The majority of enterprise AI investment in Europe over the past three years has flowed into vertical applications: products built for a specific function or industry, automating a specific workflow, sold to a specific buyer role. This has been the right starting point. Vertical specificity allows seed-stage companies to achieve the product depth required to clear enterprise reliability thresholds, and the domain expertise of the founding team creates a credible reason for buyers to trust a new vendor with a business-critical process. We have backed this pattern consistently through Fund I. The shift in Fund II deployment is not away from vertical specificity but toward something that starts narrow and has a clear, architecturally deliberate path to horizontal platform economics. The distinction matters.
A horizontal AI platform, as we are defining it for investment purposes, is not a product that works across all possible workflows for all possible buyers. It is a product where the core AI capability — whether that is natural language understanding, structured data extraction, decision logic, or workflow orchestration — is general enough that the marginal cost of extending it to a new vertical is substantially lower than the initial build cost. Taktile, in decision intelligence for financial operations, is building toward this pattern: the underlying decision engine is not financial-services-specific, it is a general framework for building, deploying, and monitoring automated decision policies that happen to be best expressed in financial operations first, but extends to any domain with structured decision logic. The distinction between "we are a credit risk automation tool" and "we are a decision intelligence platform that happens to have deep credit risk functionality" is not merely marketing language — it reflects architectural choices made from the first month of development.
The horizontal platform bet we are making in 2024 is concentrated in three capability areas. First, workflow orchestration infrastructure: the layer that connects AI-native processing components (document extraction, classification, generation) into defined business processes with state management, exception handling, and audit trail. Second, multi-modal document and data ingestion: products that can ingest, normalise, and make queryable the heterogeneous document and data formats that European enterprises actually operate with — PDFs from 1990s-era ERP exports, scanned documents from legacy archiving systems, semi-structured XML from supply chain platforms that predate modern API design. Third, enterprise-grade model serving and monitoring: the operational infrastructure for running ML models in production in a way that meets German enterprise IT standards for reliability, security, and audit transparency. These are the infrastructure layers that every AI-native application needs but that most teams rebuild independently rather than sourcing from a specialist provider.
The risk we are managing in horizontal platform bets is the go-to-market tension. Horizontal platforms have longer sales cycles and less obvious initial buyers than vertical point solutions, because the value proposition requires the buyer to understand their architecture problem rather than just their immediate process pain. The companies that have successfully navigated this tension have typically done it by starting with a specific vertical use case that demonstrates the platform capability concretely, building reference customers who can speak to both the immediate workflow value and the broader architectural potential, and only expanding the commercial narrative to "platform" once there is enough proof that the architecture genuinely delivers horizontal value. Announcing yourself as a platform before you have demonstrated platform-level value to multiple distinct buyer types is a sales mistake that costs pipeline and credibility simultaneously.
Our conviction on this thesis is grounded in a specific market observation: the enterprise software stack at a typical 300-person German company contains a growing number of AI-native point solutions that do not talk to each other, do not share data models, and do not have a common governance framework. The procurement team is using one vendor for invoice processing, a different vendor for contract management, and a third for spend analytics — each with its own integration to the ERP, its own data processing agreement, its own model monitoring responsibility. The operational overhead of managing this fragmentation is becoming visible at exactly the stage of enterprise AI adoption where the next generation of buyers is making architectural decisions. The horizontal platforms that can consolidate this surface — without forcing buyers to abandon the specialised functionality they depend on — are addressing a real and growing pain that the current landscape of vertical point solutions cannot resolve.