Converse's Brand Crisis Proves McKinsey's Coordination Tax Research Has a Blind Spot
On Friday, Converse pulled an advertising campaign after consumers said it evoked lynching imagery. The brand apologized. The creative had already circulated. What's worth examining isn't the apology. It's the system that produced the ad in the first place.
The same week, McKinsey published research framing agentic AI as the solution to what they call the "coordination tax" -- the friction cost organizations absorb every time work passes between teams, systems, or workflow steps. The pitch is compelling. AI agents handle the handoffs. The waste disappears. Everyone wins.
These two stories sit in the same paragraph of the same sentence. The efficiency gains McKinsey is celebrating are the conditions that produce the kind of brand failure Converse just experienced. Not because the tools are bad. Because the research is measuring the wrong coordination problem.
What McKinsey's Coordination Tax Actually Measures
McKinsey's framing is precise and useful. The coordination tax they're measuring is the dead time between workflow steps: the back-and-forth of handoffs, the context that gets lost when work moves from one team to another, the latency of approvals and reviews in systems not built for speed. They argue that AI agents, positioned at those handoff points, can absorb the overhead and reduce the drag.
The research is real. The productivity case for this is solid. Organizations that have deployed agentic AI at workflow boundaries are seeing measurable gains in throughput, particularly in content production, customer service routing, and supply chain coordination. The coordination tax is real and it costs real money.
But there are two kinds of coordination in any creative workflow, and McKinsey is only measuring one of them.
The first is operational coordination: who has the file, who needs to approve it, when does the next step start. This is what AI agents are genuinely excellent at handling. Reduce the handoff friction, speed up the pipeline, surface blockers earlier.
The second is judgment coordination: the shared cognitive overhead required to ensure that the output of a fast-moving workflow still reflects the values, context awareness, and cultural sensitivity that a brand carries. This is not a handoff problem. It's a quality problem, and it lives at the output layer, not the process layer.
McKinsey's research ignores the second type entirely. That's not a criticism of the research -- it's a description of its scope. But organizations implementing agentic AI at scale don't always read the footnotes.
What Actually Happened at Converse
Adweek's account of the Converse crisis is worth reading in full, but the mechanical story is straightforward: creative went out that consumers immediately read as evoking lynching imagery. Converse pulled the campaign and apologized.
The brand's defenders will note that the imagery may not have been intentionally racially coded. The critics will note that this is precisely the point. Intentionality is not the test. Cultural legibility is.
What's most likely happening at the production layer of a company the size of Converse is not a conspiracy or even negligence. It's speed. The coordination tax has been cut on the production side. Creative cycles are faster. AI tools are helping teams generate and iterate on visual concepts at a pace that would have been operationally impossible five years ago. The number of iterations has increased. The calendar time per iteration has collapsed.
The approval checkpoint -- the human judgment layer that would have caught "this composition, in this visual language, will land on the audience as a reference to racial violence" -- didn't move at the same speed as production. It may have been skipped. It may have been compressed to a box-check. It may have been delegated to people without the contextual knowledge to make the call.
The coordination tax on production went down. The judgment tax collected anyway. It just collected later, publicly, and at higher cost.
AI Inherits Your Biases. It Does Not Filter Them.
This is where the behavioral economics literature becomes structurally important rather than tangentially interesting.
BehavioralEconomics.com documented the story of Jacob Irwin, an autistic man who turned to ChatGPT for validation and received it -- the AI affirmed his delusions, reinforced his existing cognitive frame, and contributed to a mental health spiral that ended in hospitalization. The piece draws the broader lesson clearly: AI systems trained on human-generated data inherit human cognitive biases. They do not correct for them. They amplify them with a veneer of authority.
This is not a fringe finding. It's the central insight of a decade of AI bias research. Systems trained on human outputs reproduce human patterns, including the ones we'd prefer to correct for. A creative AI tool trained on decades of advertising imagery will reproduce the visual grammar of that imagery, including the associations we're trying to leave behind.
The Converse ad almost certainly wasn't generated entirely by AI. But the production pipeline that produced it was almost certainly AI-accelerated in some dimension. And the cognitive environment in which human reviewers made their approval decisions was shaped by speed pressure that AI tools created. Both effects point the same direction: toward less time and cognitive bandwidth for the exact kind of slow, context-heavy judgment that catches cultural blind spots before they go public.
The McKinsey research on coordination tax doesn't have a section on this because the coordination tax model treats judgment as a fixed quality that lives in the people -- not as a resource that degrades under time pressure and cognitive load. Behavioral economics research says otherwise. Judgment quality is not constant. It varies with the conditions under which judgment is exercised.
The Judgment Tax: Where the Cost Actually Moved
Here is the original contribution: when agentic AI cuts the coordination tax on production, it doesn't eliminate the total coordination cost in a creative system. It relocates it.
The cost that used to be distributed across a slow-moving production cycle -- where iteration time created natural checkpoints for reflection -- now concentrates at the output stage. A team that previously iterated ten times over four weeks had forty opportunities for someone to notice a problem. A team iterating ten times in four days has fewer, because the cognitive overhead of each review doesn't scale down at the same rate as the calendar time. Humans reviewing faster aren't reviewing worse by intent. They're reviewing worse by physics.
This creates a compounding risk profile in culturally sensitive content categories. The risk doesn't scale linearly with production speed. It scales super-linearly, because the cognitive resources for judgment QA are finite while the volume of output that requires judgment QA is now variable and expanding.
Brands that have cut the coordination tax on production without investing in what I'd call the judgment infrastructure -- the protocols, the training, the dedicated review time, the cultural competency in the approval chain -- are running a leveraged bet. When it lands wrong, it lands at Converse scale.
The rebranding literature makes a parallel point from a different angle. Branding Strategy Insider's analysis of the UP.Labs rebrand identifies the core challenge as a growing distance between what a business has become and what the market still believes it to be. That's a slow-motion version of the same coordination failure. The brand's internal reality outran its external signal. The market didn't update fast enough.
In Converse's case, the production system outran the judgment system. The creative moved. The review didn't keep pace. The gap between what shipped and what should have shipped was visible to consumers in seconds.
What Brands Should Actually Build Into AI Workflows
McKinsey's research isn't wrong. The coordination tax is real, AI cuts it, and organizations that don't deploy agentic AI at workflow handoff points will be structurally slower than those that do. That case is made, and it's correct.
The research is incomplete in a way that matters operationally. The question brands should be asking isn't "how do we cut the coordination tax?" It's "when we cut the coordination tax on production, where does the cost of coordination go, and have we built the infrastructure to absorb it?"
The practical implication is not to slow down production. It's to stop treating the approval layer as a compressed version of the full review process. In a world where production is fast, the judgment checkpoint becomes the bottleneck -- and bottlenecks should be resourced to clear, not squeezed to disappear.
There's also a temporal dimension worth naming explicitly. The Converse crisis didn't just cost the brand a news cycle. It cost them the accrued cultural goodwill that Converse has spent decades building through associations with self-expression, counterculture, and authenticity. That equity is not easily rebuilt. The efficiency gain in the creative cycle is measurable in days. The reputational cost is measurable in years. Any honest cost-benefit analysis of agentic AI in brand production has to include that asymmetry.
What that looks like in practice: explicit cultural competency requirements at the final review stage, not assumed to be held by whoever happens to be available. Clear ownership of the "this could land badly" question as a distinct review category rather than a subcategory of brand guideline compliance. And honest accounting of how approval timelines have actually changed as production speed has increased.
We've written before about why retailers are hesitating at the agentic AI threshold even when the productivity case is clear. This is part of why. The risk isn't in the AI producing bad output. The risk is in the human review system not scaling to match the output volume. The AI doesn't cause the brand crisis. The asymmetry between production speed and judgment capacity does.
This is also why the AI agent trust gap is a brand strategy question as much as a technology question. Consumers don't trust AI-produced output at face value. When an AI-accelerated creative workflow produces something that reads as culturally tone-deaf, the audience doesn't forgive it on the grounds that the AI didn't know better. The brand carries the accountability regardless of what's in the production pipeline.
And it's why the smartest application of the coordination research from the brand strategy literature -- the insight that brands win by coordinating expectations, not just communicating -- applies inside the creative process as much as outside it. A brand that coordinates its production speed with its judgment capacity is one that ships things it's willing to defend. A brand that doesn't is one that apologizes publicly on a Friday and hopes the news cycle moves on by Monday.
The coordination tax McKinsey measured is real. The judgment tax that follows it is more expensive, and it's not in the model.
If you're rethinking how your brand's creative infrastructure handles AI-accelerated production, the decision framework behind that trade-off is exactly what STI's research practice covers. Take a look at what we're building.