There is an old saying about planting a tree: the best time was twenty years ago, and the second best time is today. I have been thinking about that proverb in the context of procurement data because that’s what I do, but also because after decades of treating data quality and integration as tomorrow’s problem (or somebody else’s problem), it is time to take ownership and act.
Poor procurement data is not a “dirty little secret;” it’s an industry standard. (Sidebar: When I was a strategic sourcing consultant, I ran multiple sourcing projects for my clients where the only data source was from the requests the team made to its suppliers. I ran one project where only 12 of the 41 contingent workforce staffing firms sent their spend. Despite that data/knowledge gap the eSourcing event identified an overall 36% savings compression on admin mark-up rates; in part, because of that data/knowledge, new contracts were never administered.)
Procurement’s data problem is a foundational issue that has existed for a generation. CPOs (and their consultants) have spent years working to improve spend visibility, clean up the supplier master, link contracts to orders, and organize and centralize information spread across S2P solutions, ERPs, risk systems, spreadsheets, and inboxes. Major progress has been made, but overall fragmentation persists.
The Data Problem We Already Know
Ardent Partners’ research for our recent The Path to AI-First Procurement: Closing the Gap Between AI Ambition and Execution reinforces the issue in a direct and clear way. Of the 300+ CPOs and senior procurement leaders surveyed, only 11% of the say they operate with a unified procurement data model across Source-to-Pay. Another 41% have integrated environments that require manual reconciliation, while 40% describe their data as fragmented across silos. The remaining 8% still have meaningful procurement information trapped in unstructured formats.
That means 89% of procurement organizations in 2026 operate without a unified data model. To me, the more interesting part of the story is how well procurement has progressed despite this problem. It’s a testament to the savvy procurement teams that know where the reliable data sits, which supplier record is right, what information should be ignored, and what to do when systems disagree. Institutional knowledge compensates for the gaps between systems that were never designed to work together (Add this to the long list of reasons why I have been recommending for the past year+ that CPOs should reinvest their AI dividends).
For years, the consequences showed up as inefficiency, extra work, weaker analytics, and inconsistent visibility, but also as suboptimal outcomes that took longer to generate. Livable consequences, if not desirable. With a majority of procurement teams piloting or using AI in some capacity, solving the underlying problem has become much more urgent.
Time to Plant the Tree
The traditional argument for better procurement data centered on visibility (reporting, analytics, and opportunity). AI expands the issue into something much more operational: whether an intelligent system has trustworthy context to understand what is happening and act reliably.
Take a common incumbent supplier data issue that can impact contract renegotiations. A sourcing pro will typically review the supplier record, current contract terms, recent purchase orders, and performance history, before deciding if and how to engage. That person can move across several systems, recognize that differently named records refer to the same supplier, account for an outdated field, and fill in missing context from conversations with stakeholders and general experience.
An AI agent May be able to move through that information much faster, but speed only helps when the information is reliable, connected, and sufficiently complete. The agent needs to understand which records refer to the same supplier, be able to access the relevant history, and operate within clear boundaries around what it can recommend (or execute).
Agentic agents working across S2P processes need to understand the interrelationship between sourcing and contracts and orders and invoices and performance and carry context from one activity into the next. AI pilots can succeed inside organizations with weak data foundations. A tightly defined use case can be fed a clean dataset, limited to one workflow, and produce impressive results. The problem becomes much harder when CPOs try to scale the pilot across the function.
That is why the 11% finding is so important. As AI moves from assistance towards execution, partial context becomes a larger constraint because the human workarounds that have masked the problem for decades are less available.
CPOs Need to Own Their Data
This is why I think the role of the CPO has to expand. Data ownership should now be part of the CPO mandate. IT will remain essential to the architecture, integration, security, and technical management of enterprise data, but procurement leaders have to define the information their function needs, the quality standards that matter, the relationships that must be preserved across systems, and the business context AI requires to operate effectively.
That starts with understanding the current environment with much greater precision. Which supplier records can be trusted? Where is contract intelligence stored? How consistently are categories, entities, and transactions defined? Which data requires manual reconciliation? Most importantly, which weaknesses directly constrain the AI use cases that matter most to the business?
This is not a call for a massive data-remediation project aimed at some near perfect state. Perfect is the enemy of done so the better approach is to link data priorities directly to your AI priorities. If autonomous sourcing is strategic, identify the supplier, category, contract, and sourcing-history data it requires and strengthen those foundations first. If invoice exception resolution is the big near-term opportunity, focus on the relationships across invoices, POs, receipts, contracts, and supplier records. If supplier risk is the priority, start with entity resolution, external data feeds, and the supplier master.
Important point –> data ownership also extends to governance. It’s important to determine which sources are authoritative, who can access them, how changes are controlled, and whether an AI-generated decision can be traced back to the information that informed it. Those choices will shape trust, and trust will ultimately determine how much autonomy procurement teams will give to their AI systems.
The best time to build this foundation may have been twenty years ago. The second-best time is now because AI has transformed procurement data from a persistent operational headache into a strategic constraint on how far and how fast the function can advance.
CPOs have spent years asking technology providers to help make their teams smarter. Now they need to make sure their own data gives new technology something intelligent to work with.
In Part Three: As AI takes on more work and greater decision-making authority, the next control problem for procurement is governance.
This article is the second in a six-part series based on the Ardent Partners report The Path to AI-First Procurement: Closing the Gap Between AI Ambition and Execution, one of the most comprehensive studies conducted to date on how AI is reshaping procurement. The full report, produced by Ardent Partners in partnership with Ivalua, is available from Ivalua.