The Lifecycle Stage Where Most Shops Lose People

Why this matters

Ask an owner where they lose customers and you will hear about the ones who left loudly: the price argument, the callback that went badly, the account that moved to a competitor. Those are memorable and they are a rounding error. The bulk of the loss happens at one transition, silently, to people who were never unhappy, and it does not register because nothing happened - a customer simply did not call, at a moment nobody was watching.

This card is about locating that transition in your shop with a cohort table you can build in an afternoon from completed jobs, and about the reason the biggest hole is not automatically the best place to spend. Where the stages themselves are defined and named is a separate card, listed in References; here the subject is measuring the drop between them.

Every stage has two numbers, and shops only ever look at one

A stage has a survival rate (what share of the people who reached it went on to the next one) and a headcount (how many were standing there). Neither one alone tells you where to spend.

Survival rate alone sends you to whichever transition looks worst in percentage terms, which is almost always the very first one and is partly structural rather than fixable. Headcount alone sends you to your biggest segment, which is usually your established customers, where the marginal save is smallest because they were mostly going to stay.

What you actually want is recoverable loss: headcount at the stage, times the share that leaves, times the share of those leavers who were ever winnable. That third term is the one everybody omits, and it is why the worst-looking stage is frequently not the best-paying one to work on.

Build the cohort table from completed jobs

You need one thing your history already has: for each customer, the dates of their completed paid jobs.

1. Pick a cohort by first job date. Everyone whose first ever completed job with you fell in a given twelve-month stretch. Use a stretch that ended at least two full service intervals ago, and preferably three.
2. Count how many reached each subsequent job. How many of that cohort ever got to a second completed job, a third, a fourth.
3. Compute survival between consecutive stages, always dividing by the population of the stage before, not by the original cohort. Mixing those two bases is the most common arithmetic error in this exercise and it makes every later stage look worse than it is.
4. Add the last row: still active. Of the ones who reached four or more jobs, how many have booked inside their own expected window most recently. That separates "reached the stage" from "is still there."

The cohort has to be old enough. If your typical customer interval is 18 months and you run this on a cohort whose first jobs were 12 months ago, you will report a catastrophic first-to-second collapse that is nothing but customers who are not due yet. This single mistake has sent more shops into a panicked win-back campaign than any real attrition ever has. Cohort age is the load-bearing assumption; state it out loud when you present the table.

Worked example: a 400-customer cohort read three years later

A shop pulls everyone whose first completed job landed in a single twelve-month period. That is 400 customers, and the read happens three years after the end of that period, which is comfortably past two typical intervals for their mix of work.

The four loss counts are 252, 52, 22 and 13, which total 339, and 400 minus 339 leaves the 61 customers still active. The arithmetic closes, which is worth checking every time because a table like this is easy to build with a base slipped by one row.

The obvious read: first-to-second is both the worst rate and the biggest number, so put everything there. That is half right, and acting on it alone leaves the better opportunity on the table.

The correction: work the third term. The shop sorts the 252 first-to-second losses by job type and address status. Ninety of them turn out to be structurally one-time: an address the caller no longer occupies, a turnover clean or repair ordered by a landlord between tenants, an out-of-area emergency call, a job type that genuinely happens once per property in a decade. Those were never a second job. That leaves roughly 162 of the 252 as customers who had an ordinary reason to call again and did not.

Now compare two realistic improvements. Lifting first-to-second survival by 5 points, from 37% to 42%, is 5% of the 400-customer base, or 20 additional customers reaching a second job. Lifting second-to-third by 10 points, from 65% to 75%, is 10% of the 148-customer base, or about 15 additional customers.

Twenty against fifteen, on the raw count. But the 15 saved at the second-to-third transition are customers who have already demonstrated an interval, already have a known property, already accepted your rate twice, and cost far less to reach because you know when they are due. The 20 saved at the first transition are unproven, and the 5-point lift itself is harder to get: it depends on first-visit execution, follow-up sequencing, and whether the job type ever repeats.

So the honest conclusion is not "fix the second-to-third stage instead." It is that the two are close enough in value that the shop should work both, with different tools: process changes at the first visit for the top of the funnel, and a due-window follow-up habit for the customers already inside it. A shop that had only looked at the survival column would have spent everything on the first transition and never noticed that the cheaper win was one row down.

The three stages that compete for the title

First job to second. Almost always the worst rate. Partly structural, because a share of first calls are genuinely one-time events, and partly execution, because the customer has one data point about you and no habit. The tell that yours is execution rather than structure: a low rate that persists even after you strip out the one-time job types.

Second to third. The most under-watched, and often the best return. These customers proved they will call twice, which means the relationship is real, and losing them is nearly always a timing failure rather than a preference. This is the stage where the shop stops paying attention because the customer no longer feels new.

Established to lapsed. The smallest count and the largest individual loss, because each one is a customer with years of history and a predictable interval. Rare enough that each departure is worth investigating individually rather than statistically. If you lose several in a season, look for a common cause: a tech departure, a rate change, a slipped response standard.

Why the loss is invisible exactly where it is biggest

Every one of these transitions fails as an absence, and absences are only visible against an expectation. If you have never written down when each customer is expected to be due, a customer who did not call is indistinguishable from a customer who is not due yet, and you will find out at a random moment, usually when somebody notices the name has not come up in a while.

That is the whole reason a shop can lose 63% of its first-time customers before a second job and describe its retention as good. Nobody experienced the loss. There was no cancellation, no complaint, no last call. There was a phone that did not ring on a Tuesday, and it did not ring on a Tuesday 250 more times over three years, and each of those Tuesdays looked normal.

The structural fix is not a campaign. It is having an expected due window per customer so that a non-call becomes an event with a date on it. Everything else in lifecycle work depends on that one field existing.

What changes the answer

• Trade mix. In work that is genuinely episodic and years apart (a major system replacement, a one-off installation), a low first-to-second survival is normal and the leverage moves to referrals and to the property record rather than to repeat frequency.
• Recurring plan or agreement customers. They have a contractual due date, so the interesting transition moves to renewal, and the cohort table should be built on renewal periods rather than job counts.
• Heavy commercial or property-manager mix. The unit of survival is the account, not the person, and turnover in their staff can look like attrition when the account never left. Track at the account level and note contact changes separately.
• Your cohort is too young. Everything above is void. Re-read the cohort age warning and rebuild it on an older cohort before drawing any conclusion at all.

How to verify you got this right

• Every survival percentage divides by the previous stage's population, and the loss counts sum to the cohort minus the survivors. If they do not, a base has slipped.
• The cohort's first jobs ended at least two typical intervals before the date you ran the report, and that date is written on the table.
• The first-to-second losses have been sorted for structurally one-time work before anyone declares a crisis. A retention problem and a job-mix fact look identical until you do this.
• Any statement about a percentage names its base in the same breath. "37% survival from first to second" is safe; "we lose 37%" is the same number pointed backwards and will be repeated wrongly within a week.
• Someone can point at the field on a customer record that says when they are expected to be due. Without it, the table you just built is a post-mortem and not a warning system.

References

• See related: The Customer Lifecycle Stages a Service Shop Actually Has (the stage definitions this table measures between)
• See related: Why the Second Job Is Harder to Win Than the First
• See related: How to Use Service History to Predict the Next Call
• Trade-standard practice for cohort retention measurement in recurring service businesses