On a good day, enterprise AI looks like a model making a useful prediction. On a normal day, it looks like six people asking who owns the data, whether legal approved it, why last Tuesday’s result changed, and which cloud bill just caught fire. Dataiku sells the place where those questions are supposed to get answered.
The company began in Paris in 2013, when four founders - Florian Douetteau, Clément Stenac, Thomas Cabrol and Marc Batty - built Data Science Studio, or DSS. The original premise was almost stubbornly practical. Data scientists could code in Python, R and SQL; analysts could work through visual recipes; business experts could inspect the result; operations teams could deploy it. One project, several levels of technical fluency, fewer files named final_v7_really_final.
That collaboration layer became the business. By October 2025, Dataiku said it had passed $350 million in annual recurring revenue, served more than 750 organizations and employed over 1,250 people. Its customers include manufacturers, banks, drugmakers, insurers and retailers - the sort of companies where a model can change a factory schedule, flag a payment, reorder inventory or help a researcher find an answer faster.
The model was never the whole job
Dataiku prepares data, builds dashboards, trains machine-learning models, deploys them as services, monitors their behavior and wraps the work in permissions and approvals. Customers can click through a visual flow or drop into code. The point is not to make every employee a data scientist. It is to let a pricing analyst contribute business rules without pretending to be a machine-learning engineer, while letting the engineer see exactly what entered the pipeline.
That distinction matters. A bank’s fraud model is not valuable because its notebook has elegant code. It is valuable when it reads current transactions, survives a change in source data, produces an explainable decision, passes a risk review and reaches the right system at the right time. Dataiku tries to hold that chain together. Its software connects to warehouses and clouds rather than requiring a customer to place every bet with one vendor.
“What we learned over time is that it takes lots of energy and will to fill the gap between technology and business objectives.”Florian Douetteau, co-founder and CEO
Then the chatbots escaped the lab
Generative AI changed the shape of the mess. Companies could suddenly call models from OpenAI, Anthropic, Google and others, often through several clouds and app frameworks. Experiments multiplied faster than oversight. Dataiku responded with LLM Mesh in 2023, a gateway for model access, security and cost controls. In 2025 it added Agent Hub, a catalog and workspace for approved agents. By 2026 the company had recast the portfolio as the “Platform for AI Success.”
The new names are revealing. Agent Management inventories and measures agents even when they were built elsewhere. Cobuild turns a natural-language request into an inspectable visual project, including pipelines, models, agents and applications. Reasoning Systems coordinate agents with business rules and human checks for jobs such as manufacturing operations. Dataiku is moving from helping a team make AI to helping an institution decide which AI deserves to keep running.
This was not a repudiation of the old thesis. It was the thesis getting a larger blast radius. The company saw that the scarce resource had shifted from model access to organizational control. When every department can start an agent, the enterprise needs an inventory, permissions, evaluation, budgets and an off switch. A technically healthy agent can still be useless, expensive or wrong.
The first thing to break was the price tag
Dataiku has raised roughly $846 million across its disclosed rounds. The expensive chapter came during the software boom: a $400 million Series E in 2021 valued the company at $4.6 billion. In December 2022, Wellington Management led a $200 million Series F at $3.7 billion. The business had not vanished; the multiple had. That $900 million haircut is the clearest public setback in an otherwise tidy growth story.
The cost to customers is less visible. Dataiku sells enterprise subscriptions through quotes rather than a public price card. That makes the buying test straightforward: does the platform remove enough duplicated tooling, manual review, abandoned pilots and specialist time to justify another large software contract? For a global bank with hundreds of workflows, the answer can be yes. For a startup with three engineers and four models, the arithmetic gets awkward quickly.
The company itself has scaled with a similarly mixed operating model. It kept a substantial product and engineering identity in Paris while building headquarters and enterprise sales in New York, eventually spreading across 13 offices and remote locations. Its stated values include “accountable camaraderie,” learning and work-life balance; about 30 percent of employees were listed as fully remote. The whimsical office in the photograph is not proof of culture, of course. The more revealing detail is a $500 annual learning stipend described in its impact report - modest, specific and consistent with a company that needs salespeople, product managers and engineers to keep learning the same fast-changing vocabulary.
A neutral layer in a land of giants
Dataiku competes with several kinds of company at once. Databricks and Snowflake increasingly combine data, analytics and AI in their own platforms. DataRobot, H2O.ai and Domino Data Lab cover model development and operations. Alteryx and KNIME court visual workflow users. AWS, Microsoft and Google offer native services from infrastructure through model deployment.
Its answer is optionality. Dataiku works across those systems, mixing visual and code-first work while keeping governance above the infrastructure. The company can partner with Snowflake and Databricks even while their product boundaries overlap. In 2026, Snowflake named Dataiku its product partner of the year for AI platforms for the fifth time. The relationship works because a shared customer can keep data in Snowflake and use Dataiku as the collaborative work surface.
There is a catch. Neutral layers earn their keep only when the underlying environment is genuinely mixed. A customer committed to one cloud may prefer that cloud’s native tools. A sophisticated engineering team may assemble notebooks, Git, orchestration and open-source monitoring for less license cost, though rarely for zero operating cost. And every platform that promises one control plane risks becoming one more plane to control.
“No amount of prompt engineering replaces structured orchestration.”Clément Stenac, co-founder and CTO
What a reader can steal
The transferable move is not “build an AI platform.” It is to find the expensive handoff around a scarce expert. Dataiku did not try to erase the data scientist. It let analysts prepare inputs, domain experts add context, risk teams inspect decisions and engineers operate the result. Each adjacent user became more useful without receiving a fake promotion to expert.
The second move is to make the work legible. A visual flow can be dismissed as no-code decoration, but its larger value is inspectability. When an auditor, product owner or factory manager can follow the path from source data to decision, the organization can argue about the same artifact. Cobuild follows that logic: generate quickly, then leave behind something people can review rather than a mysterious block of generated code.
The third move is less glamorous: attach value and ownership before deployment. Who pays for the model calls? What result makes the agent worth running? Who signs off when it changes? What stops it? Dataiku is productizing those questions because enterprises routinely postpone them until an experiment becomes infrastructure.
When the Dataiku approach works - and when it does not
- It works when many technical and business teams share governed workflows across clouds, models and sensitive data.
- It works when review, lineage and production handoffs cost more than the license they replace.
- It struggles when a small expert team has few models, little regulation and one deeply standardized cloud stack.
- It fails if a visual interface merely hides complexity instead of making ownership and decisions clearer.
Dataiku’s future therefore hangs on an unromantic question: as models get cheaper and agents get easier to build, does coordination become more valuable? Its reported growth suggests large organizations already pay for the answer. But the company must stay above features that its partners can absorb and below the organizational problems they cannot solve alone.
That is a narrow ledge, but a useful one. The AI industry loves the moment a model produces an answer. Dataiku has spent thirteen years on everything that happens before and after: the cleaning, the argument, the approval, the deployment, the drift and, occasionally, the decision to turn the thing off.