The most honest read on enterprise AI adoption in 2024 and 2025 isn't that the technology failed. It's that the wrong products were often deployed into the wrong environments, without enough thought about what success actually looked like.

Large organisations across financial services, real estate, legal and construction for example, spent significant sums on AI strategies that tended to produce one of two things: tools that were too narrow — single-use automation like lease abstraction or contract summarisation — or tools that were too broad — general-purpose copilots layered onto existing workflows without deep integration into the data people actually used. Neither tended to stick. Narrow tools solved isolated problems but didn't change how work got done. Broad tools, disconnected from proprietary systems and institutional knowledge, gave people little reason to change behaviour.

Both paths often led to the same outcome: underwhelming adoption and leadership teams trying to explain what went wrong. For service businesses where the core offering is expert human judgment, that pressure is particularly acute.

The premise of vertical AI is that industries with high document volumes, complex domain-specific reasoning and legacy cost structures can support platforms built specifically for that context — not AI adapted from a general tool, but systems architected around the workflows, data structures, and decision logic of a particular industry.

The potential moat, when it works, compounds in interesting ways. Every workflow embedded, every data source connected, every decision logged makes the platform more useful and harder to replace. The system of record becomes the system of intelligence. That's a different dynamic from horizontal tools, which face ongoing commoditisation pressure as foundation models improve and switching costs stay low.

The verticals that seem most interesting to us tend to share a few characteristics. They run on fragmented data — documents, signals, institutional knowledge spread across disconnected systems — where there's no shortage of information but a real shortage of intelligence over it. They involve high-stakes, time-sensitive decisions where faster and more accurate analysis has obvious and quantifiable value. And they've historically underinvested in software, relying on people and process in ways that leave them exposed when a genuinely capable alternative emerges.

That said, we're aware this is a thesis with real risks. Domain specificity is a strength until the domain shrinks, or until a sufficiently capable horizontal tool closes the gap. Building deep takes longer and costs more. And the behaviour change required to get adoption in entrenched industries is never straightforward.

Our first investment in this space was Fifth Dimension, which is building an AI-native platform for real estate — connecting fragmented data sources, automating complex workflows, and helping professionals move faster on deal analysis and asset management. What drew us in was the team's starting point: rather than adapting a general capability to a new context, they began with deep domain knowledge and built the architecture around it. That approach is playing out, with strong growth and expanding use cases since we first backed them.

More recently, we've been looking at companies applying similar thinking in other verticals. In retail and CPG, for example, we've been in conversation with businesses tackling the complexity of supplier negotiations — an area where email and spreadsheets remain the norm despite the significant contract value at stake, and where deep domain expertise in how deals are structured and negotiated creates a real edge over generic tools.

The pattern we find compelling across these companies is usually consistent: they start with workflow and work backwards to the product, rather than starting with AI capability and looking for problems to solve. They treat proprietary data as infrastructure. And they're building into verticals where the incumbents — consultancies, advisory firms, service businesses built on human expertise — have tried and largely struggled to solve the problem internally.

We continue to look for founders building in this space across professional services, infrastructure, and any industry where the combination of data fragmentation, high-stakes decisions, and structural underinvestment creates a clear opening.

The question we keep coming back to is a simple one: does this product refocus an entire job function on what matters, or does it make an existing process marginally more efficient? The former is the harder thing to build, but it's where we think the most interesting companies will emerge. If you're working on something in this space, we'd love to hear from you.