Reply to BCG: The Redesign Is Right. Here Is What It Costs You.

BCG is right that AI in procurement requires workflow redesign, but a CPO still has to price that redesign: where freed capacity lands, what happens to the suite contract wrapped around the old workflow, and who may let an agent act without asking.

BCG is right about the redesign

Wolfgang Schnellbächer and his BCG coauthors make a point procurement leaders need to hear: stop placing AI inside the same people-centric workflow and start “Redesigning around workflows, not tools.” Their enterprise-level recommendation matters because this is more than a technology purchase: redesign source-to-pay around agents and have the CEO or COO lead the change so the procurement architecture serves the whole enterprise.

The numbers need more discipline than the thesis. The piece says a redesigned model can free up buyer capacity by 60%, explicitly labeling that figure as BCG modeling, and it estimates 70% of the effort needs to be spent on people, organization, and process redesign. Those are modeled capacity and estimated effort, not observed procurement outcomes. The piece also presents three outcome ranges as observations, while the pinned public passage gives no sample, comparison group, or method next to them. I would use the figures to frame questions, not to build a business case.

Freed capacity needs somewhere to go

A separate 2026 BCG technology-procurement study surveyed more than 200 CIOs, procurement leaders, and specialized buyers across North America, Europe, and Asia-Pacific. It says organizations combining deployment with capability building, process redesign, and governance reported stronger outcomes than those focused on technology alone. That is a useful direction and a self-reported association from a consulting-firm study, not causal proof that an org-chart change creates value.

Independent evidence on where the capacity lands is thinner than the confidence of most AI plans. Gartner says many adopters are seeing uneven ROI or falling short of expectations, while peer-reviewed procurement research proposes hybrid models that use human expertise to validate AI outputs and make adjustments. Neither source tells a CPO which roles disappear, which category owners gain scope, or which budget pays for the central agent capability. Those are leadership decisions, which is exactly why a capacity percentage by itself is not a workforce plan.

• Name the central owner of agent standards, controls, reusable workflows, and evidence retention.
• Keep commercial strategy, specifications, supplier selection, awards, and relationships with named category and cross-functional owners.
• Decide before the pilot whether released time becomes broader category scope, deeper supplier work, a smaller team over time, or simply unused capacity.

The legacy-suite renewal belongs in the plan

BCG recommends module-by-module upgrades that gradually decouple and replace legacy systems, while its sourcing-agent example sends work to people only when exceptions or strategic tradeoffs arise. Operationally, those ideas belong together: changing which system executes a step also changes the seat, module, integration, data, support, and termination terms the buyer may still be carrying from the old workflow.

Here the public record goes quiet. This research run found list-price commentary and generic comparisons, but no evidence that could tell a specific enterprise what its renewal, unbundling, exit, or transition would cost. I am treating that silence as a boundary, not filling it with a convenient percentage. My position is that the renewal sitting in the inbox must be priced inside the operating-model decision, because the commercial commitment can outlive the workflow it was bought to support.

• Which modules and integrations still carry value after the workflow is redesigned?
• Which dates, notice periods, data-export rights, transition duties, and dependencies constrain the sequence?
• What can be retired, what must coexist, and what evidence would justify each decision?
• Who owns the negotiation when the technology roadmap and procurement operating model disagree?

Human on exceptions needs an actual document

The evidence supports a hybrid boundary, but it does not write one for you. Guida and colleagues describe AI spend classification with human validation and adjustment. For systems legally classified as high-risk, EU Article 14 says they must be effectively overseen by natural persons while in use, including the authority not to use the system in a particular situation. I am not claiming ordinary procurement tools fall into that legal class. The useful operating lesson is narrower: oversight needs visibility, authority, and a stop path.

A Kinaxis-sponsored IDC InfoBrief, reported by DC Velocity, makes the governance gap visible but should be discounted as sponsored, self-reported supply-chain evidence rather than procurement proof. It says only 1 in 8 organizations had governance fully embedded for a wide AI rollout, in a survey of more than 2,000 supply-chain leaders across nine markets. The same coverage says 52% cited trust in AI-driven decisions as a top barrier, which is why the decision-rights record matters: it gives people a concrete reason to trust how the operating model will behave.

For every step, write whether the agent may execute, recommend, or must wait. Then name the permitted data, policy, threshold, evidence, accountable owner, required reviewer, override, fallback, and stop authority. Awards, negotiation strategy, risk acceptance, and material supplier commitments stay with named accountable people under existing delegation-of-authority and segregation-of-duties rules; agents may execute only pre-authorized, bounded transactions with explicit thresholds, evidence, override, and stop controls.

What changes when the operating model is designed for agents?

What I would do on Monday

1. Choose one consequential workflow. Use historical replay or an approved shadow run, and pick work where a recommendation can be evaluated without giving an agent uncontrolled supplier or financial authority.
2. Map the current handoffs and the intended decisions. Separate copying, checking, analysis, recommendation, approval, commitment, and relationship ownership.
3. Put the contract calendar beside the workflow map. Record modules, integrations, renewal and notice dates, data-export rights, dependencies, and owners without guessing at an exit price.
4. Write the decision-rights record. For each step, define execute, recommend, approve, override, stop, evidence, and fallback.
5. Name the destination of any released capacity before measuring it. Broader scope, deeper supplier work, resilience, negotiation, governance, or eventual structural reduction are different choices and need different baselines.

Our guide to AI in procurement maps the execution-versus-decision boundary across the source-to-pay cycle. For this redesign, I would add the org destination and contract calendar to that map before asking for another pilot budget. On Monday, I would be staring at the org chart, the decision-rights page, and the renewal sitting in my inbox for a workflow I am about to redesign. Which of those three is actually slowing you down?