes) and a Technical Owner (infrastructure and compliance), and argues that without that split, organizations accumulate what it calls a governance vacuum — nobody formally responsible when an agent acts on stale or wrong context. Salesforce’s Slackbot incident is what that vacuum looks like from the inside: a plausible-sounding number, no single owner positioned to catch it before a real forecast conversation used it.

For sales orgs deploying internal agents against CRM and BI data, the lesson isn’t to trust the tool less — it’s to name, in writing, who owns the certified data source it’s allowed to query, before the agent ships an answer that sounds exactly as confident when it’s wrong as when it’s right.