The Return of a Familiar Builder

On October 22, 2025, Anthony Goldbloom and Ben Hamner brought Sumble out of stealth and announced $38.5 million in funding: an $8.5 million seed round led by Coatue and a $30 million Series A led by Canaan Partners.

That launch mattered for more than the funding number. Goldbloom is not just another founder entering the AI application wave. He built Kaggle into the most important community platform in data science, sold it to Google in 2017, and then spent years watching machine learning move from specialist tooling into mainstream software. Sumble is his attempt to capture what happens next: the point where large language models become useful not because they can talk, but because they can continuously synthesize scattered commercial signals into operational judgment.

According to people familiar with the company, Sumble had already signed 17 enterprise customers by the time it emerged from stealth, including Snowflake, Figma, Wiz, Vercel, and Elastic. The company says its knowledge graph covers about 2.6 million companies and updates from sources such as hiring data, company sites, social feeds, and regulatory filings. People close to the business described revenue growth of roughly 550% year over year, though that number should be read in the context of a very small base.

The market Goldbloom is targeting is not tiny. Grand View Research estimated the global sales intelligence market at roughly $4 billion in 2024 and projected it could reach $12 billion by 2030. But the bigger point is not market size. Goldbloom is returning to the same structural idea that made Kaggle work: take a noisy, fragmented technical landscape and build an organizing layer that turns it into practical advantage.

Before Sumble, There Was Kaggle

Goldbloom was born in Melbourne in 1987 and trained first in economics and econometrics. He worked in macroeconomic modeling for the Australian Treasury, then spent time at The Economist, where an assignment on the emerging field of big data changed his trajectory.

What he noticed early was that machine learning talent was abundant but disconnected from real business problems. Kaggle, founded in 2010, was his answer. The first competition, sponsored by Merck, asked participants to predict HIV progression. The winning external model reportedly outperformed Merck’s internal approach by a wide margin. That result gave Kaggle its founding argument: if you open a problem to a global pool of talent, the best answer often comes from outside the building.

Kaggle expanded quickly. Jeremy Howard joined early. Venture investors including Index Ventures, SV Angel, Khosla Ventures, Yuri Milner, and others backed the company. Over time Kaggle evolved from a contest platform into a full working environment for data scientists, with hosted datasets, shared code, notebooks, and reputational rankings that became meaningful hiring signals.

By the time Google acquired Kaggle in 2017, the platform had become something like the GitHub of applied data science. By 2023 it had grown to more than 15 million users across nearly every major market. Kaggle had become a training ground, a hiring filter, a social graph, and a credentialing system for machine learning talent.

That history matters because Sumble is not a random category jump. It is another attempt to build infrastructure around an information bottleneck.

The Limits of the Google Years

Google gave Kaggle scale, infrastructure, and distribution. It also exposed the limits of community products inside large platforms.

Former employees have long described Kaggle as strategically important but financially awkward inside Google Cloud. It was a strong brand and a valuable community, but it did not fit neatly into enterprise SaaS revenue logic. Kaggle was where people learned, experimented, and built reputation. It was not, for most users, the software they directly bought to run production workflows.

That tension matters because Goldbloom seems to have learned something important from it. A large community can create influence, but influence alone does not guarantee a durable software business. If you want to build a company with clearer monetization, you need to attach your product to a recurring operational decision, not just a learning or prestige loop.

Goldbloom and Hamner left Kaggle in 2022. Publicly, the move was described as a planned transition. Privately, people familiar with the story have long framed it as a divergence over what Kaggle should become inside Google. Whatever the exact mix of reasons, the outcome was clear: both founders left with a rare combination of credibility, distribution relationships, and a deep understanding of how machine learning practitioners actually work.

That combination gave Goldbloom an unusual starting point for a second company. He did not need to prove he could build community or product. He needed to decide which workflow AI would transform next.

Why Sales Intelligence Became the Next Target

After leaving Kaggle, Goldbloom spent time as an investor at AIX Ventures and looked across dozens of AI startups. That perspective appears to have sharpened the next thesis.

The old generation of sales tools was built around static data: contact records, org charts, email sequences, and CRM fields. The new opportunity is dynamic context. Which account is hiring aggressively? Which company is migrating infrastructure? Which buyer just started talking publicly about a strategic priority? Which regulatory filing suggests a budget shift? Which leadership hire changes the internal power map?

That is the hole Sumble is trying to fill. Its core argument is that sales teams do not just need more names or more email addresses. They need a better answer to a harder question: why should I reach out to this company right now?

That question became more important as B2B buying behavior changed. Buyers do more research before speaking to a sales rep. Generic outbound messages have lower yield. Timing, relevance, and context matter more than they did in the era when merely having the right contact database created advantage.

Goldbloom’s bet is that LLMs and knowledge graphs finally make this synthesis layer commercially viable. If a company is posting data engineering roles, publishing infrastructure migration content, discussing platform modernization, and increasing technology spending, an AI system can turn those scattered clues into a coherent commercial story. That is not just research. It is a go-to-market trigger.

The Product Thesis: A Knowledge Graph for “Why Now?”

Sumble says its graph spans around 2.6 million companies and pulls from a wide range of public and semi-public signals: company sites, job posts, social content, filings, news, technical blogs, and other external data sources.

The important distinction is that Sumble is not selling raw aggregation. It is selling interpretation. Traditional sales intelligence vendors have usually been strongest at the “who” layer: who works there, who reports to whom, how to contact them. Sumble is trying to own the “what changed” and “why it matters” layer.

That puts it in a different position relative to incumbents:

ZoomInfo remains strong on contact data and buyer coverage.
Apollo is strong on workflow and outbound execution.
Gong is strong on call intelligence and deal insight once the conversation has already started.
Outreach is strong on orchestration and sales engagement.

Sumble is trying to sit one step earlier in the motion. It wants to become the system that tells a rep why an account deserves attention before the rep opens the cadence tool.

If that works, it is strategically important. The best sales software categories tend to control one of three moments: whom to target, how to engage, or how to manage the deal. Sumble is aiming at the first moment, but with more reasoning built in than previous data providers offered.

The problem, of course, is that this is also where product quality becomes very hard to judge. A contact database can be audited for coverage and freshness. A sales workflow tool can be judged by usage and throughput. A system that claims to generate high-value commercial insight has to prove that its interpretation is correct often enough to change behavior.

The Early Signals Look Strong, but They Need Context

The company says 17 enterprise customers were onboard by the time it emerged from stealth. The customer list is notable because it includes software companies with sophisticated go-to-market organizations: Snowflake, Figma, Wiz, Vercel, and Elastic. These are not naive buyers.

People close to the company also say that about 30% of users convert to the Pro tier, well above typical SaaS conversion benchmarks in many self-serve products. Sumble’s reported pricing structure includes:

a free tier with limited searches,
a Pro plan priced around $149 per user per month,
and enterprise contracts with API access, dedicated support, and administrative features.

Average contract value is said to range from roughly $50,000 to $500,000 depending on seat count and usage profile. If those figures hold, Sumble is not building a lightweight prospecting tool. It is moving toward a higher-value intelligence layer sold to serious revenue teams.

The investor roster also matters. Coatue led the seed. Canaan led the A round. Bloomberg Beta and Zetta, both linked to the earlier Kaggle story, reportedly returned. Angel investors include Marc Benioff and Nat Friedman. That list does more than signal prestige. It shows that sophisticated investors see a plausible bridge between Goldbloom’s history in machine learning infrastructure and a new software layer for commercial teams.

But the same facts can be read another way. Enterprise design partners are not the same as a scaled market. Early traction inside technical software companies does not yet prove that Sumble’s product generalizes across manufacturing, healthcare, finance, or industrial sales environments where public signals may be weaker and workflows more fragmented.

What Investors Are Really Betting On

The funding round is easier to understand if you treat it as a founder bet first and a category bet second.

Coatue’s seed participation was notable because Coatue usually enters later. Canaan led the A round. Bloomberg Beta and Zetta reportedly returned after backing Kaggle. Angel investors included Marc Benioff and Nat Friedman. That roster suggests investors saw something more durable than a fashionable AI application. They saw a founder who had already built a large technical network, turned it into platform power, and then carried that credibility into a second market at exactly the moment when synthesis models became usable.

The investor logic runs on three assumptions.

First, Goldbloom and Hamner know how to build around a technical community. Kaggle was not just software. It was a trust network. That matters because Sumble’s technical hiring advantage likely comes from the same reputation flywheel. People who spent years watching the founders build one of the most respected brands in machine learning are more willing to join an ambitious second act.

Second, they understand how to translate specialist infrastructure into repeatable workflow value. Kaggle made machine learning talent legible to organizations. Sumble is trying to make commercial change legible to sales teams. Different customer, same pattern.

Third, investors appear to believe the category could become strategically important before incumbents adapt. If the “why now?” layer becomes a standard part of go-to-market tooling, the prize is not just another feature. It is influence over which accounts get attention first, which messages get written, and which revenue teams become structurally faster.

This is why the investor list matters. The round was not priced like an experiment in lightweight prospecting software. It was priced like an attempt to define a new system layer in revenue operations.

The Kaggle Network Is Now a Recruiting Advantage

One underappreciated advantage in Sumble’s early buildout is talent access.

Kaggle spent more than a decade accumulating not just users, but a reputation hierarchy. Grandmasters, competition winners, highly ranked practitioners, and open-source contributors all learned to treat the platform as a serious merit system. That kind of network does not disappear when a founder leaves.

People close to the company have long argued that this network is one reason Sumble can stay relatively lean while still assembling strong technical talent. Instead of trying to outspend larger AI startups on raw compensation, the company can recruit through founder credibility and technical challenge. In practical terms, that means it can staff an engineering team with people who already understand model quality, evaluation, data pipelines, and competitive iteration cycles.

That matters because Sumble’s product does not live or die on a single model choice. It lives or dies on a chain of work: crawling, filtering, graph construction, entity resolution, signal ranking, LLM synthesis, delivery timing, and user trust. Weakness in any layer reduces value.

The company appears to be borrowing another lesson from Kaggle too: operate with a relatively small team, release frequently, and tie product decisions to observable user behavior rather than abstract roadmap ambition. That style works well when the hardest problem is not lack of ideas, but deciding which improvements meaningfully change user outcomes.

If Sumble scales, the Kaggle legacy may end up mattering less as biography and more as org design. Goldbloom already ran one company where technical excellence and network reputation reinforced each other. Investors are betting he can do it again in a more commercial category.

Why the Timing Is Better Than It Looks

The Sumble thesis depends on several market conditions converging at once.

First, LLM quality has reached a point where summarizing and connecting weak signals is commercially usable, at least in narrow domains. This kind of synthesis would have been far harder to ship in 2019.

Second, modern B2B buyers reveal more about themselves in public than previous generations did. Engineering blogs, hiring pages, social channels, compliance filings, launch posts, and executive interviews all create clues about strategic intent.

Third, sales teams are under pressure. Quota attainment has been volatile, outbound effectiveness has declined, and generic automation has made inboxes noisier. That creates demand for tools that promise better signal quality rather than higher message volume.

Fourth, the center of gravity in enterprise AI has moved toward application-layer products that combine foundation models with workflow, proprietary data, and domain-specific judgment. Sumble fits that pattern almost perfectly. It does not need to build a frontier model. It needs to build a better commercial inference layer on top of existing ones.

In that sense, Goldbloom’s move is very much of this cycle. Kaggle helped professionalize machine learning talent. Sumble tries to operationalize machine learning for a non-technical commercial function.

The Risks Are Real

For all the promise, Sumble faces a serious execution burden.

Data quality

The product is only as good as its signal extraction. Hiring data can be stale. Social posts can be misleading. Company blogs can overstate internal priority. Regulatory filings lag reality. If Sumble delivers noisy or wrong narratives too often, trust will break quickly.

Unit economics

This kind of product is not cheap to run. Continuous crawling, normalization, storage, graph construction, and LLM inference all create real cost. People close to the company have described gross margins around 60%, below the profile investors typically expect from top-tier software businesses. That does not make the model broken, but it means pricing power and product value have to stay high.

Competitive response

Incumbents will not ignore this category if it works. ZoomInfo, Apollo, Salesforce, HubSpot, and Microsoft all have reasons to add richer signal interpretation into their own stacks. If the “why now?” layer becomes clearly valuable, larger players will move toward it.

Privacy and regulatory pressure

The product relies on collecting and structuring external information. Even when data is public, the line between publicly visible and commercially reusable is increasingly contested. Privacy regulation and platform restrictions could raise both legal and operational costs.

Category education

Sumble is selling something slightly harder to explain than a contact database or a sequencing tool. It has to teach buyers not just what the product does, but how to operationalize its insights inside day-to-day sales behavior.

Geographic and domain expansion

The current customer profile makes strategic sense, but it also reveals the next challenge. Technical software companies produce unusually rich external signals: engineering blogs, hiring data, cloud migration language, public APIs, executive commentary, security updates, and developer footprints. That makes them favorable training ground for Sumble’s approach.

The question is what happens when the company pushes into sectors where the signal layer is thinner or messier. Manufacturing firms, healthcare providers, financial institutions, and professional-services organizations still emit useful clues, but not always in the same density or format. Expanding globally compounds the challenge. Crawling English-language company data is one thing. Building reliable signal coverage across Japanese, Portuguese, or multilingual regional markets is another.

If Sumble wants to become a real platform rather than a high-end wedge inside technology sales, it will eventually need to prove the model generalizes.

What Success Would Actually Look Like

There are three plausible end states.

One is the classic venture path: scale rapidly, define a new category around AI sales insight, and build toward a public company. That would require more than growth. It would require predictable retention, clear ROI proof, and enough product depth to avoid getting flattened into a feature inside a broader revenue platform.

Another is strategic acquisition. Salesforce, HubSpot, Microsoft, or another major commercial platform could decide that owning the intelligence layer is easier than building it from scratch. Marc Benioff’s presence as an investor does not prove anything, but it does make that scenario easier to imagine.

The third is narrower but still meaningful: Sumble becomes a profitable, specialized, high-end platform for sophisticated revenue teams without trying to be a universal system of record.

The milestones that matter most are straightforward:

proving repeatable value at larger customer counts,
expanding beyond technology-forward design partners,
showing that the graph and reasoning layer truly improve conversion or sales efficiency,
and defending enough proprietary product advantage that incumbents cannot neutralize the category immediately.

There is also a practical milestone that matters more than headline ARR: can Sumble become part of the daily operating motion of a revenue team rather than an occasional research destination? The difference is decisive. A tool used once a week for account prep is helpful. A tool that influences pipeline prioritization, outbound sequencing, deal review, and account planning becomes embedded. Embedded software survives budget scrutiny much better than “interesting” software.

That is the path to independence if the company wants one. It has to move from smart research assistant to workflow-shaping system.

The Deeper Pattern in Goldbloom’s Career

Goldbloom’s first company helped data scientists become more effective. His second company is trying to help salespeople benefit from machine learning without needing to understand the machinery behind it.

That arc mirrors a broader industry shift. Early machine learning infrastructure products served technical specialists. The new wave serves knowledge workers in domain workflows. The value no longer comes from exposing the model. It comes from hiding the complexity well enough that the user simply gets a better decision.

That is what makes Sumble more interesting than a conventional “former founder starts another AI company” story. Goldbloom is once again building around an information bottleneck. Kaggle organized machine learning talent. Sumble is trying to organize commercial signal.

Whether that becomes a category-defining business depends on the same question facing many AI application startups: can a product that looks magical in demo form become reliable enough, cheap enough, and differentiated enough to deserve a permanent budget line?

Goldbloom has earned the right to be taken seriously. He has not yet earned the outcome.

Source Notes

Grand View Research on the global sales intelligence market
Company launch and funding disclosures around Sumble’s October 2025 debut
Historical Kaggle investor, acquisition, and user growth records
Public category positioning from ZoomInfo, Apollo, Gong, and Outreach
Market and workflow interpretation based on the company, customer profile, and competitive set described above