An operating system for supply chain

40% of large companies say they've scaled AI agents. Only 6% see it in EBIT. McKinsey, BCG and PwC all land in the same place: the next gain comes from changing how decisions get made, not another tool.

40% of large companies say they have scaled AI agents. Only 6% say AI delivers at least 5% of EBIT. That’s McKinsey’s 2026 State of AI, and the gap has barely moved in a year. In supply chain, BCG finds the same pattern: the highest AI adoption of any function at 44%, yet only about 30% report measurable value in planning.

Adoption is no longer the problem. How the enterprise runs is.

Three reports from the past few months reach the same conclusion from different directions. McKinsey’s QuantumBlack team on the AI operating system. BCG on AI’s new mandate in supply chains. PwC on how AI is turning plans into decisions. One argument runs through all three. The next gain won’t come from another tool. It will come from changing how decisions get made and executed.

1. From reporting the past to deciding what’s next

McKinsey describes the AI operating system as a “second brain” for the enterprise. It holds business context, reasons across functions, recommends or executes decisions, and learns from the outcomes.

I ran an analytics team in one of my earlier roles. We built good reports and good dashboards, and I came to a simple conclusion. Even when the data was only a day old, we were reporting on history. The real work started after the report: someone had to spot the issue, work out what it meant, agree the trade-off with other functions, and make the change in another system. We rarely went back to check whether it worked.

Reporting still matters. It’s how you keep score. But the value sits in the decisions ahead of you, understood clearly enough to act on. That’s the shift an operating system makes.

2. Supply chain is where it lands first

Supply chain makes thousands of high-frequency, cross-functional trade-offs every day. Most are limited by analyst hours, not by data. BCG calls this decision density, and their advice is to start where it meets value: inventory allocation, supplier rescheduling, disruption response, transport against service. These are the decisions that wait for the next meeting today.

3. From process flow to decision flow

PwC puts it well: planning organizations need to stop owning processes and start owning decisions. BCG shows what that does to a planner’s week. Today about 70% of planning time goes to running the machine, 25% to resolving exceptions and 5% to improving the system. By 2030 they expect the weight to move toward resolving and designing. The planner stops being an integrator and becomes the steward of decision quality.

4. It’s still about people

BCG attributes 70% of the challenge to people, organization and process, 20% to technology and 10% to algorithms. 88% of respondents cite silos. BCG’s work with Thomson Reuters makes the same point from the inside: value showed up where people adopted the tools and trusted the outputs, not where tools were simply available.

PwC finds only 37% of operations leaders are comfortable letting agents run full end-to-end processes. That’s not a technology gap. It’s a trust gap. Trust comes from seeing the reasoning, being able to challenge it, and staying in control.

What an operating system for supply chain has to do

Read together, the reports set a clear bar. Here’s how we built Optaro against it.

Sit on what you already have. BCG recommends a hybrid: keep your planning systems as the system of record and add the orchestration layer on top. McKinsey says build on the technology you’ve already paid for. Optaro connects to ERP, planning, WMS, TMS, procurement and the spreadsheets that actually run the business. It uses the same data your reports already draw on, so that investment carries forward.

Decide, then act. Not dashboards. Decisions. What to buy, hold, make, move or expedite, and why.

Show the work. BCG’s agent explanation scorecard sets out the recommendation, the why, the trade-offs, the confidence and the data sources. Every Optaro decision carries its value and its reasoning. Take one SKU at one location: raise safety stock to lift service from 90% to 98%, and you see roughly $1.7M a year of net value, every number traced to its driver, plus the ~$0.6M a smaller step would leave on the table. (Illustrative.)

Act inside guardrails. The AI recommends. A person approves. Optaro writes back to your systems. Decision rights stay with your people.

Learn and compound. Every decision the system captures sharpens the next, so early movers build a lead late movers can’t easily close. Every accept and reject in Optaro trains our Supply Chain Language Model.

Count its own cost. McKinsey notes token unit costs have fallen more than 95% while AI spend at many companies has more than tripled. Our TokenOps routes each task to the cheapest model that can do it well.

The mid-market question

McKinsey’s examples are large companies with large teams. One manufacturer connected about 100 data sources and had its first phase running in around 16 weeks. That’s real progress for a big enterprise. Most mid-market supply chains don’t have the team, the budget or the time for that build. They still deserve the operating system.

That’s the gap we built Optaro to close. Pre-trained supply chain agents, on your own data, live in 72 hours, with no SI build. Start with three decisions. Expand from there.

McKinsey’s advice to CEOs is to manage by intervention, not by calendar. In supply chain, that means fewer meetings about last month and more decisions about next week, made with the reasoning in plain sight.

If you’d like to see that on your own numbers, message me. We’ll have it running in 72 hours.

Learn more at https://optaro.ai/ and read the full series on our blog.

Sources: McKinsey/QuantumBlack, The AI operating system (Oct 2026); BCG, From Copilots to Enterprise Transformation: AI’s New Mandate in Supply Chains (May 2026); PwC, From plans to decisions (Sept 2026); BCG and Thomson Reuters, Why workforce readiness is critical to AI transformation value.

*Optaro value figures are potential outcomes, not guarantees.