The Cost Nobody Line-Items
Deloitte's workforce analytics research on limited-service restaurants puts hard numbers on a problem most operators carry without measuring. Hourly frontline turnover was 135% in the third quarter of 2024. Management turnover reached 55%, up from 45% in 2019.
Using a Cornell School of Hospitality Management estimate of $5,864 as the average total cost of a frontline departure, a 50-employee restaurant running at 135% turnover experiences about 67 departures a year—roughly $393,000 in annual turnover cost.
In a sector defined by slim margins, that is not an HR expense. That is the margin.
$393,000 a year, at one location, for a cost most P&Ls never name.
The Data Is Already in the Building
The research identifies the data sources that predict attrition, and every one of them already exists inside a typical operator's systems: compensation and benefits detail, scheduling records including hours worked and schedule consistency, manager tenure and performance, employee satisfaction surveys, performance ratings and disciplinary actions, training completion logs, and operational context like business volume, seasonality and local labor market conditions.
The gap is not collection. It is that nobody has connected the tables and asked the question.
The Catalyst Read
In our experience the strongest early signal is rarely wage. It is schedule volatility and manager tenure. When an associate's hours swing week to week, or when a unit changes managers twice in a year, departures follow on a predictable lag.
That is good news, because both are operating decisions rather than market conditions. You can stabilize a schedule this month. You cannot change the local labor market this month.
The caution: an attrition model that produces a monthly report and no intervention is theater. The value is created when a flagged associate gets a conversation, a schedule adjustment or a development step within days—and when the intervention itself is measured.
We reduced associate turnover by 30% in one organization while improving engagement 15%. It was not a new compensation plan. It was scheduling discipline, manager stability and a defined path forward.
How to Start Without a Data Science Team
- Baseline the cost
Departures times a defensible per-departure cost. Put the number on the P&L discussion.
- Start with two variables
Schedule consistency and manager tenure by unit will explain more variance than most full models.
- Segment by tenure band
First-90-day attrition and 12-month attrition have different causes and different fixes.
- Define the intervention before the model
Decide what happens when someone is flagged, then build the flag.
- Measure the intervention
Track retention lift on flagged associates against a control. That is the ROI.
