Conversation
Highlights

How Dataiku Reached $350M ARR by Positioning Between Infrastructure and Business Users

When Florian Douetteau started Dataiku in 2013, talking about AI in enterprise contexts meant academic research and futuristic robots. Meanwhile, a new practice called “data science” was emerging in the Bay Area—companies like LinkedIn and Yahoo were building teams to optimize recommendation engines and advertising algorithms.

Florian saw the pattern forming and made a contrarian bet: this centralized approach would fail for most enterprises.

“My thought was that it was not the way to think about it, that for most enterprise, a lot of the context is actually in the business,” Florian explained in a recent episode of Unicorn Builders. His thesis: business operators with domain expertise should build AI solutions themselves, not wait for centralized data science teams.

The reasoning was specific. A junior data scientist lacks the context to optimize a bank’s customer onboarding process that touches 20 people and costs millions to run. They don’t understand railway safety protocols or pharmaceutical clinical data management workflows. “They don’t have the context of this real life matters,” Florian said.

This wasn’t philosophy—it became Dataiku’s entire go-to-market positioning.

The Platform-Not-Services Decision That Enabled Scale

When business users don’t understand how to apply complex technology, the default path is professional services. High margins, direct customer contact, proven revenue model.

Dataiku explicitly rejected this.

“We actually managed very well to avoid, I would say this consulting trap, even if it’s not necessarily a trap, but I think for us it would have been,” Florian explained. “And build the tool so that you can have people in the business that own it and start using it and find their way.”

Instead of implementation services, they built training programs, partner ecosystems, and self-service capabilities. The strategic bet: customer independence would drive faster adoption and better retention than dependency on Dataiku consultants.

Two years after launch, this decision paid off when they successfully sold to Fortune 500 companies in the US—a jump from their comfort zone that required a platform sophisticated enough for enterprises to deploy without hand-holding.

Targeting the “In-Between” Buyer Nobody Else Served

Dataiku’s buyers aren’t CTOs managing infrastructure budgets. They’re not business unit leaders focused purely on outcomes.

“Our buyers are typically people sitting most of the time in IT, but very much business focused,” Florian explained. “People in charge of AI analytics, data strategy, AI transformation strategy. They are kind of like the person in between.”

This persona faces a specific structural problem. On one side: enterprise IT infrastructure that “very candidly change every two or three years”—Hadoop to cloud platforms to Snowflake to Databricks, with new layers constantly emerging.

On the other side: business users who need to deliver projects that generate measurable ROI.

The in-between buyer must translate between these worlds—making infrastructure accessible without sacrificing governance, enabling business teams without creating shadow IT.

Most vendors optimize for one side or the other. Infrastructure vendors build for technical sophistication. Application vendors abstract away complexity. Nobody was building translation infrastructure as the primary product.

Competing as Infrastructure by Becoming “The Glue”

By the time Dataiku reached scale, competition intensified from both directions. Snowflake and Databricks added AI features. Cloud vendors launched AI platforms. The question became: how do you compete when infrastructure vendors move up the stack into your territory?

Florian’s answer: double down on the translation layer positioning.

“The role we have in the enterprise is from the perspective of business users, you want them to be able to use those tools in an easy way, but easy also to master and govern from the perspective of the enterprise,” he explained.

When asked about positioning as “the glue,” Florian didn’t hedge: “Being the glue is perfectly fine for me.”

The strategic insight: infrastructure vendors will always optimize for technical capabilities, not business user experience. The enterprise data stack will continue fragmenting and evolving. “You’ve got a multiplicity of vendors and also this ecosystem very candidly change every two or three years,” Florian noted.

That permanent gap is Dataiku’s sustainable competitive position. Every infrastructure shift reinforces the need for a stable translation layer.

Measuring Retention Through Project Velocity, Not Platform Stickiness

Most enterprise software companies track net revenue retention, usage metrics, or feature adoption. Dataiku measures something fundamentally different: how many AI projects business teams can deliver to production.

“The way we think about it is the acceleration and the multiplication of AI projects,” Florian explained. “How many can I have running in production? So many projects can my business teams delivered and so forth. Can I measure and track their value and so forth. And that’s actually how you build the retention.”

This isn’t semantic—it’s a different retention model entirely. Traditional enterprise software creates stickiness through data lock-in, workflow integration, or switching costs. Dataiku’s retention comes from enabling customers to generate compounding business value.

“Retention is not built out of thin air or being just a virtue of being perceived as indispensable,” Florian said. “At the end of the day, retention is a value is just derived from the business value you generate.”

The model works because each successful AI project creates organizational momentum for the next one. Teams that deliver five projects in production become equipped to deliver ten, then twenty. The platform becomes valuable through accumulated capability, not technical lock-in.

Building Global Leadership While Keeping Engineering in France

Dataiku made an unconventional structural decision: keep product and engineering in France, but recruit experienced executives globally from proven US public companies.

“A lot of the people with the experience of scale and the experience of speed have got this experience in the largest U.S. companies,” Florian explained. “So you have to find this talent and build that mix.”

The leadership team includes executives from Zoom, Twitter, and ServiceNow—companies that successfully scaled to billions in revenue and navigated public company requirements.

“We are building our leadership team globally,” Florian said, noting they deliberately brought in “people in finance that have run public companies for years” to prepare for eventual public company readiness.

This hybrid model let them maintain technical excellence in their original base while accessing operational expertise required to sell and scale enterprise software in the US market—where most of the Fortune 500 buying decisions happen.

The 12-Year Relationship: Choosing Customers You’ll Serve for Decades

Before starting Dataiku, Florian received advice that shaped his customer selection: “If you fail as an entrepreneur, it will be for two years. You don’t really care. But if you succeed, you’ll be there for 10 years or 15 years or 20 years.”

The implication: “Whatever the domain, and in particular, whatever the customer, the type of customer, the persona you’re serving, you really need to love this persona because it’s a very long relationship.”

Twelve years later, Florian still serves the same buyer: “I love the person I’m serving, meaning the AI, data, science, data people, the people that are in between in the business. I love those guys, I relate to them. I’m like, I can sit with them for hours.”

This wasn’t about passion—it was recognition that building enterprise infrastructure means decade-long customer relationships. If you don’t genuinely connect with your buyer persona, those relationships become exhausting rather than energizing.

The Path to $350M ARR and 700+ Enterprise Customers

In October 2024, Dataiku announced $350 million in annual recurring revenue. The company has raised over $850 million and serves more than 700 enterprise customers, including 25% of the Fortune 500.

“It’s not a straight line,” Florian acknowledged. “Ideally, when you do entrepreneurship, it’s never a straight line because you’re looking for exponentials.”

But the trajectory validates their core thesis: enterprises need translation infrastructure to operationalize AI at scale. As foundation models improve and infrastructure vendors compete more aggressively, business users still can’t directly access the technology without a translation layer.

Florian’s view on the next five to ten years is that enterprises “will have to redesign and automate significantly those processes in order to stay competitive.” The economic logic is straightforward: “It makes no sense to invest a trillion in data centers if you can’t make those very much information driven processes automated.”

Every trillion dollars invested in AI infrastructure increases demand for the translation layer that makes that infrastructure accessible to the business operators who actually understand enterprise processes.

That’s the fundamental insight Dataiku executed on: in enterprise software, being the glue isn’t a compromise position—it’s often the most defensible one.