You Can’t Automate Chaos: What AI in Procurement Actually Looks Like From the Inside.
At RAUS Global, we spend our time tracking structural shifts in the marketing and procurement economy — the kind that look like technology stories on the surface but turn out to be business model stories underneath. So when KonnectHouse convened its Agentic AI event in New York on March 10, we were in the room not to sell anything, but to listen. What we heard from procurement leaders at Pfizer, Hershey, DXC, and Ramp, sitting alongside builders, vendors, and technologists — was one of the most honest industry conversations we have witnessed in (AI) years. Not the polished version. The real one.
There is a version of the AI in procurement story that gets told on conference stages and in vendor decks. It involves seamless automation, dramatic cost savings, and procurement teams liberated from administrative misery to focus on strategy. It is a compelling story. It is also, for most organizations, not the story that is actually unfolding.
The real story is messier, more instructive, and ultimately more hopeful. It was on full display throughout the day, where procurement leaders did something rare: they told the truth about where they actually are.
The honest summary? Most organizations are somewhere between stumbling and finding their footing. The ones making real progress share a common insight that has nothing to do with which AI model they chose or how much they spent on software. They figured out that AI readiness is a people and data problem first, and a technology problem a distant second.
The Chaos Problem
Arie Barendrecht, General Manager at Omnea, opened the day with a provocation dressed up as a framework. He invoked the old I Love Lucy chocolate factory scene: Lucy and Ethel trying to keep up as the conveyor belt speeds up — as a metaphor for what happens when organizations bolt AI onto broken processes. “If processes are fragmented,” he told the room, “automation simply makes the chaos go fast.” It landed because everyone in the room had lived it.
Gartner research cited during the session suggests that roughly three quarters of enterprise AI pilots never make it to production. Barendrecht argued this is not a capability failure. The models are genuinely powerful. The failure is organizational.
Teams are being pressured from the top to “do AI,” so they stand up a pilot, drop it on top of fragmented workflows, siloed data, and manual handoffs, and wonder why it doesn’t deliver the ROI they promised their CFO. “The truth about agentic procurement,” he said, “is that it’s actually about data. If you don’t capture the right data, you will simply scale the challenges you already have.”
That framing set the tone for everything that followed.
From our vantage point at RAUS Global, where we work with brands and procurement teams navigating similar transitions every day, what made this day unusual was the candor. Industry events tend toward the aspirational. This one kept pulling back toward the practical, sometimes uncomfortably so. Peer procurement leaders were not comparing polished success stories. They were comparing notes on where they had stumbled, what they had learned, and what they still had not figured out. That kind of conversation is rare. It is also the only kind that actually moves the industry forward.
What Stumbling Actually Looks Like
The most valuable moments of the day came when practitioners described their first year with AI not as a triumph, but as a controlled stumble.
Subs Tripathy, Head of Corporate Services Procurement at Pfizer, was refreshingly direct. When Copilot was rolled out across the company two years ago, the procurement team moved fast — and immediately ran into the wall that Barendrecht had described. “We did not know how to load all the SOWs into Copilot so that they could access it,” she said. “When we tried, it just blew up.” The data was there. The governance wasn’t. The foundation hadn’t been built.
Victor, who leads the Procurement Center of Excellence at Hershey, described a similar reckoning. His team’s first priority wasn’t deploying AI tools. It was building what he called a “digital data tech roadmap” — getting category managers to take ownership of data quality, connecting spend data with contract and risk data, and establishing governance before attempting anything agentic. “You cannot really trust only a center of excellence to create reliable spend data,” he said. “You also need the feedback of the category managers.”
Both stories share a structure: enthusiasm, collision with data reality, recalibration, slow progress toward a genuine foundation. Neither organization is embarrassed by the stumble. They treat it as the cost of learning something that vendor demos don’t teach.
We recognize this pattern at RAUS Global. We see it with marketing procurement teams wrestling with AI-driven agency workflows, with brands trying to govern AI use across creative and media supply chains, and with procurement leaders being asked by their CEOs to have an AI strategy before they have an AI foundation. The pressure arrives before the readiness does. What separates the organizations making progress is not that they avoided the stumble. It is that they used it to build something more durable on the other side.
The Change Management Nobody Budgets For
If data is the first obstacle, culture is the second — and in some ways the harder one. Anders, Chief Procurement Officer at DXC Technology, a 120,000-person company formed from the merger of CSC and HPE’s services business, described a procurement function that had historically been perceived as a roadblock. Slow turnarounds. Last-minute requests taking a long time to get response. The organizational reputation that procurement teams everywhere recognize and quietly (well, loudly) resent.
AI, in his framing, offered procurement something it had never quite had before: the chance to prove it could move at the speed of the business. But getting there required more than deploying tools. DXC ran what they called an “AI challenge” — a week-long initiative where teams competed to build and deploy AI use cases, scored on a points system. Creating a prompt earned one point. Building an agent earned five. Teams competed on leaderboards. Leaders showed up to watch presentations. “Seeing your colleagues on a leaderboard, seeing real AI output, seeing our COO excited,” Anders said, “created a fuss and an excitement about something I haven’t seen before in the time I’ve been with DXC.”
The most telling result wasn’t a specific efficiency metric. It was a tail spend negotiation agent that one of his team members built without being asked — simply because the environment made it feel possible. Output on tail spend negotiations improved by roughly 80%, measured by time to review, time to respond to suppliers, and time to close. A human remained in the loop throughout. Nobody’s job disappeared. The work got better.
Subs described a similar gamification approach at Pfizer, segmenting the procurement team into three categories: 1) users (roughly 70%), 2) translators who can convert business needs into AI specifications (around 20%), and 3) developers (about 10%). Rather than treating AI adoption as a mandate, they gave people a menu — three doors, each with its own learning path — and let momentum build from the inside out.
Build Versus Buy: The Question That isn’t Easy to Resolve
One of the day’s most honest moments came from Mateb Alfayez, Head of Procurement at Cribl, who delivered a live demo of AI agents he had built himself using Claude Code — a contract redlining assistant, a vendor risk scout, a procurement dashboard — and then freely admitted that the build-versus-buy question has no clean answer. “I don’t know the answer,” he said. “I think that’s the most honest answer I could come up with.”
What he offered instead was a framework: if a task is internal, repetitive, and doesn’t require external data or ongoing maintenance, build it yourself. If it requires third-party data, complex integrations, or scales across the enterprise, the math usually favors buying. But even that heuristic breaks down depending on company size, engineering resources, risk tolerance, and how fast the tools are evolving.
Jake Heenehan, formerly a procurement leader and now founder of Cyrus AI, made the case that the cost of building small, targeted tools has dropped so dramatically that procurement leaders have almost no excuse not to experiment. “The cost of actually building something like this for small, manual workflows is nothing,” he said. He described building a contract negotiation email tool at his previous company — feeding a contract to an AI agent, getting back a personalized email in his voice following his standard playbook — and rolling it to a team of five. “We were sending 30 emails and it takes me five minutes per email. That’s a massive throughput.”
Aubrey Zimmerman, who runs cloud infrastructure procurement at Ramp, added the caveat that gets lost in the excitement: “The best software isn’t just built, it’s cultivated. You’re going to have to continuously train it as your policies develop.” The build decision is not a one-time call. It is a maintenance commitment.
What the Organizations Getting It Right Are Actually Doing
Across the sessions, a set of behaviors separated the teams making real progress from the ones still spinning.
They started with pain points, not technology. Johannes, a lead engineer at Lio, described working with a German manufacturing company , where the starting question was never “how do we implement AI?” It was “what problems do procurement users have today, and how do we solve them as fast as possible?”The first wins were modest: guided buying, automated PR quality checks, order confirmation processing. Those wins built credibility that made the next use case easier to fund and adopt. The client treated data as the real product.
Victor’s team at Hershey built what he called “procurement data products”, integrated, reliable data sets connecting spend, contracts, and supplier risk, before attempting anything agentic. The insight was that AI is only as trustworthy as the data it runs on. Cleaning that up first is unglamorous work that never appears in a vendor demo, but it is the difference between an AI that produces consistent results and one that produces confident-sounding noise.
They kept humans in the loop — deliberately. Anders, Subs, and Victor all described the same governance instinct: agents earn autonomy gradually, based on demonstrated accuracy. Validation tasks first. Drafting with human approval. Only then, selective automation. “If you would not allow a junior employee to update contracts without oversight,” Danielle McQuiston from Zip observed, “you should not allow a digital one to do so either.” They documented the wins loudly. Mateb at Cribl made the point that AI adoption inside a company is also a branding exercise. When procurement is seen as AI-forward, helping colleagues solve problems, not just managing compliance, the internal reputation shifts. “If your business partners view you as an AI-forward team,” he said, “you become an enabler. That’s how you maintain your brand.”
The Gap That Still Exists
Cassye Cook Provost, a former CPO at Paylocity who now runs her own consultancy, opened the AI literacy session with a simple audience poll. How many people in the room had a formal AI literacy program in place? Two or three hands went up. How many had a change management plan around AI? About the same. For a room full of procurement professionals at a conference specifically about agentic AI, it was a striking result. The tools are advancing. The organizational infrastructure to use them is lagging badly.
That gap — between personal experimentation and enterprise readiness, between pilot excitement and provable ROI — is where most procurement teams currently live. Closing it requires less focus on which AI to buy and more focus on the foundational work that makes AI trustworthy: clean data, honest governance, cultural permission to experiment, and the patience to build momentum from small wins rather than betting everything on a transformation program.
A Closing Reflection
From where we sit at RAUS Global, working at the intersection of marketing, procurement, and commercial strategy, the day reinforced something we see with clients regularly. The organizations that will win the AI transition are not necessarily the ones with the biggest technology budgets. They are the ones with the clearest picture of their own operational reality — honest about the mess, disciplined about the foundation, and patient enough to build momentum from small wins rather than betting everything on transformation.
That is not a technology lesson. It is a leadership one.
The organizations sharing their stories at KonnectHouse had figured that out. Not because they were smarter than anyone else in the room. Because they had stumbled, recalibrated, and kept going. And in a room full of peer procurement leaders, vendors, and technologists all trying to make sense of the same seismic shift, watching that honesty surface and spread felt like exactly the kind of industry conversation that moves things forward.