Is a $10 trillion proprietary data asset a moat or a risk?

For years, enterprise software vendors have viewed data accumulation as a competitive advantage. In the AI era, that belief has only grown stronger, with larger datasets increasingly positioned as a foundation for better decisions and outcomes.

‍

At Coupa Inspire 2026, a billboard on the Las Vegas Strip made the claim concrete: agents informed by $10 trillion in business spend deliver better data, better decisions, and better outcomes.

‍

It's a striking and carefully constructed figure, but volume and value are not synonymous. The quality of an AI agent is only as good as the data feeding it, and data quality is shaped by how it was captured, when it was captured, and on what infrastructure. Those factors matter as much as headline volume, and they raise questions that $10T in spend data alone cannot answer.

‍

This article breaks down why the quantity of spend data and the quality of AI-driven outcomes are not synonymous and what procurement leaders should be asking instead.

‍

Quantity ≠ Quality

Data quality is not a function of volume. As IBM notes, data quality is assessed through dimensions such as accuracy, completeness, consistency, timeliness, uniqueness, and validity, not raw scale.

‍

That distinction matters here. A large dataset can still be unreliable if it was captured inconsistently, structured unevenly, or assembled over years of changing procurement processes or behavior.

‍

In other words, more data does not automatically mean better decisions. In procurement, the real question is not how much data a vendor has, but whether that data is clean, current, and fit for how teams operate today.

‍

The legacy tech problem

Much of legacy tech data was built over years of ERP integrations, supplier onboarding, taxonomy decisions, and procurement workflows that were never designed for today’s AI-driven operating model. As those processes evolved, the underlying data often remained fragmented, inconsistently structured, and shaped by older business rules rather than current reality.

‍

Procurement data is notoriously difficult to standardize. Industry definitions of spend analysis explicitly describe the need to cleanse, normalize, and classify spend data before it can be reliably analyzed.

‍

Coupa's integration documentation reflects the complexity of connecting procurement workflows across APIs, ERP systems, custom fields, and exception-handling processes. The result isn't necessarily bad data, but it does mean that scale alone is not evidence of consistency. 

‍

The question procurement leaders should ask isn’t: how much data does a vendor have? But more so how was it captured, how consistently was it structured, and does it reflect today's procurement reality, or the habits of a previous era?

‍

‍

Can legacy tech buy its way out of legacy?

Legacy tech is trying to buy its way out of legacy not because they want to but because they arguably struggle to build AI fast enough. Normal times are different. Architectural limitations, talent gaps, and emerging players moving fast pose challenges.

‍

Coupa CEO Leagh Turner put it plainly when announcing the Tonkean deal: "To deliver the promise of increasingly autonomous execution…you need three main things: industrial-grade and unified orchestration, best-in-class buyer and supplier workflows, and clean, trusted data."

‍

This matters because outcomes don't come from data sitting in a platform. They come from successful platform adoption. A system that requires months of implementation, external consultants, and acquisitions to function as promised creates the adoption barriers that compromise ROI. No dataset, however large, produces better decisions if the teams it's meant to serve have already found workarounds.

‍

Complexity kills adoption, and adoption is what drives outcomes

Gartner projects that by 2027, 40% of enterprises will demote or decommission autonomous AI agents due to governance failures, which is almost inevitable when a platform was never fully adopted in the first place.

‍

Outcomes are driven by process adoption, and when it fails, governance fails with it, purchasing decisions made outside the platform don't get captured or fed back into the system. The dataset grows, but its accuracy doesn't. In short, that’s not a moat, that’s drift. 

‍

Final thoughts

While the $10T figure is striking, procurement leaders evaluating platforms today are asking sharper questions, not just about what a vendor has built, but about what it actually reveals.

‍

Coupa’s dataset, built on legacy infrastructure, reflects a procurement environment that has changed significantly, and sits within a platform that, by its own CEO’s admission, was still missing critical capabilities as recently as last month.

‍

Yet just weeks later, Leagh said, “While some are bolting AI onto aging systems, we have one platform that scales.” This claim is grounded in the architecture behind $10T of spend data.

‍

Better decisions and better outcomes do not necessarily come from the size of a dataset. They come from clean data captured at the point of intake, strong process adoption, and a platform architecture built for today’s procurement demands, not retrofitted to keep pace with them.