When a consumer asks ChatGPT or Gemini for a product recommendation, the role of traditional marketing crumbles in an instant. It is no longer enough to dominate Google: the last mile of product discovery shifts into chatbots, where winning relies not on sponsored links but on the semantic relevance of generated answers. It is against this backdrop that Serpier, a martech startup based in Aarhus, has just secured €1.4 million from True Collective and the Export and Investment Fund of Denmark (EIFO) to automate that very visibility.

Navi, the agent that writes and publishes without touching the backend

Founded in 2024 by Steffen Sørensen, Simon Holm, Søren Fuhr and Thomas Grástein, Serpier has built an AI agent – named Navi – capable of spotting opportunities to improve an e-commerce's online presence, creating content, and publishing it autonomously. This is not just an optimized text generator: Navi orchestrates the full cycle, from analysis to distribution, with the explicit goal of freeing marketing teams from execution. "Our agent can already handle operational tasks; now we want it to build landing pages, launch campaigns, and automate additional workflows," said co-founder Søren Fuhr.

The speed of execution has yielded tangible results: more than €2.5 million in revenue in its first fiscal year, and profitability that is far from guaranteed for a startup. But the striking figure is not merely financial but architectural: Serpier optimizes not only for traditional ranking but for how Large Language Models (LLMs) assemble responses. That means accounting for conversational context, how often a brand is cited in training datasets, and the coherence between a query and the structured information available online.

What changes when LLM inference becomes the new store shelf

For an e-commerce, being surfaced on ChatGPT or Gemini is not an early-adopter whim; it is a competitive necessity that ties revenue to the quality of model inference. If a user asks "best trail running shoes under one hundred euros" and the generated answer includes two competitors but not your catalog, the economic damage is immediate. Serpier’s approach tries to mitigate this risk by actively managing presence, but it raises a crucial question for anyone handling product data, price lists, and purchase preferences: where does this information end up?

Every interaction with a public chatbot entails sending potentially sensitive prompts to external servers, often outside European jurisdiction. GDPR imposes strict constraints on data transfer and processing, and for companies operating in regulated sectors or with stringent digital sovereignty policies, relying exclusively on non-domiciled cloud services can become a friction point. It is no coincidence that self-hosted architectures and on-premise LLMs are gaining traction in the enterprise world: they allow control over the entire flow, from end-user input to generated recommendations.

The sovereignty knot and the paths open for on-premise deployment

Serpier is born as a SaaS platform and currently does not declare any on-premise installation options. Nevertheless, the direction the market is taking – autonomous agents that write, publish, analyze, and soon launch campaigns – forces a reflection on where the decision engine actually resides. If the Navi agent handles content and campaigns, it means it accesses commercial data, promotional strategies, and conversion metrics: an information asset that many companies prefer to guard within their own infrastructure.

For those evaluating an on-premise deployment, there are well-known trade-offs: the need for GPUs with adequate VRAM, the management of quantization to balance performance and accuracy, and the overall TCO that must compare hardware CapEx with cloud OpEx. But when an AI agent starts touching core processes such as pricing or personalized landing page generation, the cost of decision latency entrusted to third parties can outweigh the investment in a dedicated server fleet. This is not about replacing Serpier with an internal stack – the platform’s value lies in vertical automation – but about asking how strategic it is to control the inference layer on which these automations rely.

The horizon: agents that execute, marketers that govern

The €1.4 million round will be used to turn the platform into a workspace where multiple AI agents collaborate on heterogeneous marketing tasks. Fuhr’s vision is clear: "Marketing is moving toward a model where AI agents handle analysis and execution, while professionals focus on strategy, prioritization, and key decisions." This perspective has the merit of making the paradigm shift tangible: from tools that assist, to machines that execute.

In this scenario, the underlying architecture becomes a non-secondary competitive factor. For enterprises that want to integrate similar automations without giving up data residency or regulatory compliance, the dialogue between SaaS platforms like Serpier and on-premise managed LLM models could define the next standard for digital marketing. Not forgetting that when an agent publishes content on behalf of a brand, the party responsible for the correctness of the information always remains the data controller.

AI-RADAR closely follows these evolving scenarios, offering analytical frameworks at /llm-onpremise for those who need to evaluate the trade-offs between control, cost, and performance when adopting AI agents in e-commerce marketing.