We analysed 28,002 active employees across 143 hospitality operators and 1,479 locations. The most engaged employees leave 69% less often than the least engaged - a gap that survives every robustness check we ran. After normalising for time on platform, the gap is still 33%; after also controlling for location, manager, pay, and tenure, each additional 10pp of daily login rate is associated with 12% lower odds of leaving.
Hospitality has a turnover problem other industries don't. While most US industries see 12-15% annual turnover, hospitality runs at 74-80%. Quick service restaurants regularly exceed 100% - meaning the entire workforce turns over in less than a year, sometimes twice.
The standard playbook is more pay, better benefits, faster onboarding. All useful. None of them are the lever this report is about. We wanted to know if a different lever - the everyday experience of working at a hospitality business - is associated with how long people stay.
The short answer: yes, by a meaningful margin. The longer answer is the rest of this page. If you'd rather skip ahead, the calculator lets you plug in your numbers, and the playbook section tells you what to do.
Replacing a frontline hospitality employee costs $3,000 to $5,000 in recruitment, training, and lost productivity. Plug your numbers in below.
Default reflects the least-engaged third on our normalised (time-adjusted) metric (9.1% per quarter). The most-engaged third sit at 6.2%. The top-engagement scenario reflects what your numbers would look like if every employee engaged like the most-active third in our dataset - in practice no team gets there. Industry-wide annual turnover (US/QSR-weighted) is 70-80%; see FAQ on why the numbers aren't directly comparable.
Replacement cost band: $3,000-$5,000 per frontline hire (Cornell). Calculator uses the midpoint. Numbers are associations from our dataset, not guarantees of intervention effects.
Across 28,002 active employees at 143 customers, the most engaged third (top tercile by total logins) leaves at 3.8% per quarter. The least engaged third leaves at 12.2%. That's a 69% gap.
That's the eye-catching number. Raw tercile gaps have a known limitation: a 90-day stayer simply has more days to log in than a week-3 leaver. The next section walks through normalising for time, then controlling for location and tenure - to show what the gap looks like once mechanical distortion is stripped out.
The 69% gap is real but coarse. We address two real concerns one at a time: time on platform, and confounding from location-level differences.
Even with total logins, stayers have a structural advantage - more days = more chances to open the app. We normalise to a daily login rate: login days divided by days the employee was active in the window. A week-3 leaver who logged in every day now scores 100%, not 21 days.
Re-running the tercile cut on the normalised metric: the most-engaged third still leaves about 33% less often (6.2% vs 9.1% per quarter). Smaller than the raw 69%, but real and structural.
Even after normalising, low-turnover companies might simply have more engaged workers. So we compare employees within the same physical location (same manager, same pay, same local job market) and add tenure as a control to rule out new-hire churn.
Each additional 10 percentage points of daily login rate is associated with 12% lower odds of leaving, p < 0.0001.
All four lines indexed to 100% at 10% daily login rate. Steeper drop = stronger engagement effect. The effect strengthens with stricter controls - the opposite signature of unobserved confounding. n = 28,002 active users. p < 0.0001 across all specifications.
Most spurious correlations weaken under tighter controls. This one strengthens - strong evidence we're measuring something real, not an artefact of company-level or location-level sorting.
Location fixed-effects logistic regression, conditional-logit sample (n = 20,988 / 769 locations with within-org outcome variation; full dataset 28,002 / 1,479). p < 0.0001.
How to read this: each percentage is the predicted probability of leaving that quarter at the median location. At a 10% daily login rate, about 1 in 7 leave; at 75%, about 1 in 14. The model accounts for everything shared at a location (manager, pay, local job market), so the curve isolates engagement itself.
The skeptic's question: is this a fluke at the extremes? No. Each engagement bin shows fewer leavers than the one before, with a small noise blip in the middle that we don't smooth over.
n = 28,002 active users. The 30-50% bin shows a small non-monotonic blip (−15% vs −19% in the 20-30% bin) which reflects noise from small leaver counts; the overall gradient is strongly downward.
This rules out "fluke at the tails." The gradient is clean, strongly downward, and gets steeper as engagement deepens.
Logins can mean anything: checking a shift, clocking in, glancing at the feed. Chat messages mean your team is actually talking to each other. That's the sharpest curve we measured.
n = 28,002 active users. 90-day window. Direct messages and group chat combined. Monotonic across all 7 bins.
Chat volume isn't the goal - connection is. Chat is just the most visible artefact of connection in a platform's data. But the dose-response is steep enough that volume by itself is a fair leading indicator. If you're a multi-site operator and you can see your sites' chat volumes, the low-volume sites are probably your retention risk sites - independent of pay or hours.
So far we measured engagement one way: how often someone logs in. But that conflates two things - consistency (showing up at all) and intensity (using the app deeply on active days). A daily quick-check has high consistency, low intensity. Twelve logins on Monday only has the opposite pattern.
Splitting them in the same model: between two employees at the same site, same tenure, who log in equally often, does deeper use matter?
Both contribute, independently and significantly, after controlling for location and tenure. In plain terms: opening the app counts; what you do once inside counts more. Features that encourage doing something - responding to a pulse, reading a post, sending a message - are where the strongest signal lives.
The data tells us engagement and retention move together. It doesn't tell us why. Here's our best read on the underlying mechanisms.
A hospitality job pays roughly the same wherever you go. What differentiates a place worth staying at is whether you feel known. Our chat dose-response shows employees who exchange messages with their team are progressively less likely to leave. Pay matters; it's a hygiene factor. But pay alone explains roughly nothing in our data once we control for location.
When new starters don't know what's happening, who to ask, or what's expected, they fill the gap with assumptions - most of them negative. A platform that surfaces what's going on (team wins, recognition, new openings) reduces that gap. People who feel informed feel included. People who feel included stay.
Up to 22% of all hospitality turnover happens in the first six weeks. That's the window where engagement either takes hold or doesn't. Structured preboarding is associated with roughly 40% better 90-day retention. See our pieces on preboarding and structured onboarding.
A "great shift today" message in a public team chat costs the manager five seconds. To the person who got it, it's a memory they take home. Multiply across thousands of small interactions and you get the dose-response we measured.
Beautiful platforms with low adoption don't move retention. Average platforms with high adoption do. The product that gets opened four times a day wins the retention race - so the things people open it for (chat, feed, schedule) matter more than the things they don't.
If you read this report and do nothing, the data was for entertainment. Three things we'd recommend, ordered by leverage.
Preboarding is associated with about 40% better 90-day retention. Send the welcome content, role explainer, and meet-the-team feed before they walk in. Day one is too late - by then they've already formed a first impression of how organised you are.
How to do it well: our preboarding guide.
People talk to people they actually work with. Auto-syncing chat channels from your rota does the configuration work for you. Generic 200-person chats go quiet. Site-specific 12-person chats hum.
Posts about a specific site's wins, hires, and celebrations get more engagement than HQ-level updates. Train managers on what good looks like: short posts, photos, names. One realistic target: each site posts something three times a week.
For thirteen more tactics, see retention secrets that top restaurants and bars swear by.
For every 100 active platform users, the gap between light and deep engagement is 3.0 fewer departures per quarter, or about 12 per year. Scaled to representative customer sizes:
| Operator size | Fewer leavers / year | Annual savings range |
|---|---|---|
| Small (50 users) | ~6 | $18,000 - $30,000 |
| Mid-size (150 users) | ~18 | $54,000 - $90,000 |
| Large (500 users) | ~60 | $180,000 - $300,000 |
To put that in concrete terms: a 150-user operator that closes the engagement gap could fund a full additional manager, or a meaningful team-wide pay rise - just from money currently spent replacing staff who didn't have to leave.
Engagement and departures are measured in the same 90-day window. The gradient strengthens with controls and survives every robustness check the data permits. But we cannot yet prove higher engagement causes employees to stay - rather than people who plan to leave disengaging first. A within-person panel study reads out Q3 2026 to settle this. Until then, all findings on this page are associational.
The most-cited figure is 70-80% annual turnover, but it's heavily US-weighted and QSR-weighted (Bureau of Labor Statistics, National Restaurant Association). UK hospitality typically runs at 30-40% annual; Nordic markets lower; premium operators run lower than fast-food chains. So "the industry average" depends a lot on geography and operator type.
Three reasons stacked, and we want to be upfront about all of them.
One: selection. This study only includes employees who logged into the platform at least once in the quarter. We exclude zero-login accounts deliberately - if someone hasn't opened the app in 90 days, they've most likely already left, and including them would inflate leaver counts in a way that doesn't tell us anything about engagement.
Two: who's in our sample. Our 143 customers are predominantly UK and Nordic hospitality operators, skewing toward mid-market and premium brands. UK hospitality runs at roughly 30-40% annual turnover; Nordic markets lower again; premium operators run below QSR averages.
Three: how we count. Industry statistics often use total separations divided by average headcount across the year, which can produce numbers above 100%. We use confirmed system-event leavers divided by activated employees, annualised linearly.
The 69% raw gap, the −33% time-normalised gap, and the −12% per 10pp regression coefficient are relative effects measured within our sample. Same employees, same locations, just different engagement levels. Those internal comparisons are what the methodology is designed to be valid for. The absolute turnover percentages are not meant as a benchmark against US industry totals.
The Cornell Center for Hospitality Research puts the cost at $3,000 to $5,000 per frontline hire when you include recruitment, onboarding, trainer wages, and lost productivity. Manager turnover costs roughly $13,000-$14,000. Most operators underestimate the real cost because they don't account for the productivity dip during the new hire's ramp.
Engagement with a platform is strongly associated with retention. In our study of 28,002 employees across 143 hospitality operators, the most engaged third leave at 3.8% per quarter versus 12.2% for the least engaged - a 69% gap. After normalising for time on platform and controlling for location and tenure, each additional 10 percentage points of daily login rate is associated with 12% lower odds of leaving (p < 0.0001). Whether this is a causal effect of the app or a marker of underlying engagement is something we're testing with longitudinal data through Q3 2026, but the directional evidence is consistent and robust.
Because raw login days mechanically advantage stayers, who simply have more days in the window to accumulate them. A week-3 leaver who logged in every day they were employed only has 21 days of opportunity; a 90-day stayer has 90. Normalising to a rate (login days / days available) removes that advantage. The corrected curve - x-axis labelled in daily login rate from 10% to 100% - is the result.
Logins can mean almost anything: checking a rota, glancing at a notification, clocking in. Chat messages are evidence of a relationship. They mean the employee has someone at work to talk to, asks questions of, gets recognised by, or simply hangs out with digitally. Relationships create reasons to stay that pay alone cannot create. The chat dose-response in our data is the steepest curve we see, dropping from 0% (non-chatters) to −69% (heavy chatters).
The first 90 days are when adoption habits get established. Most operators see meaningful engagement metrics inside 60 days of launch. Retention impact follows on a slight lag because turnover is measured in quarters. A reasonable expectation: visible engagement gains in quarter 1, measurable retention shift in quarters 2-4, full ROI realisation in year one.
Sources. All Gravy Retention Impact Report (v3, April 30, 2026 snapshot. n = 28,002 active employees, 143 paying customers, 1,479 locations, predominantly UK and Nordic hospitality). Cornell Center for Hospitality Research on cost-per-hire. Industry benchmarks from BLS, NRA, DailyPay, and SHRM.
Methodology. Headline figures reflect the raw tercile gap (top vs bottom third by total logins). The robustness regression (location fixed effects + tenure, daily login rate per 10 percentage points) confirms the direction with a more conservative magnitude (12% lower odds per 10pp, p < 0.0001). The calculator uses the normalised tercile reductions to give the most honest counterfactual. All three views are available in the underlying report.
© 2026 All Gravy. UK and Nordic hospitality data; results may not generalise to other industries or regions. Findings are associational; the within-person panel test runs in Q3 2026.
All Gravy Limited (UK)
Waterhouse Square
London EC1N 2SW
Company No.: 14103635
All Gravy ApS (DK)
Struenseegade 15A
2200 Copenhagen
Company No.: DK41335874
All Gravy NO AS (NO)
Hvamstubben 17
2013 Skjetten
Company No.: 931 624 032
All Gravy Sweden AB
Postfack 2381
116 74 Stockholm
Company No.: 559445-3077
%20(1).png)
.png)