AI Can Think Fast, But You Can’t: Why Direct Materials Hits the Judgment Wall First
McKinsey dropped an article last week that I’ve practically memorized at this point. It’s called The judgment bottleneck: Balancing AI and human speeds in strategy, written by Alex D'Amico, Bruce Delteil, Laura LaBerge, and Matt Banholzer in McKinsey Quarterly on October 15.
Their main point is beautifully straightforward: AI generates analysis at blinding machine speeds, but human brains still move at, well, human speeds. Naturally, the bottleneck moves. Crunching the numbers isn't the slow part anymore; the slow part is the humans standing around trying to figure out what the numbers mean, questioning the underlying assumptions, and taking actual responsibility for the call.
Now, McKinsey was writing about high-flying corporate strategy teams. But as I read it, all I could think about were category managers in food, beverage, and CPG. Because if you want to see this bottleneck in its purest, most chaotic form, look at direct materials first.
Why direct materials feels the pain first
Corporate strategists get to make a few monumental calls a year. Direct materials teams make high-stakes judgment calls before lunch on Tuesdays. Do we buy more cocoa now or play chicken with the market? Do we swallow the resin price hike or fight back? Do we qualify a second packaging vendor or cross our fingers? Every single decision involves serious money and requires reading market trends, supplier politics, and internal plant capacities all at once.
Throw AI into the mix, and suddenly your team is inundated with price signals, weather alerts, supplier risk warnings, and scenario models faster than they can blink. One executive told McKinsey that five weeks of financial modeling now takes under two days. That sounds amazing until you realize your category manager still has the exact same 24 hours in a day, the same endless vendor calls, and the same weekly status meetings.
McKinsey highlighted a company where a junior team generated analysis up to 20 times faster and pumped out five times as many scenarios. The outcome? Executives needed more time to review the work, not less, simply because they couldn't tell which assumptions were buried in the output. I’ve seen this exact movie play out in procurement: more shiny dashboards, more automated alerts, and decisions grind to a halt because nobody actually trusts what's inside the box.
The three ways this blows up (procurement edition)
McKinsey identifies three classic failure modes. Unsurprisingly, each has a perfect direct materials equivalent.
1. Speed over substance. Teams use AI to cut analyst hours, skipping the crucial step where someone with actual gray hair pressure-tests the logic. In procurement, this is how an auto-generated should-cost model lands on a supplier's desk without anyone noticing the energy price assumption was completely absurd.
2. The decision avalanche. Faster analysis triggers an overwhelming flood of decisions to review. A category lead gets bombarded with 40 automated alerts a week and naturally misses the subtle one warning that a crucial tier-2 supplier is about to go under.
3. Starving the next generation of judgment. This is the one that keeps me up at night. I learned this trade by building models by hand, making embarrassing mistakes, and getting grilled by a senior trader on why I was wrong. As McKinsey points out, judgment isn't built by reading finished summaries; it's forged through practice, feedback, and minor scars. If AI builds every model from scratch, junior category managers never get their reps in. In five years, who's going to step up to make the big calls?
McKinsey’s playbook for survival
The piece wraps up with three core recommendations. I agree with all of them, and frankly, they're even more critical in the world of direct materials.
First, preserve senior bandwidth like it's gold dust. Don't dump a 50-page slide deck on leadership; show them the critical "what you'd have to believe" assumptions. Categorize decisions into tiers: routine stuff handled automatically within set guardrails, medium-risk calls led by humans with AI assist, and massive, high-stakes bets owned squarely by executive leaders.
Second, make judgment building intentional. Force AI tools to show their work—formulas, sources, and underlying logic—and insist that team members be able to explain the reasoning, not just quote the answer.
Third, keep accountability crystal clear. Every single major decision needs a human name next to it. McKinsey even suggests deploying "challenger" AI agents to play devil's advocate from different perspectives (like risk or finance) before a proposal ever hits the decision-maker's desk.
My favorite line in the whole article uses a baseball metaphor: the difference between "seeing a fastball coming and being able to hit it." Being a Brit and a rugby fan, I think of it as spotting an opening in the defense versus actually crashing through it. Either way, procurement has spent the last decade getting really good at spotting things. Very few teams have built the actual muscle to act.
How we’re designing Kodiact to fix this
When we founded Kodiact, our mantra was "data is cheap, judgment is expensive." McKinsey’s research basically proves that point. Here is how we’re designing our product to tackle the judgment bottleneck head-on.
Start with real context. Kodiact connects internal data (spend, BOMs, contracts, inventory) with external market signals (weather, freight, index prices). The goal isn't to drown you in noise, but to deliver a few razor-sharp alerts linked directly to specific SKUs and plants. A category lead should start their day with three actionable decisions, not a inbox full of 40 articles to read.
Expose the machinery. Every scenario in Kodiact lays out its cards: which tariff, which volume, which price index. We build a clear "what you have to believe" view for every recommendation, so a CPO can instantly challenge the two key drivers instead of auditing the whole spreadsheet from scratch.
Set explicit decision tiers. McKinsey’s decision tiers map directly to procurement operations. Routine reorders belong in Tier 1 (automated within guardrails). Re-hedging after a market jump belongs in Tier 2 (human decision, AI supported). Multi-year supplier commitments sit firmly in Tier 3 (executive sign-off). We let teams configure these triggers upfront so the system knows when to execute and when to ask for permission.
Automate the devil's advocate. We’re building challenger perspectives right into our reasoning layer. Before a recommendation lands in front of a category manager, it gets stress-tested against potential supply risks, vendor pushback, and margin impacts.
Turn the decision ledger into a flight simulator. This is how we solve the junior training crisis. Kodiact's decision ledger logs the original market signal, options considered, owner, reasoning, and final outcome. When a junior manager faces a cocoa price spike, they don't start from zero—they inspect how the team handled the last spike and what happened. Better yet, they can practice making calls on past market scenarios to build real intuition without risking actual dollars.
The ledger also keeps people honest. Every major call has a named owner. And because strategy in Kodiact is a living system, teams review decisions dynamically whenever market triggers fire, rather than once a year during budget season.
The bottom line
McKinsey finishes with a thought that belongs on every CPO's office wall: the ultimate winners won't be the companies that blindly automate the most decisions. They’ll be the ones that leverage AI's speed to empower and protect the human judgment that AI can't replace.
In direct materials, judgment is your most valuable asset. At Kodiact, our mission is to give that judgment clearer context, sharper tools, and a longer memory.
If you're wrestling with this bottleneck in your own procurement org, drop me a line—I'd love to hear how you're tackling it.