Why Procurement Is Finally Ready for AI: Insights from SAP’s Former Chief Procurement Officer and Startup Investor

A conversation with Dr. Marcell Vollmer

Probably the scariest moment of building my previous company came during a trip to Matamoros, Mexico to visit one of our suppliers, just months after a group of Americans were kidnapped and murdered by a cartel there. After that meeting, we had to drive on to Monterrey, passing through Reynosa, the second-deadliest zip code in Mexico after Matamoros, to meet a customer.

Photos from some factory tours in Matamoros, Mexico.

The second scariest moments came during negotiations with procurement teams. To say procurement teams are penny pinchers is not an understatement – they literally will prevent a contract from closing because of a penny. I remember one Head of Procurement of a large multinational company who was arguing against 1 cent on a single part of a project that was worth multi-millions at production scale.

For that reason, I’ve historically been bearish on procurement as an area worth investing in. I mean, if procurement teams are so precious about a single penny on a project, there’s no way they are going to be willing to spend 6-figures or more on AI tech. However, as I started talking to more people in the space, I realized that wasn’t quite accurate. In fact, procurement teams are very willing to invest in their tech stack if it gives them leverage to better negotiate those pennies per project.

My exploration into this space brought me to Dr. Marcell Vollmer. With more than fifteen years in the function, time as Chief Procurement Officer, a decade-long leadership career at SAP where he also served as COO of SAP Ariba post Ariba’s $4b acquisition by SAP, and over twenty startup investments, Marcell has lived procurement from every angle.

When we sat down to talk about procurement’s AI moment, he acknowledged that “selling to procurement is a nightmare… and procurement processes very often suck in the perception of the businesses (and suppliers) .” But he also said that for the first time in decades, procurement is becoming a center of transformation towards a value generating function rather than a back-office afterthought.

Below are insights from our conversation.

Marcell’s Path Into Procurement

Marcell’s path to procurement began during his tenure at SAP, where he helped define the company’s approach to enterprise-wide finance, supply chain, and procurement transformation. It was a period when the category was still emerging and underfunded, but it sparked his long-term interest in the space.

“You can describe the supply chain world with four words: plan, source, make, deliver. So simple it is.”

Procurement (source) is universally considered as painful, and universally the same. Across industries and geographies, he says the fundamentals simply don’t change. “You have one process that could fit potentially all businesses around the world… purchase requisition to sourcing to contract to purchase order to goods receipt to invoice. This process is the same, it doesn’t matter if you’re buying military goods, services, or complex products.”

Marcell has seen countless data-cleaning initiatives fail, “I have found so many data cleansing projects and usually after day one or two your data is already outdated again.” Master data cannot be fixed manually, and no amount of human effort will ever solve it. “The only way you can solve it is with artificial intelligence,” he tells me. “Let the machines do the work so procurement can focus on value-generating tasks.”

Why AI Is the First Real Breakthrough in Procurement

If you zoom into a typical enterprise procurement workflow today, it’s shocking how fragile the systems are. Quote normalization often means buyers are manually copying and pasting numbers from half a dozen supplier formats into one master Excel file. Negotiation strategies live in people’s heads or scattered PowerPoint decks rather than in consistent, data-driven playbooks. Savings tracking is retrospective and buried in spreadsheets that only get opened at quarter-end, long after the window to course-correct has passed. Meanwhile, buyers spend more time wrangling data and chasing confirmations than actually negotiating value.

That said, procurement has always been a strong candidate for automation because of the nature of the work: structured workflows, repetitive tasks, rule-bound processes, and high dependency on data that people don’t want to manually maintain.

“Why do I still need to look in catalogs for stuff I need? Why do I need to manually process things I already have as a material master number?” Marcell asked in our conversation. Immediately following-up with the answer: “you shouldn’t. And with AI, you finally don’t have to.”

And because procurement processes are remarkably standardized across industries and geographies, AI systems can finally be deployed in a way that truly scales. As Marcell put it, “procurement is basically one solution and one process everywhere,” which removes the usual barriers to adoption and makes broad, repeatable deployment far more achievable than in most enterprise functions.

AI agents thus start to become basic infrastructure. They can sit in the flow of work. Reading order confirmations from suppliers, extracting key terms, matching them to material master data, and flagging discrepancies automatically. They can normalize quotes submitted in wildly different formats and present a unified view of options. They can surface anomalies in pricing or delivery performance that would be impossible to spot manually across thousands of SKUs. In other words, these agents give procurement the clean, structured, continuously updated data it always needed but could never maintain by hand.

How Procurement Leaders Actually Buy AI

The budget narrative in procurement is always “There is no budget,” says Marcell. “If you ask, there is no budget. Period. Forget it.” But that doesn’t mean they won’t buy, “If the solution works, the Chief Procurement Officer will find the budget… or take it out of the existing IT budget,” he said.

The early adopters will be the teams under intense pressure to deliver savings with leaner staff and compressed sourcing cycles. “They’re desperate to claw back a few hours a week per buyer and avoid expensive mistakes,” Marcell noted. For them, an AI agent that can be integrated quickly and demonstrably saves even just three to five hours per week can be meaningful.

This is why he believes the right go-to-market playbook always starts with a small, low-friction entry point: “Make it easy to deploy and to start. The POC helps you get a foot in the door and then expand.” In practice, the most natural starting point isn’t a grand “end-to-end” overhaul, but a focused wedge.

Many enterprise buyers point to order confirmation automation as the obvious first step: an agent that parses confirmations, checks them against purchase orders, and highlights exceptions before they become delivery issues. Once that’s in place and trusted, it’s a short path into adjacent areas like automating RFQs, reconciling invoices, enriching risk intelligence, or continuously monitoring supplier performance. Each new use case builds on the same core capabilities: understanding messy inputs, grounding them in master data, and handing clean outputs back into the ERP.

Another particularly powerful wedge is the long tail of spend that everyone knows is painful but no one has time to optimize. In many organizations, ~80% of suppliers or SKUs sit in this long tail. It’s small dollar amounts on a per-item basis, but enormous in aggregate. It’s where process breaks and data quality decays because no one has the time to focus here. It’s also where money is often left on the table because negotiations rarely happen or, if they do, are driven by gut feeling. If an AI system can bring structure and bottom-up calculations into this chaos by suggesting alternative suppliers, highlighting inconsistent pricing, or bundling fragmented purchases into meaningful negotiation events, it can quickly earn trust. Once procurement sees that the tool can handle the messy tail, they become more open to letting it influence larger, more strategic categories.

As with much of AI technology, the key is to minimize lift, integrate into existing systems, and create fast, visible value. Marcell strongly recommends focusing less on pitching the multi-year vision and more on the first 30-90 days: “how quickly the product can be connected, how much manual work it eliminates, and how easy it is for a Chief Procurement Officer to explain the ROI to their CFO.”

The Integration Imperative: SAP, Oracle, and Master Data

Most large organizations already run on one or two dominant ERP backbones: “in Europe it is definitely SAP. In the US it is very often Oracle,” said Marcell. Procurement tools that pretend those systems don’t exist rarely survive. The winning pattern is not “rip and replace,” but “plug in and enhance.” That means speaking the native language of the ERP: understanding how material master data is structured, how vendor master data is maintained (or neglected), and how transactions flow from requisition to invoice.

Becoming part of the master data orbit is what Marcell sees it as the #1 requirement for getting in and staying in the enterprise. “You need to have APIs… and know how the master data works. That’s the main learning to make the solution sticky and an integrated part of the procurement IT landscape.”

Master data will never be “fixed” through one-off cleansing projects. By the time a team finishes cleaning a dataset, the data is already outdated whether due to new suppliers getting added, materials being updated, plants reorganizing, or what have you. The only sustainable answer is a system that continuously reconciles and enriches data in the background. AI is uniquely suited to this job: it can infer missing attributes, cluster similar items, flag duplicates, and keep supplier records consistent as new information arrives.

What Founders Should Focus On

As both an operator and an active startup investor, Marcell recommended 4 areas that builders should focus on:

1. Focus relentlessly on customers who engage quickly: “If someone is not responding, forget it. Put them to the end of the list.” Procurement cycles are long enough that you can’t afford to waste cycles on the uninterested.

2. Seek fast, scalable wins: A procurement team adopting a new tool is taking a risk. Your job is to minimize that risk and accelerate time-to-value.

3. Build around real, daily pain, not abstract transformation: Tools that automate repetitive workflows, interpret messy data, or cleanly hand back structured outputs will win long before grand “end-to-end” platforms.

4. Invest early in integration and master data intelligence: It’s the difference between “a pilot” and “part of the enterprise.”

Marcell also cautioned against “the over-engineered way” of building products for this space. The temptation, especially for technically strong teams, is to design elaborate, comprehensive platforms that try to solve every possible workflow before really nailing one. In his experience, the winning founders do the opposite. They focus on a narrow, painful slice of the journey, integrate deeply enough to make that slice feel magical, and then scale from there.

A recurring theme in our conversation was how many procurement decisions are still driven by history and instinct rather than real-time intelligence. Category managers lean on what worked last year, or on rules-of-thumb built over decades, even when the underlying cost structures and supplier landscapes have shifted. But with AI, that can finally change.

For me, this feels like a full-circle moment. When I think back to my own experience negotiating with procurement leaders – people willing to debate a one-cent difference on a product within a multimillion-dollar contract – it seemed impossible that this function would ever spend money on new technology. But Marcell’s perspective reframed it: procurement isn’t resistant to spending; it’s resistant to spending on tools that don’t demonstrably help them negotiate those pennies.

AI finally gives them that leverage.

—

Note: Procurement is a horizontal category, not a vertical one. But the themes in this conversation carry across both and the patterns Marcell highlights apply just as much to vertical AI startups as they do to broad, horizontal platforms.