Used by RevOps teams at
Three workflows it owns end-to-end — plus a fourth your CRO will ask for in week two.
We ran RevOps at Gong, Clari and Outreach for a combined 19 years. Every "AI" feature we shipped was great in a demo and impossible to defend in a board meeting — because none of them showed their work. Cobalt is the opposite. Every forecast change is a clickable list of the calls, emails and stage moves that caused it. If you can’t defend the number, it doesn’t make it into the dashboard.
Our forecast accuracy went from ±17% to ±4% in two quarters. The board now asks for the Cobalt number, not the CRM number. That is a sentence I never thought I’d say.
Every tier includes forecast, deal-risk, coaching notes, Salesforce + Gong ingest, SOC 2 and SSO.
For sales orgs up to 25 reps.
For sales orgs of 25–300 reps.
For 300+ rep orgs with custom data residency or VPC.
Two things. First, the unit of explanation is the deal, not the cohort — every forecast number is a clickable list of the actual calls, emails and stage moves that caused it. Second, we ship per-rep coaching notes as a first-class workflow, not a hidden tab. That’s where most of the customer ROI lands inside 90 days.
For a typical Salesforce + Gong + Slack stack: 12 minutes for connectors, 24 hours for the historical backfill, first usable forecast on day 2. We do this without a Cobalt SE on the call. Larger orgs (300+ reps, multiple CRM instances) get a dedicated 2-week implementation team.
No — ever. Your CRM, calls and Slack never train a shared model. We use stateless inference against frontier providers under DPAs, with EU and US ingest endpoints. Customer chooses residency at signup; data never leaves the chosen region.
Every claim Cobalt makes is grounded in named source artifacts — the call ID, the email message ID, the CRM field history row. If a claim has no source, it isn’t shown. We publish our retrieval-grounded accuracy numbers in the trust report (94.1% at GA).
Email support on Team (4hr response, business hours). Shared Slack channel on Scale, average first response 9 minutes. On Enterprise, a named CSM plus the founders’ personal line on Signal. Every founder reads every escalation.