Enterprise Search AI
BI tools finally reach the
90% of data they couldn't see.
Until now, BI agents could only answer from the numbers sitting in your data warehouse. ThoughtSpot's Spotter 3 changes that design, letting a single question pull in the chatter, tickets, and deal notes buried in Slack and Salesforce alongside the warehouse tables.
What Changed
What it means to become
an "MCP host"
This isn't just another connector — Spotter itself now reaches out to call other tools.
Per ThoughtSpot's official blog, Spotter 3 now runs as an MCP (Model Context Protocol) host, pulling unstructured data from tools like Slack, Salesforce, Jira, and SharePoint into the same conversation as structured warehouse data. Six months ago it could only answer from warehouse numbers; anything from internal chat had to be tracked down by hand and cross-referenced manually.
The premise behind it, per diginomica: roughly 90% of enterprise data is unstructured, and most existing BI and AI tools simply don't look there. The reason a number moved is usually buried in a Slack thread or support ticket, not in the warehouse itself.
Why It Matters
Finally closing the gap
between "analysis" and "context"
Traditional BI dashboards show you what happened but rarely why. When revenue suddenly drops, tracing the reason has meant manually digging through Slack debates and CRM notes — eating up a large share of an analyst's time. Spotter 3 is designed to hand that "numbers-to-context" manual bridge over to the agent itself.
The MCP-host architecture matters here too. With Spotter itself acting as the caller of outside tools, the set of connected sources can grow as more tools adopt MCP, without necessarily waiting on ThoughtSpot to build each integration one by one.
Who Benefits
Who it helps, and how
BI/data teams (engineers)
Less manual digging through Slack and tickets to explain a number. Worth auditing exactly what access scope gets handed to Spotter over MCP before rollout.
Business leads and PMs
You can verify "why this number" on the spot during weekly reviews. During the phased rollout through December, check which capabilities apply to your team first to avoid wasted setup.
Existing ThoughtSpot users
No need to rebuild dashboard assets — unstructured-data search is added incrementally on top of your existing analytics setup.
The number lived in the warehouse.
The reason lived in the chat.
Recommended Actions
What to line up now
Design access scope first
Decide the policy on which Slack channels or Salesforce objects Spotter is allowed to read before company-wide rollout — it's much harder to walk back afterward.
Pilot with a small team
Since capabilities differ across the September-December rollout, pilot with a low-risk team first and check answer accuracy before deciding on a wider deployment.
Map overlap with existing tools
If you already run Slack search or an internal knowledge-base AI, sort out ahead of time where their scope overlaps with Spotter's unstructured-data search to avoid confusion.
Counterpoint
Cross-source search is also cross-source leakage risk
If DMs or private Slack channels end up in scope, one access-control mistake means information that was never meant to be visible can surface inside a BI dashboard. The MCP-host pattern is convenient, but an agent that reaches across tools also multiplies the surface area for permission mistakes. It's also worth noting the full feature set doesn't land until December — September ships only part of it — and several competing BI and agent vendors are chasing the same "structured plus unstructured" pitch around the same time, so how long this differentiation actually lasts is an open question.