The number Zelara leads with is a result, not a raise. Working with a European neobank, the Berlin startup says it lifted customer reactivation by 66% without touching the bank’s CRM or redesigning a single customer journey. Reactivation, getting a dormant customer to come back, is one of the hardest and most valuable things a consumer business does, and a two-thirds improvement on it is the kind of figure that makes a pre-seed worth writing about. Zelara just raised €3 million on the back of it, which is a reminder that at the earliest stage investors are not buying a product so much as a single proof that the thing might work.

The problem it is attacking is old and expensive. Brands have spent two decades and enormous budgets on CRM and marketing technology, and what most of them got for the money was the ability to run the same predefined journeys and rule-based campaigns at everyone in a bucket: the “lapsed users” bucket, the “high value” bucket, the “win-back” bucket. The software was built to execute campaigns at scale, not to learn from them. The industry promised personalisation and shipped segmentation, which is a different and much blunter thing. A segment is a confession, really: it is what you build when you cannot afford to know the actual person, so you sort them into the nearest crowd and message the crowd.

A layer, not a rip-and-replace

Zelara’s product is a learning layer that sits on top of a brand’s existing CRM rather than trying to replace it. It decides the message, the channel and the timing for each individual customer, treats every interaction as feedback, and improves over time, while the marketers keep control of the strategy and the guardrails. “Instead of optimizing campaigns for segments, the system learns how to engage each individual customer in ways that create more value for both the customer and the business,” said co-founder Björn Heckel. The segment, in his framing, was always a shortcut for not knowing the person, and the shortcut is no longer necessary once a system can learn fast enough to treat a million people as a million people.

The architecture choice is also the commercial one, and it is shrewd. Sitting above the existing stack means low switching cost, fast deployment, and no migration project for a procurement committee to kill. The incumbents own the budget and the system of record; Zelara is not asking a marketing team to tear that out, which is the request that gets a startup shown the door. It is asking only to be allowed to make the existing investment smarter. That is a far easier yes.

The catch is trust, not technology

The round was led by NAP, with Heartfelt and Angel Invest joining. Heartfelt is a corporate-backed investor, which matters at pre-seed: it suggests a strategic reader of the space, not only a financial one, looked at the thesis and decided it had legs. Founders Nikolas Schriefer and Björn Heckel have spent years building this kind of technology, which is partly why the pitch is “layer on top” rather than the more naive “rip and replace” that first-time founders in martech tend to reach for.

The real obstacle is not the model, it is the handover. The product asks marketers to give up the moment-to-moment decisions, the message, the channel, the send time, to an automated system, and marketers are paid for and judged on exactly those decisions. Handing them to software is a question of professional trust as much as software quality, and the entrenched platforms Zelara sits on top of already own both the budget and the relationship. There is also a quieter risk in being a layer: a layer is, by definition, dependent on the platforms beneath it, and the largest of those platforms can decide to build the same learning capability themselves and bundle it in.

One neobank result is a genuinely strong start, the kind that wins a pre-seed and earns the next ten conversations. It is not yet a pattern. Whether the next ten buyers will hand over the controls, and keep them handed over once the novelty fades and a campaign underperforms, is the question the €3 million is meant to answer.