Contract Renewal Rate and the Link That Has To Exist

Why this matters

A renewal that gets keyed in as a brand new agreement, with nothing tying it back to the one it replaces, is indistinguishable from a customer who walked away. The rate reads low, the shop concludes it has a churn problem, and it goes looking for a cause that is not there. This is not a rare or exotic failure. It is the default outcome of ordinary data entry, because a renewal written up from scratch and a renewal written up as a replacement both produce an agreement the customer is happy with, and only one of them leaves any trace of what it replaced. What follows is one shop's month of chasing a problem they did not have, and the audit that found it.

What the number is made of

Of the agreements whose term ended inside the window, the share that have another agreement pointing back at them as their replacement.

Two things about that composition are worth stating before the story. First, both halves are drawn from the same cohort: the denominator is the agreements that came up for renewal, and the numerator is the subset of those that renewed. That symmetry is what keeps the figure between nought and a hundred and what protects it from the truncation that distorts any rate counting outcomes in the same window as the opportunities that produced them.

Second, the numerator is opt-in. It exists only where somebody recorded a link. That is the mirror of a rate whose denominator is opt-in, and it fails in the opposite direction: an opt-in denominator flatters you the less you record, while an opt-in numerator punishes you the less you record. Sloppy entry makes this number worse than reality, every time, and never better.

The signal

The shop is eleven years old and runs a few hundred maintenance agreements. Reading a twelve-month window that closed 60 days before the day they ran it, 234 agreements had reached the end of their term. Of those, 136 carried a link to a replacement agreement. That is 136 of 234, or 58 percent.

Fifty-eight percent felt wrong to the owner, who could not name fifty customers who had left, let alone ninety-eight. But the number was there in the report and it was computed correctly from what the records said, so they went hunting.

Three explanations killed on evidence

Price. They had raised plan pricing by about 9 percent at the start of the window and the theory was obvious. So split the cohort by whether the agreement took the increase or was grandfathered at the old price. Of the 151 that took the increase, 88 renewed, which is 58 percent of that group. Of the 83 grandfathered, 48 renewed, which is 58 percent of that group. The two groups sum back to the 234-agreement cohort and to the 136 renewals, and they read the same rate. Whatever is depressing this figure is not the price increase.

A technician. The second theory was that one tech was souring accounts. Their busiest technician was the last person on site for 31 percent of the cohort's plan visits and was the last tech on 30 percent of the 98 recorded non-renewals. Those two shares are computed on different bases and they are close enough that the theory has nothing to stand on: if he were driving customers out, his share of non-renewals would sit well above his share of visits.

A cluster around the price letter. They expected the non-renewals to bunch in the two expiry months after the letter went out. The 98 spread across all twelve expiry months, and the heaviest single month held 13 of them against an even spread of about 8 per month. Call the excess 5 agreements at most. Five agreements out of the 234-agreement cohort is about 2 points of rate, which is real and is not a 58.

What the record actually said

Having run out of theories, they pulled the list of 98 and called twenty of them at random. Eleven of the twenty said they were still on a plan and named their most recent visit. Two of the eleven had already had their next plan visit.

That reframed the whole question. It was not a churn investigation any more, it was a records investigation, so they ran a search instead of more calls: for each of the 98, is there a later agreement on the same customer starting within 60 days of the old one ending?

There were 51. That is 51 of 98, or 52 percent of the recorded non-renewals, and it sits right alongside the 11 of 20, or 55 percent, that the phone sample produced. Two independent reads of the same population, one from the records and one from the customers, agreeing inside their sampling noise. That agreement is what let them trust the search rather than calling the remaining seventy-eight.

Then they checked the service address on each of the 51 new agreements against the address on the old one, because a customer can buy a second agreement for a second property without renewing the first. Seven of the 51 were second properties. Those are new business and they do not belong in a renewal numerator.

The corrected rate, and what it is worth

Forty-four of the 98 were renewals keyed as new agreements. The corrected numerator is 136 plus 44, or 180, over the same 234-agreement cohort. That is 77 percent, against a reported 58 percent.

Note what did not change: the denominator. Both figures are computed over the identical cohort, so the 19-point gap is entirely a numerator correction and nothing about the comparison is doing extra work.

Is 77 good? There is no industry renewal rate worth quoting here, and any figure you find published will have been computed on somebody else's cohort rule and somebody else's link discipline, which is exactly the thing that was wrong with this shop's own number an hour ago. The honest benchmark is your own corrected rate over prior windows, and this shop could not have one until they fixed the link, so their first corrected year becomes the baseline.

The gate for calling a move real: restate it as a count before you react to it. Five points on this shop's 234-agreement cohort is about 12 agreements, and twelve customers is a list the owner can read out by name. On a cohort near 100 agreements, which is about the floor at which this rate is worth computing at all, five points is five agreements, which is a bad week rather than a trend. Read the movement in agreements, then convert to points for reporting, never the other way round.

The window boundary moves this number too

An agreement that ended in the last week of the window and renewed three weeks later is a renewal, and it lands outside the numerator if you compute the window the day it closes. So a freshly closed window always reads low, and the same window recomputed a month later reads higher with nothing having happened.

Lag the window. Compute the twelve months ending 60 days ago, which is the same 60 days the audit search used, and for the same reason: 60 days is the period inside which a signature is a renewal rather than a win-back. Tune it to your own notice cadence. A shop whose renewal notices go out 45 days ahead and whose follow-up stops a month past expiry is well served by 60. A shop that chases for a full quarter should use 90 and should say so wherever the number is reported, because a rate computed on a 90-day tail is not comparable with one computed on a 60-day tail.

Fixing entry so the audit does not have to repeat

Two habits, and both are about where the work starts rather than about care.

Start the renewal from the agreement that is expiring, not from the customer. Work that starts at the expiring agreement carries its own reference: whoever writes the new one has the old one in front of them, so what it replaces gets recorded as a by-product of doing the work. Work that starts at a customer name starts from nothing with a term attached, and the link back depends on somebody remembering, which after a year of Fridays somebody will not. What decides it is which list the renewal gets worked from, not how careful anyone is being.

Nothing gets closed as not renewed without a reason and a last-contact date. This costs about fifteen seconds and it converts a residue into a callable list. It also makes the failure visible: a non-renewal list where most rows have no reason recorded is telling you the rows were never worked, which is a different problem from customers leaving and needs a different response.

Auditing your own last twelve months

The cohort, the search and the address check are all above. One step is not, and it is the one that decides whether the result is worth acting on: call ten to twenty of the unlinked agreements at random and ask whether they are still on a plan, then hold that share against the share the record search found. Two reads agreeing inside their sampling noise is what buys you the right to stop calling. Two reads far apart means one of them is counting something other than renewals, and finding out which is cheaper than acting on either.

Then repair the links on the rows the search turned up, so next year's audit measures the year instead of re-measuring this one, and write the corrected figure down with its date, because it is the baseline every future window gets compared against.

References

• See related: Estimate Conversion Rate and the Cohort Problem, for what happens when a rate's outcomes and opportunities share a window
• See related: SLA Compliance Only Counts Jobs That Carry a Deadline, for the mirror case of an opt-in denominator
• See related: Contract Attach Rate Is Measured Against Every Customer You Ever Had
• See related: The Renewal Cadence That Reduces Churn, for the outreach that produces the renewals this card counts