The summer I was seventeen, I mowed football fields and school properties with a small crew across our district. It was not anyone’s idea of a leadership development program. But the person running the grounds crew let me make calls he did not have to let me make: which property to prioritize when weather was moving in, how close to cut around a track without scalping it, when a piece of equipment was worth stopping for and when it was not. He was not training me on purpose. He was busy, and he trusted me with a little more each week, and I learned some of it by getting it wrong in front of him.
I spent two different stretches of high school and college working for horse trainers, and later in college worked my way into two specialty departments at The Andersons, a large department store in Toledo, Ohio, that does not exist anymore. Every one of those jobs handed me small decisions before anyone had officially decided I was ready for them. Someone was usually watching, not formally mentoring, just close enough to let me try, and close enough to catch it if I got it wrong.
None of those jobs would read as leadership experience on a resume. But they were where I actually built it, one modest, forgettable call at a time, under people who seemed to understand, even if they never said so, that they were training me.
I think about those jobs now because that exact mechanism, someone quietly handing a person real decisions before they have technically earned the title, is what a lot of frontline retail work no longer offers by default. This is not really a story about AI removing dull tasks. It is a story about AI removing the same kind of small, low-stakes decisions that built my own judgment, at scale, across an entire industry, mostly without anyone noticing what is disappearing along with them.
Nearly every retail and consumer organization currently rolling out AI-driven store operations describes the change the same way: less admin, more time for people leadership. Fewer hours spent building schedules by hand, reviewing inventory counts, or chasing markdown timing. More hours, in theory, spent coaching a team. It is a reasonable story, but it is also incomplete.
Those tasks were never just admin. For decades, they were how a cashier became a shift lead, a shift lead became a store manager, and a store manager became a district manager. Nobody designed it this way on purpose. It happened because retail and consumer leadership has always been built on repetition: thousands of small, low-stakes calls, made under real conditions, with real feedback. Should we mark this down today or wait for the weekend traffic. Do we pull someone from electronics to cover a register during the lunch rush. Is this a supplier problem or a demand problem. Every one of those decisions was a rep. Every rep built judgment.
AI is now making a growing share of those calls. Nobody meant to stop training managers. The training ground just eroded out from under them, one automated decision at a time.
What is actually disappearing
It is worth being specific about what AI is absorbing in retail operations today, because the language of “automation” tends to flatten very different kinds of work into one category.
Markdown timing, once a judgment call built on floor experience and gut feel for a category, is increasingly a generated recommendation. AI tools now suggest discount percentage and timing to clear slow-moving inventory while protecting margin, a task that used to separate a merchandiser who was still learning from one who had seen enough cycles to trust their instincts. Replenishment and reorder decisions, historically the province of an experienced buyer or store lead, are following the same path, with AI-driven forecasting reducing the guesswork that used to require someone to notice a pattern and act on it.
Scheduling has moved the same direction. Building a shift schedule was never glamorous work, but it forced a new supervisor to hold a dozen variables in their head at once: foot traffic patterns, labor cost targets, individual availability, who was ready for more responsibility and who was not. AI scheduling tools now analyze foot traffic, historical sales by hour and day, and even local events and weather to generate an optimized schedule directly, and the tasks managers used to perform manually, building the schedule, reviewing inventory counts, tracking planogram compliance, are increasingly handled by the system.
The customer-facing side of the floor has followed a similar pattern, though it gets discussed less often in this context. Self-checkout, chat-based support, and automated returns processing have absorbed the routine, low-friction customer interactions that used to make up the bulk of a new associate’s day. What is left disproportionately are the harder moments: the disputed return, the price mismatch, the customer who wants an exception the system will not grant. Those are the interactions with the highest emotional cost and the least room for error, and they are increasingly the primary training ground for anyone entering frontline retail today. The easy reps that used to build confidence before someone had to handle a hard one are largely gone.
None of this is being framed internally as a leadership development issue. It is being framed as an efficiency issue, because it is one. Retailers deploying AI across inventory and workforce operations are reporting meaningful reductions in carrying costs and labor hours. That is a legitimate win. But efficiency and judgment-building used to travel together inside the same task, and separating them was never the plan. It is simply what happens when you automate the decision without noticing it was also the training.
There is also a subtler effect worth naming. A generated recommendation carries an air of neutrality that a manual decision never did. When a person had to set a markdown percentage themselves, they knew they were making a judgment call, and they braced for the possibility of being wrong. When a system generates the same number, it reads as data rather than as a decision, and the habit of questioning it, testing it, or learning from it when it misses tends to fade. The decision has not disappeared, but the feeling that a decision was made has gone.
Why this hits retail and consumer harder than most sectors
The Harvard Business Review argument for redesigning mentoring in the AI era is built around a knowledge-work scenario: an analyst whose repetition of spotting weak arguments in a memo or interpreting ambiguous data used to build professional judgment, and who now has that layer of exposure stripped away by AI doing the first draft. It is a real problem. But it assumes a workforce with a desk, a stable manager relationship, and enough tenure to notice what has changed.
Retail and consumer leadership pipelines do not have any of those advantages, and that is why the erosion compounds faster here than almost anywhere else in the economy.
First, this sector overwhelmingly promotes from within. A district manager was a store manager first. A store manager was, often, a shift lead or an associate before that. There is no equivalent of hiring a leadership pedigree from outside; the judgment has to be built on the floor, because that is where almost every leader in the pipeline came from.
Second, the leadership layer that needs this training is already churning at a rate few other functions would tolerate. Gallup research puts frontline manager turnover at 20 to 30 percent annually, notably higher than the employee base overall in many industries. That means the organization is not developing one generation of frontline leaders and moving on. It is constantly regenerating this layer, year over year, which means any erosion in how judgment gets built does not happen once. It repeats, and compounds, with every new cohort of supervisors who came up with fewer reps than the cohort before them.
Third, the stakes of getting this wrong are already high and well documented. Perceptyx research on 2026 workforce trends puts the cost of poor management at $408 billion annually in turnover for U.S. businesses, with up to $211 billion more in lost productivity, while noting that the largest leadership population, the one responsible for the majority of frontline employees, receives the least development investment of any leadership tier. That imbalance existed before AI touched a single scheduling tool. Quietly removing the informal, on-the-job mechanism that used to partially offset it does not make the gap smaller.
Fourth, promotion timelines in this sector are already compressed relative to almost any other function because turnover forces organizations to fill supervisory roles quickly rather than waiting for a fully seasoned candidate. A hiring manager in a corporate function can usually afford to wait until someone has genuinely earned readiness. A regional retail leader trying to keep forty stores staffed through a turnover rate that outpaces most of the economy often cannot. That combination, less time to accumulate reps before promotion, and fewer reps available to accumulate in the first place, is what makes this sector’s exposure compound rather than simply add up.
The cost of not naming this
If organizations continue to treat AI-driven operations purely as an efficiency story, without acknowledging what else is being displaced, the practical consequence shows up a few years out, when a new wave of supervisors reaches the role having built a fundamentally different skill than the one the job requires.
They will be fluent in evaluating a recommendation. Approve this markdown, override that schedule, flag this reorder as wrong. That is a skill, and it matters. But it is not the same skill as generating a judgment call from ambiguous information under time pressure, which is what the job still demands the first time a system’s recommendation is wrong in a way nobody anticipated, or a customer situation escalates faster than any tool can flag it, or a decision has to be made with incomplete data because the AI has never seen this particular combination of conditions before.
Evaluation and judgment look similar from the outside. They are not the same muscle. One is trained by reviewing outputs. The other is trained by generating them, getting them wrong sometimes, and living with the consequences long enough to calibrate. Retail and consumer organizations that do not distinguish between the two will keep hiring and promoting people who can competently execute a well-functioning system, and will be genuinely surprised when that same population struggles the moment the system is not enough.
The obvious objection, and why it does not hold
The natural pushback to this argument is that it romanticizes inefficiency. Manual markdown decisions were not some hallowed training ritual; they were often guesswork, made under pressure, without good data, and frequently wrong in ways that cost real margin. Nobody should want to preserve bad decision-making just because it happened to double as an informal apprenticeship. That objection is fair, and it deserves a direct answer rather than a dismissal.
The answer is that the choice was never between keeping the old, imperfect manual process and adopting the new, better-informed one. It is between adopting the better process and doing nothing else, or adopting the better process and deliberately rebuilding the developmental function it displaced. Most organizations are currently on the first path, by default rather than by decision. The second path keeps the same efficiency gain. It also restores what the new process quietly took away. There is no version of this argument that asks a company to make worse decisions on purpose so people can learn from them. There is a version that asks a company to notice what learning mechanisms are removed and replace them with something better than accidental exposure.
What deliberate development looks like now
The instinct here is to reach for a training program, and that instinct is not wrong, but it is incomplete on its own. Formal training is not how most frontline leadership judgment got built in the first place. The floor was. The fix has to happen closer to where the judgment used to form, not only in a classroom disconnected from it.
A few directions worth organizations’ attention, not as a finished framework but as a starting point for reengineering what the job used to hand out for free:
- Route edge cases and exceptions deliberately, not habitually. When something falls outside the system’s confidence range, whether a markdown recommendation looks off for a specific store or a schedule conflict needs a human call, the default is usually to hand it to whoever is most tenured, because that is fastest. It is also the option that guarantees the least experienced people never build the muscle. Building a habit of routing a meaningful share of those exceptions to emerging leaders, with real stakes and real oversight, is a direct way to reintroduce the reps the system quietly removed.
- Make the reasoning visible before the action executes. A recommendation that simply gets approved or rejected teaches almost nothing. A structure where an emerging leader has to articulate why a markdown suggestion is right, or why a schedule recommendation does not account for something the system cannot see, forces the same cognitive work the old manual process used to require, even when the system ultimately does the execution.
- Treat override and escalation decisions as a coached, visible skill, not a rare event to be reviewed only when something goes wrong. Most organizations currently look at overrides as exceptions to be minimized. There is a case for looking at them differently: as one of the few remaining moments where judgment is still being actively exercised, and therefore one of the few remaining moments worth deliberately building a coaching conversation around.
- Redefine what “ready for promotion” actually measures. Many frontline promotion criteria still implicitly assume a candidate has accumulated a certain volume of hands-on decisions, because that used to be true by default. If that volume is no longer accumulating on its own, readiness assessments need to test for judgment directly, through scenario-based evaluation or structured shadowing, rather than assuming tenure in the role is still a reliable proxy for it.
None of this reverses the automation, and it should not try to. The tools are staying, and the efficiency gains are real. The task is not to slow the technology down. It is to notice, on purpose, what it used to teach without anyone intending it to, and to rebuild that mechanism somewhere else before the pipeline feels the absence.
Three things to take from this
- AI is not primarily removing frontline administrative work. It is removing the low-stakes decision reps that used to build managerial judgment, and the two have always been bundled inside the same tasks.
- Retail and consumer leadership pipelines are more exposed to this shift than most sectors, because they promote almost entirely from within and already churn through frontline managers faster than nearly any other leadership population.
- The answer is not resisting automation. It is deliberately reengineering the moments where judgment gets built, because the job is no longer generating them as a byproduct of getting done.
The broader implication
The organizations that win the next decade of frontline leadership will not be the ones with the most sophisticated AI tooling in their stores. Most competitors will have access to comparable technology within a few years of each other. The organizations that win will be the ones who noticed, early and specifically, what that tooling was quietly taking away from the people meant to lead through it, and built something deliberate to put back in its place before the gap became visible in a P&L.
Todd, who ran that mowing crew, Lynn and Tim and Joy, the trainers I worked for, and the managers who let me take on more at The Andersons would not have called what they were doing leadership development. They were just people who noticed someone was ready for a little more, and handed it over. That is an easy thing to lose at scale, and a harder thing to rebuild on purpose than it looks. But it is not complicated. It just has to become a decision instead of an accident.
Nobody meant to stop training future managers. The organizations that recover from this will be the ones who decide, on purpose, to do for the next generation what those people did for me without ever calling it a program.
Sources: Gallup, frontline manager turnover data; Perceptyx, “Employee Experience Trends: What the Data Says About 2026”; Harvard Business Review, “Why Mentoring Matters More in the AI Era” (July 2026); industry analysis on AI-driven retail markdown, replenishment, and scheduling automation, 2026.
