How agentic AI is eliminating the two biggest productivity killers in retail media
By Dan Marc, CEO, Footprints AI
The Two Productivity Killers No One Talks About
Jason Fried has a theory, and it has nothing to do with strategy, talent, or funding. It's about the two things that quietly destroy a knowledge worker's day: managers and meetings. M&Ms.
The logic is simple enough to be uncomfortable. Managers exist to check on work; meetings exist to synchronize it. Both do their job by interrupting the work itself.
Paul Graham drew the same tension in his essay on the maker's schedule versus the manager's schedule. Makers need long, unbroken blocks — four hours of deep focus isn't four times an hour, it's worth far more — so a single meeting dropped into the middle of an afternoon doesn't cost thirty minutes, it costs the afternoon. Managers run on a different clock, sliced into hour-long slots because every slot is a different person, a different decision, a different context. Their job is coordination, and the hour is their natural unit.
The trouble is that organizations default to the manager's clock. Everyone's calendar ends up looking like a manager's, including the people who were hired to make things. Your most creative, highest-paid people spend their days in a state a neuroscientist would call chronic cognitive fragmentation.
What Fried didn't account for is a third actor in the system — one that can coordinate without interrupting, synchronize without scheduling, and manage without a manager's calendar. That actor has now arrived.
23 Minutes, 40% Loss, $37 Billion
The numbers behind this are worth sitting with.
It takes the average knowledge worker about 23 minutes to return to deep focus after a single interruption. Not two, not five — twenty-three. Gloria Mark's team at UC Irvine measured it minute by minute, and the recovery time doesn't stay linear; it compounds.
Task-switching costs roughly 40% of productive capacity, according to work quantified by the American Psychological Association. Sophie Leroy named the mechanism “attention residue”: when you leave Task A for Task B, part of your mind stays behind on A. You're physically in the meeting and mentally still on the problem you were pulled out of.
Microsoft's 2025 Work Trend Index put a number on the frequency — about 275 interruptions a day, one every two minutes during core hours, drawn from telemetry across hundreds of millions of Microsoft 365 users. Each ping restarts the 23-minute clock.
And meetings themselves rarely earn their place. In Atlassian's research, only about 11% of meetings are rated highly productive by the people sitting in them. The rest are, for the most part, expensive theater. Atlassian also puts the cost of unnecessary meetings in the US alone at roughly $37 billion a year — a figure that's now on the conservative end, with later estimates running considerably higher.
Stack these on top of each other. A retail media manager sits through six to eight meetings per campaign cycle, and each one fractures the day into fragments that each carry a 40% cognitive tax. The most expensive skill in the building — the ability to think hard about a complex problem — gets shredded by the very coordination layer that's supposed to support it.
This isn't really a culture problem; it's closer to a physics problem, and physics problems need structural fixes, not better meeting etiquette.
M&Ms Were the Best We Had
Most commentary treats managers and meetings as villains. They aren't. They're coping mechanisms, and reasonably good ones for the constraints they were built under.
Think about what a weekly status meeting actually does. It forces people who've drifted apart over five working days back into sync, surfaces problems that would otherwise sit buried in an inbox, and creates a forcing function for decisions no one wants to own alone. A middle manager does something similar: holds context across workstreams, translates strategy into tasks, arbitrates between competing priorities, and keeps a system coherent when it's grown too complex for any single contributor to see whole.
These are essential coordination functions. We didn't invent them out of stupidity — we invented them because they were the best coordination technology available to a species with limited working memory, no shared real-time state, and a painfully slow way of transferring information (we call it talking).
M&Ms worked well when signals changed slowly, when the decision space was small, when humans were the only processing layer, and when a week between insight and action was acceptable. They start to break when signals shift by the hour, when the combinatorial space explodes across every SKU, store, and shopper persona, when processing speed becomes a competitive advantage, and when a week-old insight is already dead.
That second list is a fairly precise description of retail media in 2026.
Retail: The Most M&M-Poisoned Industry on Earth
I've spent more than twenty years inside retail media operations, and the pattern is consistent.
A single campaign for a mid-sized FMCG brand at a single retailer typically runs through:
a briefing meeting
a media plan review
a creative review
a targeting and audience alignment call
a budget approval meeting
a mid-flight optimization review
a post-campaign debrief
and at least one “emergency” call when something underperforms
That's eight meetings, minimum, for one campaign at one retailer. A brand running across five retailers is looking at forty. A quarter with a dozen campaigns gets ugly fast.
The people doing this work — media buyers, trade marketing directors, retail media leads — routinely tell me they spend 60 to 70% of their hours in calls. Not planning media, not analyzing performance, not thinking about how to actually reach shoppers. Coordinating.
Here's what should bother you about that. The richest signals in retail media — real-time purchase data, store-level inventory, shopper journey patterns, promotional elasticity — get pushed through a coordination layer with roughly the bandwidth of a weekly PowerPoint. You have a data environment capable of producing millions of signals an hour, and you process them through a Thursday-at-2pm conference call with a one-week refresh cycle. That's not a 10% efficiency loss; it's a latency gap of several orders of magnitude. It's a Formula 1 engine connected to the wheels by a horse.
The Meeting That Didn't Need to Happen
At the start of 2023, Shopify ran a company-wide “calendar purge.” It cancelled every recurring meeting of three or more people, reinstated meeting-free Wednesdays, and boxed large gatherings into a single Thursday window — clearing roughly 10,000 events off employee calendars. In the first two months, average time in meetings fell about a third year over year, and the company projected its teams would ship around 25% more projects over the year as a result. What it replaced meetings with wasn't more Slack noise but systems that made them unnecessary: async documentation, decision logs, clear ownership, automated status reporting.
The meetings didn't need to happen because the information they carried could move through faster, cheaper, less disruptive channels. Map that principle onto the categories of meetings that eat retail media teams:
Status updates. “Where are we on the Unilever campaign?” That's a database query wearing the costume of a meeting. An agent monitoring performance can surface status continuously and flag only the exceptions that need a human.
Optimization decisions. “Should we move budget from display to sponsored products?” That's a calculation with known inputs — current ROAS, remaining budget, category history, promotional conflicts — that an agent settles in milliseconds rather than in a 45-minute call with six people next Tuesday.
Approval workflows. “Can we approve the pet food creative?” That's a checklist against brand guidelines, retailer specs, and compliance rules — something an agent validates in seconds instead of across three email chains and a calendar hold.
Coordination across teams. “Let's align the trade promo with the retail media flight.” That's a constraint-satisfaction problem across dates, budgets, audiences, inventory, and competitive activity, and an agent can solve it continuously instead of at scheduled intervals.
Each of these meeting types exists because humans needed a way to synchronize. Once the synchronization happens in the system itself, the meeting isn't merely less useful. It's redundant.
The Coordination Layer Gets Automated
Gartner has a prediction that tends to make boardrooms shift in their seats: through 2026, one in five organizations will use AI to flatten their structure and cut more than half of current middle management roles. Not a fringe forecast — Gartner, twenty percent of organizations, half the middle layer.
This isn't fundamentally about layoffs. It's about the coordination function those roles perform being automated. The meeting-scheduling, status-collecting, information-routing work is what gets replaced.
Klarna offered an early and genuinely messy version of this. It froze hiring and let attrition pull headcount down from around 5,000 toward 3,500, handing a growing share of coordination and service work to AI, and reported lower cost per interaction and higher revenue per employee. It's worth being honest about how the story continued: Klarna later admitted it had pushed efficiency too hard at the expense of quality and began rehiring for customer service. The episode cuts both ways — but the directional signal, that a large slice of coordination work is now automatable, held up.
The reason this lands hard in retail media is that the industry is unusually coordination-heavy. The ratio of coordination to creative work is inverted; people hired to think spend their days synchronizing. When that layer automates, three things follow.
The meeting calendar collapses — not by decree but by obsolescence. You don't schedule a weekly review when an agent surfaces anomalies in real time, and you don't hold a planning meeting when the plan adjusts continuously.
Reporting layers thin out. When information flows straight from data to decision to action, the human intermediaries who used to ferry it between levels lose their function — not because they're incompetent, but because the ferrying itself disappears.
And spans of control widen. A director who once oversaw six people, each managing six more, can oversee outcomes across whole portfolios once agents handle the granular coordination that justified the middle layer. That's not science fiction; it's a plausible org chart for 2027.
What If the Meeting Was Never the Point?
It's worth asking whether we've been solving the wrong problem all along. The question was never “how do we make meetings better?” but “what were meetings trying to solve, and can we solve it structurally?”
Meetings were trying to solve coordination across a system too big and too fast for any single person to hold in their head. In retail media, that system looks like this:
50,000 SKUs
500 stores
20 distinct shopper personas
real-time promotional calendars
dynamic inventory levels
competitive activity signals
weather, events, and seasonality
budget constraints at brand, category, and retailer level
The combinatorial space runs to billions of decision points per day. No meeting can coordinate that, no manager can hold it in working memory, and no weekly review can process signals at that speed. This is where agentic AI operating on a brand–shopper–store graph changes the game structurally.
Not “AI-assisted planning,” and not a chatbot that recaps last week. A system that runs continuously and:
Detects shifts in shopper behavior at the store and SKU level before they show up in aggregated reports. A spike in plant-based milk in suburban stores on Tuesday mornings isn't a trend that surfaces in a monthly review; it's an opportunity that expires in hours.
Predicts how those shifts cascade. If Store 247 sees a 30% lift in organic snacks after a local wellness event, which other stores share that demographic profile, and how long does the window stay open?
Prescribes specific action — shift 12% of budget from banner display to sponsored product ads for these 47 SKUs across these 83 stores for the next 72 hours. An executable instruction, not a slide.
Acts on those prescriptions inside the system: budget reallocation, bid adjustments, creative rotation, audience expansion or contraction — all within the guardrails a human strategist has set, but executed at machine speed.
Measures outcomes at the same granularity. Not “the campaign did well,” but Store 247 returned 4.2x ROAS on organic snacks among health-conscious millennials during the 72-hour window while Store 312 returned 1.1x — and here's what differs between them.
Adjusts continuously on the basis of those measurements. The feedback loop isn't weekly; every hour of data sharpens the next hour of decisions.
The underlying point is that alignment doesn't have to be achieved through human coordination at all. It can be an emergent property of a system that shares state and optimizes toward shared objectives. The meeting was never the point — coordination was, and coordination is now a computational problem with a computational solution. Fifty thousand SKUs, five hundred stores, twenty personas, billions of permutations: no deck holds that, no meeting processes it, no manager coordinates it. A graph can.
Makers Finally Get to Make
So what's left for the humans?
For the first time in the history of advertising, the people hired to think creatively about brand–shopper relationships might actually get to do it. Full days of unbroken focus stop being a luxury and become the default, because once the coordination layer is automated there's no manager's schedule left to compete with — the coordination no longer runs at human speed.
What tends to emerge is smaller, more senior teams doing more meaningful work. Brand strategists spend their time understanding shopper psychology and building creative hypotheses about behavior change, rather than sitting in status calls. Data scientists push at the edges of what the signals can predict and find patterns no meeting would ever surface, rather than assembling weekly decks. Creative directors design experiences that connect brands to real human moments, rather than sitting through three rounds of alignment before they're allowed to create.
The teams get smaller because coordination overhead was the reason they were large in the first place. Remove the coordination tax and you need fewer people doing higher-value work.
None of this is a story about eliminating jobs. It's about eliminating the parts of jobs nobody wanted and nobody was especially good at — the parts that burned out your best people and pushed senior talent out of the industry.
The makers finally get to make. The meetings finally die. And M&Ms belong in a packet, not on a calendar.
