What 50 Enterprise AI Implementations Taught Us About Procurement
Most commentary on enterprise artificial intelligence is written by people who have never participated in an enterprise procurement cycle for it. The result is a market understanding that overweights what AI can theoretically do and underweights what enterprise buyers will actually purchase. We approach this gap from the opposite direction. Our European platform has executed implementations across more than fifty enterprise clients, in industries ranging from heavy industry to legal services to public administration. The observations below are drawn from that direct operational experience. They will be familiar to any reader who has personally sat across the table from an enterprise procurement committee. They may be unfamiliar to readers whose understanding of AI adoption is shaped primarily by financial press coverage.
Observation one: The buyer is not who you think it is
In nearly every enterprise AI deal we have closed, the official buyer named on the purchase order is not the person who made the decision. The decision was made earlier, at a more senior level, often in a conversation that included no procurement language whatsoever. By the time the deal reaches the procurement function, its commercial substance has already been settled. Procurement is administering the decision, not making it.
This has substantial implications for go-to-market strategy. Founders who optimize their sales process for the procurement function are optimizing for the wrong audience. The procurement conversation is a confirmation conversation, not a persuasion conversation. The persuasion conversation happens months earlier, at the executive level, in contexts that do not look like sales meetings.
Observation two: Enterprise buyers do not pay for software. They pay for outcome.
This is the single most consistently misunderstood feature of the enterprise AI market. A founder offers a software product. The enterprise buyer evaluates that product not on its features, but on a question the founder rarely thinks about explicitly: if I purchase this product, who is accountable when something goes wrong?
If the answer to that question is “the buyer’s own organization,” the deal is structurally fragile, even if the software is excellent. If the answer is “the vendor, jointly with an implementation methodology and a defined accountability structure,” the deal closes faster, at higher prices, and with longer contract terms.
This is why pure software pricing models often underperform in enterprise AI. The market is not buying software. It is buying transferred operational risk. A vendor that can credibly transfer that risk — through implementation services, training, governance frameworks, and sustained engagement — sells a fundamentally different product than a vendor that delivers only a working application programming interface.
Observation three: The fastest path to close is rarely the most impressive demonstration
A common pattern in enterprise AI sales: a founder delivers a technically dazzling demonstration. The enterprise audience nods appreciatively. The conversation seems to be going well. Then nothing happens for nine months. The deal eventually closes — or, more often, does not — for reasons that have no apparent connection to the demonstration itself.
The explanation is that the demonstration was the wrong artifact. Enterprise buyers do not need to be impressed by what artificial intelligence can do. They have already read the same financial press as everyone else. They need to be reassured that this specific vendor can deliver a specific outcome in their specific organization, within a specific timeframe, under specific accountability conditions. None of that is communicated through a demonstration. It is communicated through a deployment methodology, a reference architecture, a credible implementation timeline, and a candid acknowledgment of what can go wrong.
The vendors who learn this lesson early skip the dazzling demonstration and instead present an implementation case study — a previous client, a similar problem, a structured deployment, and a measured outcome. The conversion rates are dramatically different.
Observation four: Training is not adjacent to implementation. It is part of it.
A counter-intuitive operational finding from our work: enterprise AI implementations that include structured training for the buyer’s own people complete more successfully, more quickly, and with substantially higher post-deployment satisfaction than implementations that focus exclusively on technical delivery. The reason is not surprising once stated plainly. The post-deployment success of any AI system depends on whether the buyer’s organization knows how to operate alongside it. A technically excellent system delivered into an unprepared organization is a system that will be quietly abandoned within six months.
This is why we treat training and implementation as a single integrated product line, not as two separate services. Many specialist vendors resist this integration because their internal economics treat training as a low-margin add-on. This treatment is, in our view, a strategic error. Training is the mechanism through which implementation becomes durable. A vendor that owns both functions captures retention and expansion revenue that a pure-implementation vendor structurally cannot.
Observation five: The regulatory environment is a selling point, not a barrier
European enterprise buyers think about artificial intelligence in the context of the European Union’s regulatory framework — specifically the AI Act, sector-specific regulation, and the broader institutional culture of compliance. A vendor that treats this regulatory environment as an obstacle to be navigated around is a vendor that will lose deals. A vendor that treats it as a feature of the buying conversation — building governance frameworks, documentation standards, and compliance scaffolding directly into the implementation methodology — is a vendor that closes deals faster.
This is a difficult posture for American-built products to adopt from a distance. It requires native presence inside the European regulatory and operational context. It is one of the reasons our investment thesis weights operational depth inside the relevant market so heavily.
Observation six: The most valuable deals start as the smallest deals
The pattern is consistent across industries. The most strategically important enterprise relationships in our portfolio did not begin as large engagements. They began as narrow, focused pilots — sometimes deliberately small, often initially funded out of discretionary executive budgets rather than formal procurement processes. The reason these small starts produce the largest accounts is operational. A focused pilot that delivers a measured outcome creates internal credibility for the executive sponsor. That credibility is the political capital that funds the next, larger deployment. And the next. And the next.
Founders who pursue large initial deals frequently lose them. Founders who pursue small initial deals with credible expansion paths frequently win the entire account over a multi-year period. The total contract value differs by an order of magnitude. The initial deal size, paradoxically, is often inversely correlated with eventual account scale.
The implications for investment
These observations shape how we evaluate companies in the applied enterprise AI category.
We weight evidence of senior executive relationships over evidence of product feature completeness. We weight implementation methodology over demonstration impressiveness. We weight integrated service offerings over pure software pricing models. We weight regulatory and governance maturity over pure technical sophistication. We weight account expansion patterns over initial deal size.
Founders who recognize these patterns instinctively are easy to identify. They speak about their customers’ organizations rather than about their own technology. They describe deployments in terms of outcomes and accountability rather than in terms of features and capabilities. They treat training and implementation as integrated rather than separate. They are comfortable in boardroom-level conversations because they have had many of them.
Enterprise artificial intelligence is, ultimately, a discipline. The companies that succeed in it are the ones that approach it as such.