We closed Telhaverde Fund I in September 2021 with a specific thesis: generative AI would restructure the marketing stack, and the value would accrue to the companies building the infrastructure layer — content generation, personalization, creative intelligence — rather than the application layer that sits on top. That thesis was right in broad direction and wrong in several specific predictions. Fund II, which closed in March 2024, reflects what three years of deploying capital into this sector actually taught us.
Writing a thesis update is an exercise in intellectual honesty. The parts of the thesis that were right are easy to describe. The parts that were wrong, or that changed substantially as the market developed, are more useful — both for our own learning and for founders evaluating whether we are the right partner for their stage and category.
What we got right in Fund I
The core bet held. The marketing stack is being restructured, not incrementally improved. The companies that built purpose-specific AI tools for marketing workflows — rather than adding an AI feature to an existing marketing platform — captured different buyer behavior and built different retention dynamics than the AI-feature-in-existing-tool category. Buyers who chose a dedicated AI content generation tool chose it because it was fundamentally better at that task than an existing platform's AI feature. That meant the dedicated-tool companies had pricing power and retention that the AI-feature-in-existing-tool companies did not.
The data moat thesis also held in its most specific form. Companies that accumulated performance-labeled training data — not just content generation volume but actual deployment performance data from campaigns, from customer engagement, from conversion outcomes — built compounding advantages that generic model providers could not easily replicate. Three years later, the companies in our Fund I portfolio that are performing best are the ones that built and protected their proprietary training data in the core workflow.
The European market specificity thesis held in a way we did not fully anticipate at the time. When we wrote the Fund I thesis, our European angle was primarily about access — we were building from Lisbon, with a team that had run marketing operations in European enterprise, and we would have access to European founders that US-based funds would not see as early. That was correct. What we did not fully appreciate was the GDPR dividend: European enterprise buyers, under data compliance pressure, were more willing to pay for marketing tools that processed their data in Europe, with contractual data residency guarantees. That created a European market access advantage for our portfolio companies that was above and beyond what the product quality alone would have driven.
What we got wrong or underestimated
We underestimated how quickly general-purpose language models would improve, and what that would mean for the entry bar for AI marketing tools. In 2021, building a language model fine-tuned for marketing content was a meaningful technical barrier. By 2023, a well-prompted GPT-4 could produce reasonable marketing content across most categories. The entry bar for building an AI marketing tool dropped substantially, and the number of well-funded competitors in every sub-category of AI MarTech increased correspondingly.
The implication for our thesis was that technical differentiation alone was no longer sufficient for seed-stage companies in AI MarTech. We needed to weight distribution moat more heavily than we had. A seed-stage AI marketing tool with genuine technical differentiation and no theory of sustainable distribution is in a harder position than we had estimated when we built the Fund I thesis. For Fund II, distribution moat moved up in our investment criteria significantly. The best technical output in the world does not build a durable company if it can be reproduced by a well-resourced competitor in six months.
We also underestimated the consolidation dynamics in content generation specifically. The broad content generation category — general-purpose AI writing tools for marketing — consolidated faster and more aggressively than we expected. The companies that survived the 2023 consolidation wave were either deeply embedded in specific workflows with strong retention dynamics, or they were building toward platform scale with multiple product lines and expanding TAM. The middle — a good AI writing tool with decent retention but no specific workflow depth and no clear platform path — got squeezed from both directions.
What Fund II looks like as a result
Fund II is $37M against a total of $62M under management across both funds. It is larger than Fund I by design — we needed more capacity to write checks that gave us meaningful ownership in the companies where we had highest conviction, and to support companies into later stages with reserves.
The category priorities shifted on the margin. Content generation in its broad form is less interesting to us now than it was in 2021 — the category has matured enough that the structural advantages required to win have compounded past the point where a new seed-stage entrant can realistically build them from scratch. We are more interested in the infrastructure layer beneath content generation: the content intelligence and performance optimization layer that tells you what to generate, for whom, and how to improve based on outcomes. That layer is less populated and the technical requirements are higher.
Creative intelligence — the intersection of generative creative tools and performance analytics — is a Fund II priority that was not as prominent in Fund I. Our investments in Smartly.io (2023) and Omneky (2025) reflect this. The companies building in this space have a different technical profile from content generation companies — more emphasis on computer vision and multimodal modeling, more emphasis on performance measurement and attribution, more sophisticated data infrastructure requirements — and they are selling to a buyer persona that is willing to pay more for measurable performance improvement.
What has not changed
Our stage (seed), our check size (€1.5M to €3M), and our geographic focus (Europe and European founder global expansion) have not changed. We still believe that the seed stage is where the most interesting value in AI MarTech is created, because the technical foundations of defensible companies are laid at seed. A company that has already built its data architecture, defined its feedback loop, and demonstrated retention in a specific workflow at seed has compounded past the point where a later-stage investor can easily evaluate whether those foundations are solid.
Our conviction that operator-background founders have structural advantages in this space has also not changed. The founders in our portfolio who have consistently outperformed are the ones who built and shipped marketing technology before they started a company. They know where the real friction is, they know what buyers will actually pay for, and they know how to evaluate whether their model output is production-quality. That last point is more important than it sounds — knowing the difference between a good demo and a production-ready product is not obvious from outside the workflow, and the founders who have been inside the workflow know the difference.