Why Your Best Staff Leave in Month Four

Ask an operator about turnover and you get a shrug. "It's the industry."

It is not random, and the shrug is expensive. Losing someone in month four costs more than losing them in week two, because by month four you have paid for the learning and someone else has the relationship.

Turnover clusters. And the cluster has a shape you can see about three weeks before it happens.

Headcount retention lies

The usual number is headcount: how many people are still here at 90 days.

It flatters you. A room with ten casual staff working two shifts a month and three core staff working twenty each has thirteen people. If the three core staff leave, headcount retention says you kept 77%. You lost about 70% of your delivered hours.

Shift-weighted retention = shifts worked this month by people who were here last month
                         ÷ total shifts worked this month

Weight by shifts, not by people. Then a core person leaving shows up as a core person leaving.

This one change usually moves the number by fifteen to thirty points, and the new number is the one that matches how the room actually feels.

The month-four cluster

Turnover in this business tends to bunch at three points, and they have different causes.

When	What it usually means	Worth fighting?
Week 1–2	Bad fit, or the job was not what they expected	No. Fix the hiring conversation
Month 2–4	The interesting one. Competent, integrated, and now comparing	Yes. This is where you lose value
Month 12+	Life change, or genuinely done	Sometimes. Often out of your hands

Month two to four is the expensive band. They are past the learning curve, guests know them, and they now have enough information to compare you to the room down the street. You have paid all the cost and are just starting to collect.

And it is the band you can most affect, because what drives it is usually not money.

What actually drives the month-four exit

In our experience running rooms, the leavers in that band rarely cite pay first. They cite three things:

One — schedule unpredictability. Not the hours. The not knowing. A person who works four nights a week on a published rota is more settled than one who works three on a rota that lands on Wednesday. The cost is on their life outside, and that cost is invisible from inside the venue.

Two — no visible path. Month four is when someone asks themselves what month twelve looks like. If nobody has told them, they will ask a room that has.

Three — being the reliable one. The competent person gets the difficult tables, the short-notice cover, and the closing shifts. Reliability is rewarded with more load. That is fair-feeling to the manager and exhausting to the person.

None of these cost money to fix. All of them cost attention.

The three-week signal

People do not decide to leave on the day they resign. They decide two to four weeks earlier, and the behaviour changes before the conversation.

Signals, in rough order of usefulness
  Availability narrows        … they stop offering the flexible slots
  Shift swaps increase        … outgoing, not incoming
  Late arrivals appear        … in someone who was never late
  Discretionary effort drops  … they do the job, not the extra

The first one is the strongest and the easiest to measure. If your rota system records submitted availability, you already have it.

Availability trend = slots offered this month ÷ slots offered last month

A drop of a third or more, in someone who was consistent, is worth a conversation. Not a confrontation — a conversation. Often it is a life change you can accommodate. Sometimes it is a rota that stopped working. Occasionally you are too late, and even then you learn why.

The conversation nobody has

The exit interview is the wrong time. They have already decided, and they will be polite.

Have it at month three. Before the band, not after. Three questions:

What would make the next three months better than the last three?
Is the schedule working for your life outside?
Where do you want to be in a year?

The third one is the one that matters and the one that gets skipped because it feels awkward. A person who cannot answer it about your room is answering it about someone else's.

Which leavers to fight for

Not all of them. Fighting for everyone burns the attention you need for the ones who matter.

Fight for   … high delivered hours, high guest-return rate, trains others
Let go      … low reliability, or a fit problem that was there from week one
Negotiate   … good but constrained (school, second job, distance)
              — often a schedule problem, not a person problem

The middle group is where rooms lose people they could have kept. A person leaving because Thursdays stopped working is not leaving because of you. They are leaving because nobody asked.

The four numbers

Shift-weighted retention   … monthly. The headline
Retention by tenure band   … week 1-2 / month 2-4 / month 12+. Where you are losing
Availability trend         … per person. The three-week warning
Load concentration         … what share of difficult shifts one person absorbs

The fourth is the one that predicts the second. If one person covers a disproportionate share of the hard shifts, you know who is in the month-four band before they do.

The system we built

tasteck is a booking and analytics system for night venues, built by people who ran them for sixteen years and grew from ¥200 million to ¥1.2 billion a year — six-fold. That growth was staffing as much as sales: you cannot scale a room that loses its core people every four months.

What it does here: shift submission and scheduling in 30-minute blocks, so availability is a record rather than a memory. Shift history per person, so tenure bands and load concentration are computable. Guest records that persist across visits, which is how you see whose guests come back. Settlement with per-person payout, so pay conversations start from the same numbers on both sides.

What it does not do: there is no HR module, no review cycle, no exit-interview workflow. The month-three conversation is yours to have. What the product supplies is the availability and shift history that makes the three-week signal visible at all.

The output no one else produces

tasteck outputs the maximum you can spend on each marketing channel next month, as an amount in your currency.

It connects to this directly. A guest who returns for a specific person is a guest you lose when that person leaves. The lifetime value that justified your acquisition spend was partly held by an employee. Rooms with high turnover systematically overpay for acquisition, because they keep re-buying the same relationships.

Nothing else in the nightlife category produces that figure.

Ask it from ChatGPT

tasteck connects to ChatGPT over MCP: ask your numbers as a question and the answer comes back in the chat — who worked which shifts, whose availability is narrowing, which guests return for whom.

Measured against every vendor listed on Japan's principal nightlife-industry directory, this is the first implementation of it in the category, and the same interface is callable from anywhere rather than being tied to one assistant.

Multi-language is built in, the operating surface itself, with your language set put in place during onboarding.

From $34 a month for up to two venues. Thirty days free on every plan, cancel any time. → Pricing

Start this month

1. Recompute retention weighted by shifts. The number will drop. That is the point.
2. Split it by tenure band. Week 1–2, month 2–4, month 12+. Find your cluster.
3. Pull availability trend per person. Anyone down a third is a conversation this week.
4. Look at load concentration. Who absorbs the hard shifts? That person is in the band.
5. Move the exit interview to month three. Ask where they want to be in a year.

Read next

• Night Venue Scheduling: Why the Roster Never Matches the Room
• Opening a Second Venue: What Breaks First
• The Regular You Lost Without Noticing

---

On benchmarks. No target retention rates, tenure curves, or turnover costs appear in this guide. We do not have a dataset broad enough to publish them, and they vary enormously by format, city, and labour market. Three months of your own shift-weighted numbers, split by tenure band, will show you a cluster that no industry average can.