Hilbert and the Hidden Machinery of Growth

A first order can be a splendid liar. A company buys attention, offers a discount and watches its acquisition chart climb. The customer accepts the bargain, then never returns. Marketing records a win. Finance records the cost. The product team sees a brief visit. All three departments have a true number, and none has the whole story. Hilbert exists in the gap between those numbers.

The San Francisco company calls its product AI-native growth infrastructure. Strip away the vocabulary and the proposition is direct: make consumer behavior legible enough that a business can act while an opportunity still exists. Hilbert targets businesses with repeat customers and complex funnels, especially retail and transactional companies. It says its system spans acquisition, retention and monetization, the three rooms in which growth teams often keep separate ledgers.

The expensive argument over a simple number

The founders did not arrive at this problem through a brainstorm about artificial intelligence. Nazlı Tan, Ceyda Erten, Özgür Akaoğlu and Cenk Batman worked on growth systems at Getir, the fast-expanding delivery business. Hilbert says its earlier work reached nine countries and involved hundreds of millions in budget. Tan led Getir's growth function; the group had to make decisions across markets, channels and products while the underlying data kept changing.

The problem that failed first, in their account, was the foundation beneath the strategy. Events were scattered. Teams used different metric definitions. Insights arrived as dashboards, then waited for a person to turn them into a decision. A marketing team could optimize cost per acquisition while a finance team worried about customer lifetime value. Neither calculation was foolish. The danger lay in treating them as unrelated.

This is the point at which the founders seem to have changed their minds about what software should do. A better chart would still leave a human to reconcile the definitions, find the causal clue and persuade three departments to act. Hilbert instead tries to install a common metric layer, label customer behavior and place agents on top of it. The company is selling the plumbing and the operator in one package.

“It's all about you learning from your company deeply.”Nazlı Tan, speaking to Axios

A growth loop with four verbs

Hilbert describes its work as detect, reason, act and optimize. Detection might flag a cohort whose repeat orders have begun to weaken. Reasoning connects that change to factors such as acquisition source, discounts, delivery experience and customer value. Action could mean a reactivation play or a paid channel adjustment. Optimization compares what followed with what the model expected, then updates the next choice.

Its public examples are concrete. Hilbert lists churn risk, predictive lifetime value, cross-sell opportunities, budget allocation and unit economics among the signals it can work with. A promotional analysis shown on its site points to 20,361 customers who used discounts on their only purchase and did not make a second order after at least 90 days. The display associates those orders with about $771,000 in promotional spend. It is a product illustration, not a published audited customer outcome, but it makes the question plain: how much of that spend bought a habit?

The answer requires joining records that companies commonly keep apart. An ad platform knows what an impression cost. An ordering system knows what went into the basket. A finance system knows the margin after discount and fulfillment. A customer experience system may know why the delivery disappointed. Hilbert's differentiation is that it attempts to keep these facts in a persistent, business-specific model rather than asking a general chatbot to rediscover their meaning in every prompt.

The customers offer a better test than the slogan

Blank Street's global head of digital praised Hilbert for thoroughly cleaning and preparing data before ingestion. That is a remarkably useful endorsement. No one buys AI to admire a tidy database; everyone notices when an untidy one gives a confident, wrong recommendation. The praise suggests Hilbert's early value can be the patient work required before the agent takes a turn.

FreshDirect's vice president of strategy and analytics describes a different payoff. The grocer had two decades of customer behavior to draw on, but static definitions of loyalty and churn could not capture how customers were changing in real time. The executive says Hilbert helped replace those fixed labels with live predictive intelligence, making it easier to prioritize retention and identify high-value behavior. Erewhon says it explored a light integration for personalization and product recommendations in its digital store. Axios also named Walmart and Levain Bakery among Hilbert's customers. Those names show range, although public case studies do not yet provide a consistent set of independently measured return figures.

That range helps explain the market Hilbert occupies. Conventional business intelligence can describe what happened. Marketing automation can send the message. Customer data platforms can assemble a profile. In-house data scientists can build tailored models, if a company can hire and keep them. Hilbert's bet is that the awkward handoffs between those jobs are themselves a product category. Its pitch to a growth leader is speed; to a finance chief, a clearer connection between spend and value; to an analyst, fewer weeks spent untangling the same definition.

What $28 million buys, and what it does not

Andreessen Horowitz led Hilbert's $28 million Series A in April 2026. ScaleX Ventures, which backed the team earlier, joined the round. The investor's published argument is revealing: consumer companies often struggle to find the people who can build and maintain the data foundation beneath experiments and campaigns. Hilbert offers those companies an alternative to assembling that function from scratch. The funding pays for building and distributing the platform; it says nothing precise about what an individual customer pays. Hilbert sells through demos, and its Startup Program advertises startup-focused pricing without publishing a price list.

The startup program also clarifies when the software has a chance to work. Applicants need early user data or clear behavioral signals. Hilbert offers guided integration, a metric layer, its AI Growth Brain, monthly reviews with operators and scientists, and playbooks for retention, average order value, cross-sell and churn. A new product with almost no repeat behavior has little for a predictive system to learn. A company that has not agreed on what counts as an order, active customer or profitable cohort will need to settle those arguments before trusting automated action.

There is a practical method here even for teams that never buy Hilbert. Put marketing, product and finance around one table. Pick a single customer journey, perhaps first purchase to second purchase. Agree on the event definitions, the full cost of the incentive and the time window in which success counts. Follow a cohort through the whole journey, including margin. Then ask which signal would have let you act earlier. The exercise is ordinary. That is precisely why it is so often postponed.

Hilbert's more ambitious claim is that this exercise can become continuous infrastructure. It is still a young company, and broad claims about autonomous growth deserve proof over many customer cycles. Yet the founders' insight is hard to dismiss: a business cannot accelerate a decision it has never defined. The first order may flatter you. The machinery beneath the second tells the truth.