ISSUE 005
What People Do After They Buy
The first four entries in this pillar were all versions of the same question: can you find out what someone will do before they do it. The answers were mostly no. This is the other side of the purchase, where the evidence is much better, because what a customer did after they paid you is a fact rather than a prediction. It is also an area where the most quoted numbers turn out to have travelled a long way from what their sources actually say.
An idea worth thinking about
The most-quoted retention statistic says something narrower than you have been told
You have met this one. A five percent increase in customer retention increases profits by 25 to 95 percent. It appears in sales decks, software marketing and roughly every blog post ever written about churn. I went looking for where it comes from.
The closest primary document I could reach is a short Bain brief by Fred Reichheld. Its actual sentence is: in financial services, for example, a five percent increase in customer retention produces more than a 25 percent increase in profit. Two things have gone missing on the way to the slide deck. The industry qualifier — this is a financial services illustration, not a cross-industry result. And the range, because the 95 is not in this document at all. I could not trace the upper bound to a primary source I could open, so I am not going to tell you where it came from.
The disclosure that belongs here: Bain is a consultancy, Reichheld is a Bain fellow and the author of several books about loyalty, and the brief is a marketing document for loyalty consulting. That does not make the claim false, and the mechanism it gives is perfectly sensible — returning customers buy more over time, cost less to serve, refer others, and will often pay a premium rather than switch to someone unfamiliar. The brief is also old: it is undated on its face, references the current downturn, and cites a book published in September 2001, which puts it somewhere around two decades ago.
The best line in it is the one nobody quotes. Reichheld writes that not every customer has the potential to be profitable and long-standing, and that cost-effectiveness means segmenting to find the subset that does. The famous statistic travels everywhere; that sentence travels nowhere, and it is the one that would actually change what a small business does on Monday.
A tool I disappeared into
Customer-base models that run in Excel, published free by the people who wrote them
Bruce Hardie at London Business School keeps a page of technical notes with working spreadsheets attached, free, no account. The relevant ones here are How Not to Project Customer Retention, Fitting the sBG Model to Multi-Cohort Data, and Computing DERL for the sBG Model Using Excel. There are also Excel implementations of the BG/NBD model and a note on building an RFM summary in Excel.
What makes this unusual is that it is the academics who developed these models publishing the implementation rather than a vendor repackaging it. The sBG spreadsheets take a cohort's survival numbers and fit a model that projects the curve forward, which is the thing you cannot do reliably by drawing a trend line — the next slot explains why.
Three limits, and they matter. The notes are written for researchers and graduate students, so they are notation-heavy and assume you are comfortable with the idea of fitting parameters. The contractual models, sBG included, assume you can observe departure, meaning a cancellation or a non-renewal; a heating company never sees the moment a customer decides to call somebody else. For businesses like that the right shelf is the Pareto/NBD and BG/NBD family, also on the page, and it is genuinely harder.
If you open one thing, make it the RFM summary note. Recency, frequency and monetary value computed from a sales list is the most useful customer analysis available to a business with a spreadsheet and no analytics stack, and it needs no model at all.
Learn something
Your retention rate will rise every year whether or not anything improves
This is a peer-reviewed paper, free as a PDF, and it contains the single most useful correction I have read about retention. The authors take survival data for two customer segments of a subscription business, reported in a 2004 book by Berry and Linoff. The Regular segment starts at 100 percent and falls to 63.1 percent after one year, 46.8 after two, 38.2 after three, and keeps going down to 17.3 percent at year twelve. The High End segment holds 86.9 percent after a year and 39.4 percent at year twelve.
Now do the year-on-year arithmetic, which I did and the paper does not print. The Regular segment's retention rate is 63 percent in year one, about 74 in year two, about 82 in year three, and roughly 94 by year twelve. The retention rate climbs steadily for a decade. It looks exactly like a company getting better at keeping customers.
The paper's model assumes the opposite. Each individual customer has a fixed probability of leaving that never changes; what varies is that the probability differs between people. The ones most likely to leave go first, so the survivors are increasingly made up of people who were always going to stay. The observed rate rises because the population sorted itself, not because anybody became more loyal. The authors borrow a phrase for this from demography: the ruse of heterogeneity. They note it is often overlooked by people trying to make sense of aggregate customer patterns, which is a polite way of putting it.
The practical consequence is blunt. If you measure retention across your whole customer base instead of by cohort, the number will drift upward as your business ages, and you will attribute it to something you did. The only measurement that is not contaminated this way is a cohort: a fixed group of people who bought in the same period, followed forward. The paper's appendix walks through implementing the model in a spreadsheet, and the method applies to contractual settings where you can see somebody leave.
What we’re seeing
The top quartile keeps 70 percent of its customers through twelve months. The top quartile.
ChartMogul analysed aggregated data from over 2,100 subscription businesses and published it as a retention report. Its finding on new customers: the top quartile of businesses at any revenue range retain 90 percent of their customers through the first three months and 70 percent through the first twelve. At the larger end, 15 to 30 million in annual recurring revenue, the top quartile manages 93 percent and 77 percent. That is the good end of the distribution losing three customers in ten inside a year.
The disclosure first, because it governs how much weight any of this carries. ChartMogul sells subscription analytics software and the population is its own customers — businesses that were sufficiently interested in retention to buy a tool for measuring it. That is not a random sample of anything. The data covers 2021 and 2022 and the report came out in 2023, so it is a few years old and sits inside a period the report itself describes as unusually hard, with more than half of the businesses recording worse retention in 2022 than 2021. And it is all software subscriptions. An HVAC company is not a software subscription, and none of these levels are a benchmark for one.
What does transfer is the arithmetic, and this is the part worth the time. There are three retention numbers, not one. Customer retention, also called logo retention, is the share of customers you still have. Gross revenue retention is the share of revenue you still have from that same group, excluding anything extra they bought. Net revenue retention counts the extra. The report's own worked example: a hundred customers a year ago, twenty gone, customer retention is 80 percent — but if those twenty were half your revenue, gross revenue retention is 50 percent. Same quarter, thirty points apart.
The same trick runs the other way, which is the version that catches small businesses. The report notes that net retention of 105 percent could be 90 percent gross retention with 15 percent expansion, or 70 percent gross retention with 35 percent expansion. Translated out of software: one good account grows, total revenue holds steady, and you never notice that eleven small customers stopped calling. Revenue is a lagging and flattering indicator of whether people want what you sell. The count of people is the honest one, and almost nobody tracks it.
Which brings this pillar to the thing it has been circling for five entries. Simulated customers could not predict individuals. Stated intent was a weak predictor of behaviour. Survey pricing methods were only as good as their validation against real purchases, and the comparisons between them were written by people selling them. The constructive version of all four is the same measurement: find out what people did after they paid, on a cohort, with a denominator.
Here is the whole method for a business with no analytics anything. Take everyone who bought from you in one month last year — one month, not the whole year. Count how many of them bought again within the following twelve months. Then add up what that same group spent in the second twelve months against what they spent in the first. The first number is your logo retention for that cohort. The second is your revenue retention. Two numbers, a spreadsheet and an afternoon, and unlike every benchmark in this entry they are actually about your business.
One question
Of the customers who bought from you a year ago, how many bought again?
A rough count is fine. So is “I would have to go and look” — that is the most common honest answer and it is the reason this entry exists.
More in this pillar as it's written.
— Shana
Sources link to the original publisher, never to a summary.