Guides

Where Restaurant Overtime Actually Comes From (and How to See It in Your Data)

A high level of demand-forecast inaccuracy can push required staffing as much as 50% higher, per Cornell hospitality research. Overtime in an independent restaurant, cafe or bar is not random: it clusters into five repeatable causes, none of which show up if you only look at the total hours. Here is how to find which one is costing you money, and why cutting hours blindly can make things worse.

Alex Riesenkampff

August 3, 2026 · 17 min read · Markdown

Restaurant overtime is not random. It clusters into five recurring causes, and the largest of them is a demand forecast that was simply wrong: a high level of forecast inaccuracy can push required staffing as much as 50% higher than a perfectly accurate forecast would need, according to Cornell hospitality scheduling research. Two venues can run identical total overtime hours for completely different reasons. A no-show cascade looks nothing like a chronically thin base schedule on paper: both show up as the same number on your labour-cost report, but they need opposite fixes. Nobody publishes a hospitality-specific number to compare yourself against, so most owners never get past the total. This guide is about getting past it.

Cafes, bars, restaurants and bakeries all run this same clock differently. A cafe covering a barista's late arrival with overtime, a bar training a new bartender through a slow Tuesday that turns busy, and a restaurant walking a stand-in line cook through his first Friday service are all paying for the same failure to plan for something that was, in hindsight, foreseeable. The point of this guide is to help you tell which failure you are actually paying for.

Nobody tracks hospitality overtime, so nobody has fixed it

There is no official benchmark for restaurant overtime, in any country, and that absence is itself worth knowing. The US Bureau of Labor Statistics publishes average overtime hours every month for manufacturing, and nothing at all for leisure and hospitality, according to its own Employment Situation release. Scheduling vendors and consultants publish plenty of causes (poor forecasting, no real-time hours visibility, understaffed peaks), but rarely a cited figure behind any of them, because the data mostly does not exist outside your own point-of-sale and rota system.

Restaurant consultant David Scott Peters draws the distinction that most labour-cost conversations skip past entirely: "Controlled overtime is management. Surprise overtime is chaos, and chaos is expensive." His illustration is worth sitting with: "Restaurant A has a labour cost of 30%, with almost no overtime. Restaurant B also has a labour cost of 30% but is running 50 hours of overtime every week. Those are not the same business." Two venues can post an identical labour-cost percentage and be in entirely different operational health. Overtime already sits on top of a labour cost line that runs a median 36.5% of sales for US full-service restaurants and 31.7% for limited-service, per the National Restaurant Association, so even a small, recurring overtime pattern is not a rounding error against an already-thin margin.

Our restaurant labour cost percentage benchmarks cover what a healthy baseline looks like before overtime; this guide picks up from there, on the part of the number nobody else is measuring.

The five places overtime actually comes from

Treat overtime as a symptom with a small number of likely diagnoses, not a single problem with a single fix. In four weeks of scheduled-versus-actual data from almost any independent venue, the overtime hours will sort into one of these five buckets far more cleanly than intuition suggests.

Before the five, make one cut that saves most of the work. Planned overtime has a single dated cause with a clear start and end — a private booking, a public holiday, a new hire still learning the job, a genuine same-day sick call — and it stops on its own once that event passes. Structural overtime recurs: the same shift, or the same name, running over in three or more of the last four weeks. Log the first and leave it alone, because chasing it to zero usually just means understaffing the event itself. Only the second is worth rebuilding a schedule for, and everything below is about the second kind.

A no-show sets off a coverage chain

One absence rarely costs one shift of overtime; it usually costs whoever picks up the slack, and sometimes the person after them too. A server calls in sick at 4pm for a 5pm shift. The manager pulls the closer from the shift before, who now runs a double. That double runs long because the venue is short a pair of hands during the rush it was not built to cover. None of this shows up as "no-show overtime" on a standard report; it shows up as ordinary-looking overtime on whoever absorbed it. If your overtime clusters around the same few names, cross-reference it against your absence log before assuming it is a scheduling problem at all.

Clopening turns one bad night into a slow, expensive morning

Half of hourly service workers report working a shift with under 11 hours between closing one day and opening the next, according to Harvard's Shift Project survey of roughly 30,000 retail and food-service workers. A closer who leaves at midnight and opens again at 7am is not just tired; they are the person most likely to run their close slowly to avoid a mistake, or run their open slowly for the same reason, and both tend to push into paid overtime at one end or the other. New York City's Fair Workweek law bans the practice outright for covered fast-food chain employers, Starbucks among them, unless the worker opts in writing and is paid a $100 premium; that is one city's rule for one category of employer, not a universal one for hospitality generally, but the $38.9 million settlement New York reached with Starbucks in December 2025, covering over 500,000 violations across more than 300 locations, is a useful reminder of how routinely this pattern happens once nobody is watching for it specifically.

The schedule was built on last week's demand, not this week's

This is the largest of the five: Cornell hospitality scheduling research found that at a high level of forecast inaccuracy, required staffing runs as much as 50% higher, and service-delivery costs as much as 39% higher, than when the forecast is perfectly accurate. A rota drafted from memory or a fixed template misses the exact days that generate overtime: the private event, the local match, the public holiday, the weather swing. "The greater the forecast inaccuracy, the greater the staffing requirements," Cornell hospitality researcher Gary Thompson wrote in the industry scheduling guide that reported the finding. It is foundational work, published in 2004, but the mechanism has not aged: a schedule built on the wrong number is a wrong schedule. These are the shifts where actual hours run furthest past scheduled hours, because nobody staffed for what actually happened. If you can pull a POS report of your busiest unplanned days over the last quarter and lay it against your rota for those same days, the gap between what you built and what you needed usually explains a meaningful share of your overtime on its own.

The base schedule is thinner than real demand, every week

Some overtime is not a process failure at all; it is the schedule quietly admitting that budgeted hours are below what the venue actually needs to run. This is the one cause that a scheduling fix will not solve, because the schedule is working exactly as designed against a headcount that is too low. The tell is that overtime shows up on ordinary weeks, not just unusual ones, and it shows up regardless of who is working. If that is your pattern, the honest fix is hiring or accepting the cost, not a tighter rota.

A manager's habit, or a punch-clock error, quietly inflates the number

Not every overtime hour on the report was actually worked; some of it is a timekeeping artefact, and it is worth ruling that out before you assume every hour is a scheduling failure. One documented case, a multi-location operator working with Restaurant365 and Toast, traced what looked like a persistent overtime problem to staff forgetting to clock out combined with the POS automatically closing the day, which pushed unworked minutes into paid hours every single shift. Separately, a habit as simple as a manager defaulting to "whoever's already here" for a last-minute gap, rather than checking who is close to their overtime threshold, adds up over a month in a way no single decision looks responsible for. The same is true of closing tasks that run fifteen or twenty minutes past the scheduled end, night after night: that pattern adds up to a part-time salary over a year without ever looking dramatic on any single day.

Run a four-week audit before you touch the schedule

A single week of overtime data is mostly noise; four weeks is usually enough to see which of the five causes is doing the most damage. This does not require new software, just discipline about pulling the same two numbers, scheduled hours and actual clocked hours, and looking at where they diverge shift by shift rather than only at the weekly total.

Some scheduling platforms already generate a scheduled-versus-actual variance report, and it is a genuinely useful number. What it will not do on its own is tell you why the variance happened; that step still needs a human tagging each instance against the five causes above, at least until you have done it enough times to recognise the pattern on sight.

Once the audit has told you how many of those hours were genuinely avoidable, the cost is arithmetic. Set the premium below to what your own market and contracts actually require: in several countries that figure is zero, and the hours still cost you their base wage.

The 1.5x overtime premium is a US federal rule, not a universal one

Under the Fair Labor Standards Act, most non-exempt US restaurant staff are legally owed 1.5 times their regular rate for every hour worked past 40 in a week, and that premium is exactly what makes the wage-theft cases above prosecutable rather than merely unfair. Some US states layer their own daily-overtime rules on top of the federal floor, so a multi-state operator cannot assume the federal rule alone covers every location. If you operate outside the US, do not import that assumption: overtime pay elsewhere is set by your own country's law and, in many places, by your employment contract rather than a universal rate, so confirm your own jurisdiction's rule directly rather than budgeting around a US figure.

Whichever market you operate in, that legal difference does not change the underlying diagnosis in this guide. Overtime still shows up as extra wage cost and extra fatigue regardless of whether a statute requires a premium on top, and the five causes above are what decide whether that cost is fixable or a genuine staffing signal.

Do not fix a labour-cost number by breaking something else

Squeezing an overtime number without checking its cause tends to relocate the cost, not remove it: across 1.44 million transactions in 25 stores of a US casual-dining chain, moving to fully last-minute, cost-minimising scheduling cut servers' sales productivity by 4.4% against schedules set with advance notice. That finding, from Kamalahmadi, Yu and Zhou's 2021 paper in Management Science, cuts against the instinct to tighten the rota the moment a labour-cost report looks high: the leaner-looking schedule cost the business revenue. Three further failure modes are worth guarding against deliberately, because they are common enough to be documented, not hypothetical.

The clearest is off-the-clock work. When a manager is handed a labour-cost or overtime target with no room to miss it, the fastest way to hit the number on paper is to have staff clock out and keep working, or to falsify the records outright. This is not a rare edge case: the US Department of Labor recovered $1.45 million from a Japanese steakhouse chain in 2022, a repeat violator dating back to 2010, for paying staff off the books to conceal unpaid overtime, and $1.65 million from a Los Angeles restaurant group in 2023 for creating fraudulent records that showed no overtime hours at all. Both were caught. Most operators do not want to find out the hard way what an investigation into their own timesheets would surface.

The second failure mode is quieter: cutting hours to chase a lower number changes take-home pay for hourly staff who were relying on those hours, and that has a real, measured cost of its own. Hourly workers' month-to-month earnings already swing by a typical 9%, with one month in four moving more than 21%, and employer-driven changes to scheduled hours sit behind roughly half of that instability, according to JPMorganChase Institute research, which also finds that workers are more likely to leave employers whose hours are less predictable. Suddenly zeroing out someone's usual overtime is not a neutral cost-saving move from where they are standing; it is the same volatility the research flags as a driver of turnover, just moving in the other direction. The Shift Project puts a figure on where that ends: six-month turnover among hourly service workers ran at 24% for those given two or more weeks' schedule notice, against 39% for those given under 72 hours.

The third is the honest counterexample: sometimes overtime is not a process failure at all, it is evidence that your budgeted headcount is genuinely short of what the venue needs. Cutting hours to hit a lower overtime number in that situation does not remove the underlying demand; it just means fewer people covering it, with the service quality, mistake rate, or sales you can actually capture during a rush absorbing the difference instead. That is the structurally-thin-schedule pattern from the diagnosis above, and it is the one case where the honest fix is hiring, or accepting the cost, rather than tightening the rota further.

None of this argues for ignoring overtime. It argues for diagnosing it before you act on it, the same way you would not treat a symptom without knowing what caused it.

Where Super44 fits, without pretending software replaces judgement

Software can make the four-week audit far faster to run, but it still cannot decide for you whether a pattern is a coverage-chain problem, a clopening habit, or an honest staffing gap. Super44's native staff scheduling drafts a rota from roles, availability and POS-informed demand, and its time tracking records actual clocked hours against what was scheduled, which is the exact scheduled-versus-actual comparison this guide asks you to run by hand. Cross-checking that against your own POS demand data on the specific days overtime spiked turns a vague sense that "overtime is high again" into a shift-by-shift answer for why.

The habit matters more than the tooling. One large riverside restaurant we work with now runs on a weekly Monday-morning business report covering revenue, top and flop products, table rotation and fresh reviews, arriving automatically rather than being assembled by hand. The shape of the fix for overtime is the same: not a blanket cap on hours, but a standing weekly look at the schedule against the data, so a pattern gets caught in week two rather than in a quarterly payroll shock.

The owner still makes the call on what to do about it: hire, adjust the base schedule, fix a coverage-request habit, or simply accept that a handful of unpredictable days will always cost more to staff. Our fair rota guide for small venues covers building the base schedule this all sits on top of, and our staff turnover cost guide uses the same instinct, cost your own real cases instead of borrowing a headline number, applied to a different line on the P&L.