loading
loading
by Jason Lemkin | AI App of Week, Blog Posts The Next Generation App To Learn What’s Really Happening In Your Target Accounts Sumble is a knowledge graph of what’s actually happening inside your target accounts. Not “does this company use Snowflake,” but which team uses it, who runs that team, how many people report to them, what they posted a job for 21 days ago, and what they’ve been quietly migrating off of since January. Built by Anthony Goldbloom and Ben Hamner, the founders of Kaggle (acquired by Google in 2017). $38.5M raised from Coatue and Canaan, with Marc Benioff and Nat Friedman on the cap table. Databricks, Snowflake, Figma, Vercel, Wiz, Elastic, dbt Labs, Snyk, and Datadog reps are using it. And the entry price is $99/month, in a category where the incumbent has historically wanted $30K and a year-long contract before you see a single record. Contacts Got Commoditized. What Didn’t Is What’s Happening Inside of THem Every GTM team in B2B has now bought the same three things: a contact database, a sequencer, and something with “AI” in the name that writes the email. The result is that everyone has the same contacts, everyone has the same automation, and everyone is sending the same personalized-by-LLM message to the same VP of Engineering on the same Tuesday morning. Contacts got commoditized. What didn’t get commoditized is knowing what’s actually going on inside the account. That’s the wedge. Sumble crawls public sources across job postings, company sites, social, and filings, then uses LLMs to assemble it into a structured knowledge graph of each account: teams, reporting lines, technologies mapped to the specific teams using them, hiring activity, and live projects like a cloud migration or a GenAI rollout. Most tools get you to “maybe Wells Fargo uses Grafana.” Sumble gets you to: the Platform Engineering team at Wells Fargo uses Grafana, that team has 72 people, it’s led by a named engineering leader in Charlotte, they put out a job post 21 days ago that mentions Grafana, and since January they’ve been expanding Grafana while pulling back on OpenTelemetry. One of those is a data point. The other one is an opening line. The Problem They’re Actually Solving Here’s the honest state of outbound in 2026. Your rep pulls a list. The list is firmographics plus a title filter. The title filter is wrong roughly half the time, because org charts in real companies don’t map cleanly to LinkedIn titles. Then an AI writes a personalized opener based on a press release the prospect didn’t read either. Three specific failures show up over and over in pre-agentic data tools: - You target by title, so you miss the actual owner. In one customer workflow, Sumble surfaced a contact whose LinkedIn title was “Implementation Manager” but whose actual job description showed she ran the entire call center. No title-based search on earth was finding her. That’s not an edge case. That’s most of enterprise. - You know the company uses a tool, but not who. Legacy technographics scrape the website and tell you a logo appeared somewhere. They can’t tell you the tool sits with a 40-person data platform team in Austin whose lead just posted three roles mentioning it. - You have no idea when the budget is live. The single best time to sell into a project is when the company is staffing it. Job postings are the most underrated buying signal in B2B, because they’re published, timestamped, and describe the work in the company’s own words. Sumble reads them as intent data rather than as recruiting noise. Fix those three and outbound stops being volume math. What Actually Makes Sumble Different Tech mapped to teams, not to companies. This is the core unlock. Knowing an account uses a technology is a filter. Knowing which department pays for it, how big that team is, and who it reports to is a plan for how to break in. Real org structure and reporting lines. You see who reports to whom and who actually controls the budget. Most tools either don’t attempt this or infer it badly from titles. Active initiatives, with timing. Migrations, GenAI rollouts, platform rebuilds, surfaced while they’re happening rather than after the vendor got picked. The whole product is oriented around “why now,” which is the question 95% of outbound can’t answer. Net-new account discovery. Once your best customers are described in terms of team shape and stack rather than industry and headcount, you can go find the companies that look like them. This is the TAM-expansion use case, and it’s the one RevOps teams get excited about. An MCP server, and that’s a bigger deal than it sounds. Sumble ships an MCP server that drops its data directly into Claude, Cursor, or ChatGPT. So instead of a rep clicking through a UI, a GTM engineer can ask in plain English: “find Boston companies growing 20% YoY using Databricks and Looker with a data engineering team under four people,” or “which teams in my accounts are running a data migration right now?” Then have the model pull the decision makers and draft the outreach in the same pass. One team ran 1,800 accounts through it to find call center usage, something no existing tool reliably surfaced, and came back with 61 qualified companies plus the actual owners inside them. That’s the workflow, end to end, in a chat window. API and warehouse delivery. Robust API, and it can write straight into Salesforce, Snowflake, or Databricks. Which matters, because the moment this data lives in your warehouse it stops being a research tool and starts being scoring logic. The Founder Story Anthony Goldbloom and Ben Hamner built Kaggle, the data science competition platform that grew into a community of millions of practitioners and got acquired by Google in 2017. They started Sumble in 2022 and launched the product in April 2024. What I like about the origin: they hit this problem from the data side, not the sales side. At Kaggle they kept running into how brutally hard it was to assemble large, clean datasets about companies. So the second company is essentially the answer to a question the first company kept asking. Founders who pick their second company out of a problem that annoyed them for a decade tend to have unusual staying power on the hard parts. And the hard part here is real. Turning messy public web data into a structured knowledge graph that a sales rep will trust is a data engineering problem, not a prompt engineering problem. That’s exactly the skill set these two have. The thing most AI GTM startups are missing right now is anyone who has actually shipped a large-scale data product. Sumble is two people who have. The investor list tracks the same story. Coatue led an $8.5M seed, Canaan led a $30M Series A, with AIX Ventures, Square Peg, Bloomberg Beta, and Zetta participating, plus angels Marc Benioff and Nat Friedman. Several of those investors backed Kaggle. Rich Boyle at Canaan was a Kaggle board observer. When the people who watched your last company from the inside write the biggest check on your next one, that tells you something a pitch deck can’t. The Numbers - $38.5M raised as of the October 2025 launch ($8.5M seed led by Coatue, $30M Series A led by Canaan Partners) - Marc Benioff and Nat Friedman among the angels - 550% revenue growth reported at launch - 19 enterprise customers signed since the April 2024 launch, with tens of thousands of total users - Customer list includes Databricks, Snowflake, Figma, Vercel, Wiz, Elastic, Snyk, dbt Labs, Baseten, Modal, Fireworks AI, Atlan, Omni, CodeRabbit, Prophecy, and Nooks - Founded 2022, launched April 2024, headquartered in San Francisco - Small team. Roughly two dozen people running an enterprise data product against ZoomInfo and Apollo Note the customer list carefully. It’s disproportionately the companies selling technical products into technical buyers: data infrastructure, dev tools, security, AI infra. That’s not an accident. That’s the ICP working exactly as designed, and it’s the segment where “which team uses what” is the entire sales motion. Pricing - Free: $0. Tech, job function, and firmographic filters, browse organizations, 500 credits a month, 7 signals a week, first page of results, Slack integration. Sign up with a work email and you get 30 days of Pro. - Pro: $99/month. Project-based search, people search inside orgs, reporting hierarchy, full text search, personalized outreach drafts, the MCP server, 9,900 credits a month, up to 20 signals a day, 10 pages of results. - Enterprise: Custom. Onboarding, RevOps data integrations (Salesforce, Snowflake, Databricks), workflow integrations, company-wide signal definitions, dedicated support. The pricing is the strategy. Sales intelligence has been an enterprise-contract category for fifteen years: annual commit, seat minimums, procurement, no way to see the data before you sign. Sumble put a real free tier and a $99 self-serve plan in front of the same dataset. That does two things at once. It lets a single AE or GTM engineer prove value on their own accounts in an afternoon with no approval, which is how bottoms-up lands in 2026. And it makes the incumbent’s pricing feel like what it is. Hard to defend a $30K floor when a rep on your own team is already getting better answers for a hundred bucks on their personal card. Who Should Use It Yes, immediately, if you sell technical products into technical buyers, your deals hinge on which internal team owns a system, or your reps burn hours per week on account research before a call. Yes, if you’re building GTM agents in-house. This is the data foundation problem. LLM-powered outreach with bad context just generates fluent noise faster. Probably not if you sell SMB at high volume where the winning motion is speed and coverage, not depth per account. The whole product is built around accounts worth researching. What This Means for B2B Founders The broader lesson here goes well past one tool. Everyone spent 2025 buying the agent layer. The models write the emails, run the sequences, and handle the follow-up. That part is now close to free and it’s approaching parity across vendors. When execution gets commoditized, the advantage moves to whoever has better inputs. Which is why the sharpest thing in Sumble’s positioning is Goldbloom’s line: reps don’t need more contacts, they need more context. Every AI GTM stack is about to converge on the same capabilities. The differentiation will be the data foundation underneath it, and almost nobody has built one worth having. There’s a second lesson for anyone looking at a category everyone calls finished. Sales intelligence is as crowded as software gets. ZoomInfo, Apollo, Clay, the technographic vendors, plus a hundred AI SDRs. A two-founder team walked in anyway, because they saw the whole category was answering the wrong question. Crowded doesn’t mean solved. It usually means everybody is solving the same 80% and nobody wants the hard part. Start on the free tier, run five of your real target accounts through it, and see whether it tells you something your current stack doesn’t. That test takes twenty minutes and you’ll know. Try it: sumble.com SaaStr AI App of the Week is a weekly series highlighting the most interesting AI tools actually being used in production by B2B companies. Not demos. Not pilots. Actually deployed, actually working, actually generating ROI. Want to be featured as App of the Week? Reach out to our team.
document fetched 2026-09-14 02:48 · canonical https://saastr.com/saastr-ai-app-of-the-week-sumble-the-kaggle-founders-rebuilt-sales-intelligence-around-context-not-contacts