AI Can Draft Your Category Strategy in 2 Minutes. That’s Not the Problem. That’s the Problem.

Andrew Speck leads Coupa’s Global Category Management Center of Excellence, an enterprise team of category management practitioners. His previous roles include the founding leader of Cirtuo Americas and global category management at Molson Coors Beverage Company.

I’ve had this conversation with dozens of procurement leaders this year. Ask any Chief Procurement Officer what their top strategic priority is this year, and you will almost certainly hear a variation of the same goal: accelerating enterprise value through AI.

Naturally, leaders are asking a logical follow-up question: If generative AI models can write text so effortlessly, can we prompt our way into building enterprise category strategies?

It is the right question to ask. But it reveals a fundamental misunderstanding of what a category strategy actually is.

Generative AI can draft a category strategy document in about two minutes. That isn’t the breakthrough. That is the trap.

A fast output creates the illusion of work done. A generic AI assistant serves up a polished, five-page strategy filled with credible industry vocabulary. It looks sharp, reads persuasively, and allows an organization to tick the “strategy complete” box. But six months later, those documents sit forgotten in shared drives while category managers make sourcing decisions in spreadsheets, completely detached from the draft.

Speed of drafting and success of execution move in opposite directions. Here’s why, and what I think companies should do about it instead.

The improv trap

A general-purpose AI assistant is a talented improv actor. Ask it for a category strategy today, and it will deliver a smooth, confident narrative. Now ask the same question next week. It’ll serve up a different strategy. Then ask a colleague to run the same prompt, and you’ll get a third version. Each sounds credible. None of them are aligned with each other.

Strategy can’t work that way. A category strategy has to apply one proven methodology the same way every time, so a plastics strategy and a logistics strategy meet the same analytical bar regardless of who wrote them, or the category in focus. Without that consistency, a plan is unstructured text dressed up as strategy.

The deeper issue is access, not intelligence. A general-purpose model has no connection to your ERP data, your supplier scorecards, or the business constraints your organization actually operates under. It can’t see what it was never given. So it fills those gaps with guesses that sound specific. No amount of clever prompting substitutes for data the model was never connected to.

What a guardrailed AI framework does differently

Should you ban AI from strategy development work? Absolutely not. What sourcing and procurement teams need to do is give AI a methodology to follow, and their company’s own data to follow it with, instead of a blank page to improvise on.

A domain-specific, guardrailed framework, such as Coupa Category Strategy (formerly Cirtuo) powered by Coupa Navi™ AI agents, walks a category manager through the same analytical stages every time: business requirements, category and spend segmentation, the Kraljic Matrix for negotiation posture, Porter’s Five Forces for market dynamics, and supplier preferencing to understand how suppliers view your business. The rigor is built into the stage, not left to the prompt. And the AI is informed by your actual, verified spend data and supplier history instead of guessing at it.

A guardrail framework also tailors the output without changing the underlying data. The Coupa Navi™ Category Strategy Storyteller Agent, for example, turns the same strategic inputs into a concise narrative a business leader can read in five minutes. The Coupa Navi™ Category Strategy Consultant Agent turns those same inputs into multi-constraint evaluation rules a team can act on. One source of truth, several audiences.

Friction was the point

Procurement teams often learn the hard way that stakeholders who don’t weigh in on a plan won’t commit to it later. A strategy built in isolation is just that, a plan that has one voice and no one will follow.

The debate between procurement, operations, engineering, and finance has always looked like friction, like the slow part everyone wants to automate away. But the debate is the very necessary human in the loop part. It’s the mechanism that builds the plan’s credibility. Let’s say a plant manager pushes back on a proposed supplier consolidation, and the category team has to sit down with that manager and defend the trade-off out loud. That conversation has the potential to turn a document into a commitment with people and financial resources behind it. Skip the debate, and you’ve automated the part that was actually doing the work.

A strategy nobody argued about is a strategy nobody runs

If a category manager can’t defend a strategy when a VP of Finance questions it, the problem isn’t the draft itself. The real issue is the lack of a stress test or any kind of preparation to counter real-world objections when they surface and a decision needs to be made yesterday.

AI should own the heavy lifting: spend classification, market indexing, first-pass strategic recommendations. Humans own the argument: weighing trade-offs, defending the plan under pushback, and locking in alignment across teams at every step of the journey who each want something different from the same category.

The stress-test before you sign off

Before any AI-assisted category plan gets executive sign-off, run it through four checks.

• Can your suppliers pass validation in time? If the plan shifts volume to a lower-cost alternative, that supplier still has to clear quality validation and engineering sign-off within whatever window a disruption gives you. A cheaper supplier who can’t clear approvals fast enough isn’t a backup plan.
• Who pays during the transition? Someone absorbs the carrying cost of extra inventory, the tariff shift, or the expediting fees while volume moves from one supplier to another. Name that someone before you sign, not after.
• Did operations, quality, and finance actually agree? Not reviewed the deck. Agreed, specifically, on who decides what, what the target cost is, and how much risk the team will tolerate. A strategy with no named decision-maker for a given trade-off will stall the first time that trade-off comes up in real life.
• Are your other categories pulling the same direction? A win on total cost of ownership that creates a new supply-assurance risk, or blows through an ESG target, isn’t a win. Check that adjacent categories are reinforcing the plan, not working against it.

Closing the loop: Turning static strategy into live sourcing constraints

Sourcing answers one question: How do we buy this well? Category management and category strategies answers a different one: What’s our approach, and why? Most category strategies fail because they answer “why” once in a slide deck that sits disconnected from day-to-day RFQ and contract work. Six months later, nobody opens the deck, and sourcing decisions are made without it.

A closed-loop system connects sourcing and category strategy. Spend and market data feed a strategy stage, where the system sets Kraljic positioning, value levers, and objectives. Those strategic rules flow directly into sourcing optimization as mathematical constraints — not as recommendations someone might remember to check. A dual-sourcing requirement or an ESG allocation minimum set during strategy becomes a binding rule inside your next RFQ. A general-purpose model can hand you a well-written plan, but it can’t make that plan an operational rule. That’s the part no prompt replaces.

Once a contract is awarded, the resulting P&L savings and performance data flow back into the strategy, so the next cycle starts from what actually happened rather than from a stale assumption.

If you’ve read this far, I hope our paths will cross someday soon. Maybe you’ll show me your AI-drafted category strategy, and maybe it was drafted in under two minutes. I’m sure the deck will look good. But I won’t forget to ask you whether your organization can execute that strategy with confidence once the deck closes.