"I see procurement becoming more of a human-focused role as opposed to a transaction-focused role." - Shantanu Verma, Partner, Operations | Leader, North America Procurement Practice, Roland Berger
Over the last few years, procurement teams have earned more influence the hard way, through spiraling freight costs and a post-COVID commodity cost crisis. AI now promises to automate much of the tactical work that fills a buyer's week, raising a practical question for every CPO: what will their people do with the hours they get back?
In this episode, I speak with Shantanu Verma, Partner in Operations and Leader of the North America Procurement Practice at Roland Berger. Shantanu has spent 12 years in procurement, and he shares where he sees the function on its AI journey, why so many organizations hesitate, and what CPOs can start doing now.
A Crisis Becomes a Ladder
"I was speaking to a CPO at a fairly large e-commerce retailer, and he used a term that he borrowed from Game of Thrones, funnily enough. He said, 'A crisis becomes a ladder for procurement.'"
Shantanu most often sees executive teams change their view of procurement because of outside pressure. Sometimes that pressure is an EBITDA target after a private equity acquisition. Other times it comes from a supply or cost shock like the ones that followed COVID.
Getting invited in comes first. The harder test is whether the team remains a strategic business partner once the panic fades, or goes back to being the cost-saving SWAT team until the next crisis arrives.
Start Indirect AI Work With the Data
"I think the starting point in indirects always has to be data, because there does tend to be a little bit of opacity, a little bit of muddiness as far as data is concerned."
On the indirect side, Shantanu often starts with supplier-level payment information and works up from there, cleansing and structuring data so teams know more about what they buy. Direct materials data tends to be better structured, so the opportunity there shifts toward reducing complexity and aggregating demand.
He also thinks procurement is better placed than many functions to gain productivity from AI, because so much of their work is systematic. He points to reviewing contracts, tracking expiry dates, and rolling out a universal set of terms to suppliers.
Complexity Gets Designed In
"Complexity, particularly when it gets injected in the design process, is a major cost driver for a lot of industrial organizations."
Shantanu cites Whirlpool, which went on what he calls a complexity blitz, deliberately reusing parts across new designs. The stated value was something like half a billion dollars. Support costs for old, low-volume parts add to the bill when nobody reviews designs systematically up front.
He describes the goal as designing for sourceability, and he admits wholesale adoption hasn't happened yet: "Unfortunately, I feel like procurement's still a little bit of an afterthought." The encouraging sign is that topics like rare earths in battery design now come up in management conversations. That gives procurement teams an opening to join the design process earlier.
Where the AI Hesitation Comes From
"It's still a little bit of the Wild West. I think every company is adopting its own guidelines around governance, and I think part of it is driven by what solution they're marrying themselves to on an enterprise level."
Shantanu traces most AI hesitation to two places. The first is concern about liability if something goes wrong. The second is IT teams that want a control structure in place before tools reach the business, mostly because of data loss risk.
In the meantime, he sees people using ChatGPT or Gemini for quick supplier research or for thinking through requirements, "more as a teacher and a research buddy." He also worries that process knowledge, and how organizations think, may be ending up in the public domain, where it could train competitors.
Better Data Still Needs Good Inputs
"Having played enough with AI at this point, what I do realize is the output is often only as good as the set of inputs that the user's providing it."
Shantanu expects AI to improve should-cost models, because teams can compile data from more sources and closer to real time. He doesn't think anyone can hand should-cost modeling over to AI yet.
On the old information gap between buyers and sellers, he sees more buyers walking into negotiations aware of commodity movements and what those movements do to the SKUs they buy. In his view, that gain comes mostly from better buying practices, with AI playing a smaller part so far.
Measure the Tactical Hours First
"I think the first thing you need to do is start figuring out, from an organization standpoint, how much of your time you're spending on the tactical and transactional."
Shantanu expects much of the tactical work, even negotiating basic contract terms, to become automated. What remains is a more human-focused role, helping shape cost and design roadmaps inside the business and building deeper collaboration with suppliers.
He laid out a sequence from a workshop he ran with one CPO's team. Measure where the hours go, decide what size and skillset you need for relationship work, and then automate the transactional tasks. He wants CPOs to make that decision actively and on their own terms.
Links:

