The first useful clue in Anthony Goldbloom's career arrived while he was temporarily doing something else. In 2008, the young econometrician took three months away from Australia's Treasury for a reporting internship at The Economist in London. He had been building economic models in government. Now he was calling chief information officers and asking how companies planned to use their swelling stores of data.
The answers exposed a market that knew it had a problem but could not confidently buy a solution. One vendor might sell an expensive support-vector-machine package, another a neural-network package, and the buyer had no clean way to know which would work best. Goldbloom returned to Australia, he later recalled, "totally obsessed by the idea of big data." His proposed fix borrowed the logic of an architectural competition: give everyone the same brief, then reward the approach that performs best.
That thought became Kaggle. The premise was unusually crisp. Organizations posted data and a measurable problem. Data scientists submitted models. A public leaderboard kept score. Credentials receded; performance sat in plain view. For a field still arguing about its own name, Kaggle offered both a clubhouse and a proving ground.
A leaderboard for ability
Goldbloom's founding insight was partly technical and partly social. Companies needed better predictive models, but talented practitioners also needed a portable way to demonstrate what they could do. A competition connected the two. Someone far from a familiar hiring hub could beat an established team on the metric that mattered. The work became the résumé.
The format drew problems from industry, science and government. It also helped turn "data scientist" from an exotic label into a recognizable professional identity. By March 2017, when Kaggle joined Google Cloud, the platform had more than 800,000 data experts. Google framed the deal around lowering barriers to artificial intelligence and giving practitioners access to cloud tools and large public datasets.
Success did not arrive with founder mythology neatly attached. Early on, Goldbloom spoke candidly about his impatience with bureaucracy and his own urge to grab every task, even while recognizing that the habit would not scale. He also described himself as broadly capable rather than singularly brilliant at one function. It was an operator's admission: knowing every part of a machine can help you assemble it, then tempt you to keep touching every gear.
“I felt like a jack in the box.”Goldbloom on the restlessness that pushed him from government work toward founding Kaggle
There were lighter tells. He told one interviewer that his kitesurfing ability seemed to decline at the rate of Kaggle's success. After a client chief executive expressed surprise that his company had hired "a kid," Goldbloom dressed more formally for later meetings. Neither adjustment sounds like a grand strategic pivot. Together they capture the awkward education of a founder selling a sophisticated idea before he looked like the expected seller.
Recognition followed: Forbes included him in its technology 30 Under 30 lists in 2011 and 2012. MIT Technology Review named him an Innovator Under 35 in 2013, the same year the University of Melbourne gave him an Alumni of Distinction Rising Star Award. His 2016 TED talk compressed his view of automation into one distinction. Machines improve quickly at frequent, high-volume tasks. Humans retain an advantage in novel situations, where experience must be recombined rather than repeated.
He closed that talk by addressing his infant niece: "Let every day bring you a new challenge. If it does, then you will stay ahead of the machines." It was career advice disguised as a forecast, and it fits his own path. The common element is not allegiance to a single model. It is the appetite for the next unstructured problem.
The quiet interval
Goldbloom stayed with Kaggle after the Google acquisition and left in 2022. He had also joined AIX Ventures as an investment partner in 2021, working with a fund built around active AI practitioners. Then he reunited with Ben Hamner, Kaggle's former chief technology officer. The pair began exploring how to build high-quality structured data from noisy public inputs. For much of the market, their new company remained barely visible.
That interval matters. Sumble was not announced as an idea and assembled in public. Goldbloom and Hamner worked with a small founding engineering team, developed the data system and launched the product in April 2024. The company emerged from stealth in October 2025 only after it could point to tens of thousands of users, 19 enterprise customers and a 550 percent year-over-year increase in revenue.
Sumble's bet is that sales teams do not suffer from a shortage of records. They suffer from missing relationships. A conventional database can say that a 70,000-person company uses a tool. A useful answer says which team uses it, which project created the need, how recently the activity appeared and who owns the decision. Sumble draws from public material such as job postings, professional profiles, corporate pages and regulatory filings, then organizes the clues in a knowledge graph supported by large language models.
The practical difference shows up in Goldbloom's writing about account scoring. A mysterious 82 in a customer database does little for a salesperson who cannot see where it came from. Company size can dominate a crude score even when size says nothing about buying intent. Goldbloom argues for observable attributes, calibration against won deals, daily reranking and a view into the people, teams and projects underneath the number. His cleanest line is also a product specification: "A good score points the rep at a first move, not a grade."
A second act with old connective tissue
The $30 million Series A carried some of Kaggle's institutional memory. Canaan partner Rich Boyle had been a Kaggle board observer. Bloomberg Beta and Zetta had also backed Kaggle. Goldbloom, because of his role at AIX Ventures, stepped out of the room when the fund considered Sumble. That small governance detail is more revealing than a glossy reunion story. Networks can repeat; responsibilities still need clear edges.
The product's distribution offers another echo. Kaggle grew because useful work attracted other useful people. Sumble, Goldbloom says, often spreads inside a customer from one user to a Slack channel, then a team, an office and eventually the broader company. In one version of that path, monthly active users rise from one to 500 over six months. The unit of growth is a colleague passing along something worth inspecting.
“The more data that we add into the knowledge graph, the richer the corpus will be.”Goldbloom on the accumulation he believes can make Sumble defensible
There is an obvious challenge. Sales intelligence is crowded, and public data is available to everyone. Goldbloom's response is structural. A list of facts is easy to imitate. A graph that continuously connects facts across millions of organizations, departments, projects and people becomes more useful as its relationships deepen. He also expects those relationships to serve machines directly, letting a user ask a language model about a company's technology stack while grounding the answer in Sumble's data.
It is possible to read this as a pivot from data science to sales software. The more interesting reading is continuity. Treasury forecasting asked which observations could support a view of the future. Kaggle asked which model survived contact with a common metric. Sumble asks which company signals deserve attention and whether the user can trace the reasoning. Each business turns on a decision that improves when the evidence is organized.
Goldbloom's latest public work has become increasingly operational: guides for sales development representatives, account executives and teams building account scores; notes on territory planning, lead routing and finding untouched accounts; hiring posts for the next layer of Sumble's commercial organization. The econometrician's abstractions keep ending up as somebody's ordered list for Monday morning.
What the score is for
A leaderboard can look like the destination. Goldbloom's career suggests it is an interface. Kaggle's ranking made an invisible capability legible, but the purpose was to discover stronger models and give talented people a way in. Sumble's ranking makes a territory legible, but the purpose is to help someone decide where to spend the next hour and what to say when contact begins.
That distinction is useful well beyond either company. Scores decay when they become verdicts. They stay alive when they expose the factors beneath them, accept new evidence and guide a specific action. The design principle is modest: do not ask a user to admire the intelligence. Help the user move.
Goldbloom is now building with a co-founder who knows his first company from the inside, investors who remember the earlier journey and a market newly crowded with systems that can generate fluent answers. His wager is that fluency raises the value of grounded context. If every interface can produce a sentence, the advantage shifts toward knowing which facts belong in it.
The young reporter who interviewed CIOs in London noticed that buyers could not compare competing technical claims. The founder he became built a public mechanism for comparison. The experienced operator now wants the score to arrive with its reasons attached. Across eighteen years, the tools have changed from economic models to leaderboards to language models. The stubborn idea underneath them has not: evidence should earn its way into the decision.