72% are exploring AI, 0 can name a use case. Why Defer is the most profitable call in indirect procurement.

In January 2025, BCG published a finding that deserves to circulate in every indirect-procurement committee: of 32 retail CPOs surveyed, 72% say they are exploring or integrating AI. Not one respondent could name a concrete use case already deployed and generating value.

The BCG verbatim is blunt: "when asked to elaborate on use cases, no respondents could articulate areas where they are using AI yet". Translation: 72% of explorers, 0 explorer who arrived anywhere.

That gap should be a signal. It is, just not the one vendors would like.

Why this matters now

For 18 months, every IT and digital procurement committee has received the same request: "we want to deploy AI agents on [finance, HR, support, sourcing, legal]". SAP Ariba ships its agentic module. Coupa pushes Coupa AI. Oracle stacks Procurement Copilot. Zip, Tropic and Ramp tell the same story in other words. And IT, caught between business pressure and the fear of falling behind, hesitates to sign while hesitating to wait.

Against that backdrop, McKinsey published in 2024 a decision framework for enterprise agentic platforms, built around a simple tree: should you Build, Partner, Buy, or Defer. Four branches. The first three are familiar to anyone who has run a classic make-or-buy. The fourth, Defer, never shows up in the committees I see.

It is exactly the one you need today.

What the McKinsey framework says, in four questions

The framework (shown above) runs on a decision-tree logic you can treat as a checklist to enforce before any agentic RFP on indirect procurement.

Question 1. Is there a strategic reason to build this capability? Is this use case in your top 3 levers of competitive differentiation? If YES, move to question 2 (Partner or Build). If NO, drop down to question 3.

Question 2. Can we partner while guaranteeing our requirements? Evolution timelines, a contract that prioritizes our needs, oversight on fine-tuning with your data, reversibility of weights and prompts. If YES, it is Partner. If NO, it is Build.

Question 3. Is there a fit-for-purpose market solution we can integrate while keeping control, transparency and roadmap influence? If YES, it is Buy, favoring open-source where possible. If NO, move to question 4.

Question 4. Is the impact of deferring larger than the TCO of building or partnering? If YES, waiting costs you more than acting: go back up to Partner or Build. If NO, it is Defer.

On paper, it is a classic make-or-buy applied to a new object. In the field, it is a device that legitimizes patience as an active decision. And patience is exactly what is missing in today's AI procurement committees.

The Defer lever, read by a practitioner

Let me be direct. Of the 30 to 40 candidate agentic use cases on indirect S2P (auto-categorizing spend, generating RFQs, contract analysis, invoice processing, supplier-risk alerts, panel scoring, category suggestions for maverick spend, and so on), between 60% and 80% are not differentiating for your company. They are cognitive commodities, already covered or being covered by every P2P vendor on the market.

For those cases, question 1 closes on NO. So you drop to question 3: is the market solution truly fit-for-purpose, with the control, transparency and roadmap influence you require? In June 2026, for most agentic P2P modules, the honest answer is NO: still-opaque building blocks, proprietary lock-in, a roadmap you don't steer. So you move to question 4.

Question 4: is the impact of deferring larger than the TCO of building or partnering over 12 months? Honest answer, in June 2026: almost always no. For three cumulative reasons.

First reason. The cost of a token has fallen roughly 80% in 18 months on so-called frontier models. What cost $30/M tokens in 2024 on a GPT-4 costs $4 to 5 today on an equivalent. Any agentic solution deployed today on a 2024 model will be structurally more expensive than an equivalent deployment 12 months from now.

Second reason. The multi-agent orchestration layer (LangGraph, AutoGen, CrewAI, n8n, MCP on the Anthropic side) only stabilized in late 2025. A project launched in 2024 paid the entry ticket for a proprietary framework. A project launched in 2026 inherits a mature open-source orchestration layer, with no lock-in cost.

Third reason, the most uncomfortable. 72% of your peers are exploring with no result. If no one around you is capturing a quantifiable competitive edge through AI in indirect procurement, then the opportunity cost of waiting is itself close to zero. You are not missing a train. You are watching a train that hasn't left the station.

In 2026, in indirect procurement, smart money waits. Defer is not surrender: it is the branch you should document first.

The senior counter-objection

A senior CPO will hit you with three classic objections when you walk into committee with the idea of deferring.

"We'll fall behind our competitors." That is the emotional objection, not the factual one. BCG has just proven no one has an articulable edge yet. The question to send back: "which identified competitor are you losing to, and on which precise use case?"

"We need to build AI skills right now, or the team won't be ready." Legitimate in form, wrong in conclusion. Building skills does not require signing a vendor RFP at €200k a year. An in-house PoC on a non-critical case, on an open-source model, costs 5 to 10 times less and teaches the team far more.

"IT is putting pressure on us." That is the real political question. Answering it cleanly requires producing the written decision grid, signed by the sponsor, that documents the Defer. Without a written record, IT pressure wins by default at the next committee.

The 4-column matrix for your next committee

To make Defer operational, I use a four-column matrix that I fill in before every AI arbitration on indirect spend.

Column 1, candidate use case. One row per use case, not per vendor.

Column 2, strategic differentiation (yes / no). A one-sentence answer with justification. If you write "yes" everywhere, you are lying to the committee.

Column 3, 12-month TCO (Build or Buy). Full total cost: license, integration, run, change management, vendor retrofit when the API changes. Not just the license.

Column 4, business impact of a 12-month delay. In euros. Not in "competitiveness" or "agility". In measurable euros: lost savings, uncaptured cost avoidance, quantified service degradation.

When column 4 is lower than column 3, you have a documented Defer. And 60 to 80% of the rows on your list will land in that box, by construction. That is the normal output of the analysis, not a failure of ambition.

What you can do tomorrow morning

Three concrete actions, in this order.

First, list your next five IT/digital committees where an agentic AI decision will land. For each, ask the business sponsor to answer the four McKinsey questions in writing before the committee meets. Not during. Before.

Then, build the four-column matrix for your indirect scope. If you don't have the full list of candidate use cases, ask your current P2P vendor for their 2026-2027 agentic roadmap. You'll get the list effortlessly.

Finally, formalize your position in two paragraphs signed with the procurement sponsor: "we are deferring use cases X, Y, Z for the following reasons, with a re-assessment scheduled at T+12 months". Without that written record, the Defer does not survive the next committee. With it, you have bought 12 months of room to maneuver.

The trap to avoid: Defer is not inaction

The mistake I see most often: confusing Defer with inaction. Deferring a vendor deployment is not the same as doing nothing. It is an active choice that assumes, in return:

• A low-cost in-house PoC on 1 or 2 non-critical use cases, to build the team's practical skills.
• Weekly monitoring of P2P vendor announcements, so you don't miss a real market shift.
• A quarterly review of the four-column matrix, to re-run the decision with updated figures.

Without those three conditions, Defer degenerates into stagnation and you'll lose the next political battle. With them, you hold the most profitable posture on the market today.

The bottom line

BCG says: 72% explore, 0 articulate. McKinsey says: there are four branches to your agentic make-or-buy, not three. The two findings converge on the same conclusion. In 2026, in indirect procurement, Defer is the branch to document first, because it is the one that withstands IT pressure while buying 12 months of better visibility, at lower cost. Build and Buy will stay relevant when a genuinely differentiating use case emerges. For now, smart money waits.

A question for you: of the agentic AI use cases pitched to you over the past six months in indirect procurement, which one would pass McKinsey's question 1? Which would pass question 4? If you don't know, the grid probably hasn't been applied yet.

Want to stress-test your agentic roadmap before signing an RFP? Book 30 minutes here.

I'm Alex Lio. 10+ years in indirect procurement, 7 of them at Amazon, now working alongside my clients on digitalization and AI.

Sources

1. Boston Consulting Group, Indirect Spend in Retail: Procurement Leader Priorities for 2025, January 2025.
1. McKinsey & Company, Rethinking enterprise architecture for the agentic era.
1. QuantumBlack, AI by McKinsey, Creating a future-proof enterprise agentic platform architecture (architecture diagram + Build / Partner / Buy / Defer tree).
1. McKinsey & Company, Indirect procurement: Insource, outsource, or both.