It’s 6:40 on a Monday morning and a store is already two people down. The duty manager works the phone, reshuffles the schedule, pulls someone off their day off, and absorbs the gap the way they always do. Ask that manager later who they would have bet on calling in, and the answer comes fast: the team that’s been run ragged through the peak, the ones covering shifts, picking up overtime, holding the schedule together when someone leaves. Burnout, the thinking goes, eventually cashes itself in as a sick note.
It’s a reasonable story. In the data, it’s mostly wrong.
We analyzed sickness absence across a large European grocery retailer: tens of thousands of workers across stores, distribution centers and hubs, several years of records joined to external signals like national flu surveillance and weather. The aim was to understand what really drives employee absenteeism, and how HR analytics can make absence management more evidence-based.
The patterns contradicted the instincts that most absence policies quietly run on. Employees who flagged their workload as “far too light” were absent more than those who called it “too heavy.” Periods of higher overtime came with lower absence, not higher. And some of the strongest risk signals weren’t in the survey answers at all. They were in the blanks where answers should have been.
None of this means overwork is fine or that wellbeing doesn’t matter. The point is narrower and more useful: the levers leaders reach for first are often not the ones that move the number. In retail, where one uncovered shift is lost sales, forced overtime and a thinner customer experience at once, pulling the wrong lever is expensive.
So what actually causes absenteeism in a large retail workforce, and how do you see it coming? Three places where the gut and the data part ways, and what to do about each.
Disengagement, not exhaustion, is the louder signal
The belief that absence rises with workload is so common it’s almost invisible. It shapes how managers read their teams, how HR designs wellbeing programs, and how leaders read a spike on the dashboard: someone got pushed too hard.
The data said otherwise. When we modeled what predicted a worker’s absence rate, self-reported overload was not the signal. If anything the relationship ran the other way: people who called their workload too light showed higher absence than those who called it too heavy. That points less at exhaustion and more at disengagement. People who aren’t stretched, aren’t absorbed, aren’t needed in the way good work makes you feel needed are the ones drifting off the schedule.
Overtime made the point sharper. Higher overtime tracked with lower absence, and the effect held even within the same person over time. This one needs a caveat, and giving it is part of the value. Some of the link is mechanical: you can’t be on overtime and absent in the same shift, so the two partly substitute for each other. We’re not calling overtime a wellness program. But the direction is clear. There’s no sign that the people leaning in are the ones heading for a wave of sick leave. The idea that the hardest workers will crack isn’t in the numbers.
The consequence is about where the money goes. If you think absence is overload, you fund load reduction, headcount relief and resilience training, and you aim it at your busiest people. The data says redirect that attention toward engagement and role design, and toward a different group: the under-used and the disengaged, who rarely show up on a “who’s stretched” radar.
Why survey non-response is an early warning signal for absenteeism
Every large retailer runs an engagement survey, and every analytics team studies the answers. Almost nobody studies the non-answers.
Here, not responding was one of the strongest flags we found. Workers who left the workload question blank, or chose “prefer not to say,” carried materially higher absence risk. In one business unit a missing workload response went with close to three times the rate of those who answered. The blank wasn’t missing information. It was information.
It makes sense once you see it. Disengagement and absence share a root, and the disengaged are the people who don’t fill in the survey, don’t open the email, don’t show up to the town hall. Not responding is a behavior. In a frontline workforce where survey coverage is always patchy, the pattern of who goes quiet is a watchlist you can build for free.
It also flips a problem most HR teams treat as a nuisance. Low survey response usually gets filed under data quality, a thing to chase up, a footnote. The data says the gaps sit exactly where the risk is highest. Two things follow: your engagement metrics flatter you by leaving out your most at-risk people, and the response rate itself is a leading indicator worth tracking on its own.
Predicting absenteeism with existing HR data you already own
There’s a tempting version of workforce analytics that always needs new data: wearables, sentiment tools, a feed you don’t have yet. It’s tempting because it’s a purchase, and purchases feel like progress.
The strongest predictor of who would be absent next period was none of that. It was the worker’s own recent absence history. It roughly doubled expected absence, outweighed every other factor, and held up under the strictest test we ran: comparing a person against their own baseline over time, so it couldn’t be brushed off as “some people are just sickly.”
That’s almost boring, which is the point. The most valuable signal in absence management is already in the HR records every retailer holds. You don’t need a new platform to start spotting risk weeks before a shortage. You need to use what’s there. Most absence management is reactive: you learn about the gap when the shift opens short and the phone starts ringing. A simple, defensible risk score, built from existing records, turns that into a conversation a manager can have early and supportively, before it becomes a coverage problem.
Why good managers get this wrong
Why is the gut so consistently off here? The reason also points to the fix.
Managers see what’s visible. The overworked are visible: on the floor, on overtime, clearly stretched, and when one of them is finally off sick, the conclusion looks obvious. The disengaged are close to invisible. They don’t put themselves forward, don’t land on the radar, don’t create the vivid moments that intuition is built from. Judgment is strong on the cases it can see and weak on the cases it can’t. Absence risk sits mostly in the cases it can’t.
Data doesn’t have that blind spot. It weighs the quiet leaver the same as the visible striver. That’s not a reason to replace a manager’s judgment. It’s the reason to pair it with something that catches what judgment structurally misses. The combination beats either one alone.
From instinct to evidence. Put the three findings together and a different operating model falls out.
Stop aiming support by visibility. Helping the obviously stretched isn’t wrong, but on its own it misses the group where absence concentrates. Let the data point you before you decide who needs help.
Treat your own data as signal. The survey blanks, the absence history, the response rates: not housekeeping. They are some of the most predictive material you have, and you already collect all of it.
Move from reactive to proactive. The same records that explain why people are absent can drive a forward view that lets you act before the gap opens. For a distributed workforce, the difference between hearing about an absence at 6 a.m. and watching the risk build over weeks is the difference between firefighting and planning.
Quantify the prize. Absence in retail isn’t a soft HR cost. It’s coverage, overtime premiums, lost sales at peak, and the churn of teams that are always short. The arithmetic is worth doing even roughly, though the real figure has to come from your own coverage economics, not a benchmark. As an illustration: take a 50,000-person workforce at a conservative loaded day-rate, running one point above target. That single point is on the order of hundreds of thousands of lost days a year, and the cost to cover them (overtime, agency, lost trade) runs into the tens of millions before anyone has the first hard conversation. Most retailers can’t state their own number, so they can’t size what fixing it is worth, or fund the work to do it. The first deliverable of a serious program isn’t a model. It’s a defensible cost figure that turns absence into a line a CFO will back.
