Kaggle's Founders Just Undercut Sales Intelligence Pricing by 300x

Anthony Goldbloom and Ben Hamner, the founders who built and sold Kaggle to Google in 2017, are back with Sumble — a sales-intelligence platform that skips the contact database entirely in favor of knowledge graphs mapping an account’s actual organizational structure, reporting lines, technology stack, and active initiatives with timing data. The pitch, as SaaStr’s AI App of the Week profile frames it, is context over contacts: sellers get told what’s actually happening inside a target company, not just who to email.

The pricing is the part worth sitting with. Sumble’s Pro tier runs $99 a month against the $30,000-plus annual contracts typical of legacy sales-intelligence vendors — a roughly 300x gap that only makes sense if the underlying economics of account data have genuinely shifted. Since its April 2024 launch, Sumble reports 550% revenue growth and 19 enterprise customers, including technical, product-led companies like Databricks, Snowflake, Figma, Vercel, and Wiz — the kind of buyer that tends to be skeptical of legacy sales-tooling pricing on principle. The company has raised $38.5 million total: an $8.5 million seed from Coatue and a $30 million Series A from Canaan Partners. It also ships an MCP server, so the account data can be queried directly from Claude, Cursor, and ChatGPT rather than through a standalone dashboard.

Sumble’s bet — that context beats contacts — has company. Clay launched Account Research Agents in July, a feature that continuously synthesizes account intelligence from Gong transcripts, CRM records, and buying signals rather than researching one account at a time, with the key differentiator being persistence: standard research agents restart context on each run, Clay’s compound it over time into a running account history. Different architecture, same underlying wager as Sumble’s: as contact-level data becomes commoditized and cheap, the sales-intelligence vendors worth paying for are the ones selling context that compounds, not lists that go stale — and MCP-native distribution is fast becoming table stakes for making that context usable inside whatever AI tool a seller already has open.