Stage 09 · Compounding: The Loop That Pays for Itself

Ariadne's ninth stage is compounding — the feedback loop that closes the system. This post is the unit economics that determine whether that loop actually pays.

The elasticity result

Gupta, Lehmann and Stuart, in the Journal of Marketing Research, valued customers as a discounted stream of margins Gupta et al. 2004:

with $m_t$ the margin in period $t$, $r$ the retention rate and $i$ the discount rate. Their headline is an elasticity comparison, quoted from the abstract:

retention elasticity is in the range of 3-7 ... improving customer retention by 1% is likely to improve customer and firm value by 3-7%. In comparison, margin elasticity is about 1 and acquisition elasticity is only 0.02-0.3.

Retention is worth an order of magnitude more per point than acquisition. For an owner-led service business — where most of the marketing budget and essentially all of the founder's anxiety goes to acquisition — that ordering is worth sitting with.

The most mis-cited statistic in business

Now the detour, because this is a good illustration of what "measure honestly" costs in practice.

You have seen: increasing customer retention by 5% increases profits by 25% to 95%, attributed to Reichheld and Sasser's 1990 HBR article "Zero Defections."

That article does not contain that range. Its exhibit is titled "Reducing Defections 5% Boosts Profits 25% to 85%," and the text reads Reichheld & Sasser 1990:

Reducing defections by just 5% generated 85% more profits in one bank's branch system, 50% more in an insurance brokerage, and 30% more in an auto-service chain.

The 25% to 95% version traces to a 2014 Harvard Business Review blog post by Amy Gallo, which states it and attributes it generically to "research done by Frederick Reichheld of Bain & Company" — not to a specific study Gallo 2014. Meanwhile Bain's own summary page states a third figure entirely: companies can boost profits "by almost 100%" by retaining just 5% more customers Bain & Company.

The honest version: quote the specific cases — 85% at a bank's branch system, 50% at an insurance brokerage, 30% at an auto-service chain — and note the provenance. Pfeifer and Farris later addressed the underlying question properly in a peer-reviewed paper, showing that how much value a given retention improvement produces depends critically on the starting retention rate Pfeifer & Farris 2004. Which is the answer you actually needed: there is no universal multiplier.

Get the formula right

Fader and Hardie document a specific, common and expensive error: there are three legitimate margin-multiple formulas in circulation, and they are not interchangeable Fader & Hardie 2012.

They differ only in whether the initial payment is included and whether cash is booked at period start or end. Formula (2) excludes the initial payment — it is the residual value of an existing customer. Fader and Hardie document a published case of formula (2) being used to set an upper bound on acquisition spend, where formula (1) was required. Using the residual figure to decide what you can pay to acquire someone systematically underspends.

Modelling customers you cannot observe leaving

Service businesses are mostly non-contractual: customers do not cancel, they simply stop. You never observe churn directly, only absence.

The literature has good tools for this. Schmittlein, Morrison and Colombo introduced the Pareto/NBD model, which combines gamma-mixed Poisson purchasing with gamma-mixed exponential dropout to make "who is still alive?" estimable Schmittlein et al. 1987. Fader, Hardie and Lee's BG/NBD replaces the exponential dropout with a beta-geometric one that can only occur immediately after a transaction — the change that makes the likelihood estimable in a spreadsheet Fader et al. 2005. On the CDNOW dataset (2,357 customers, 39-week calibration and holdout), BG/NBD fit better than Pareto/NBD by log-likelihood (−9582.4 versus −9595.0) and by chi-square goodness of fit.

The claim I will actually make

Compounding is not a growth hack. It is the observation that a system with a closed loop has different economics from one without, and that the highest-elasticity arrow in that loop is the one most owner-led firms measure least.

If retention elasticity is 3–7 and acquisition elasticity is 0.02–0.3, then the marginal hour spent making the ninth stage work is worth an order of magnitude more than the marginal hour spent on the first. That is the argument for building the whole path rather than buying the top of it.

References

1. Gupta, S., Lehmann, D. R., & Stuart, J. A. (2004). Valuing customers. Journal of Marketing Research, 41(1), 7–18. https://doi.org/10.1509/jmkr.41.1.7.25084
2. Reichheld, F. F., & Sasser, W. E., Jr. (1990). Zero defections: Quality comes to services. Harvard Business Review, 68(5), 105–111. https://hbr.org/1990/09/zero-defections-quality-comes-to-servicesPractitioner magazine; underlying data are proprietary Bain client analyses, never published or replicated. The article's range is 25–85%.
3. Gallo, A. (2014). The value of keeping the right customers. Harvard Business Review (online), 29 October 2014. https://hbr.org/2014/10/the-value-of-keeping-the-right-customersOrigin of the widely circulated '25% to 95%' figure.
4. Bain & Company (1990). Zero defections: Quality comes to services (insights summary). Bain & Company. https://www.bain.com/insights/zero-defections-quality-comes-to-services-harvard-business-review-hbr/States 'almost 100%' — a third framing of the same claim by the authors' own firm.
5. Pfeifer, P. E., & Farris, P. W. (2004). The elasticity of customer value to retention: The duration of a customer relationship. Journal of Interactive Marketing, 18(2), 20–31. https://doi.org/10.1002/dir.20006
6. Fader, P. S., & Hardie, B. G. S. (2012). Reconciling and clarifying CLV formulas. Working note, brucehardie.com/notes/024. https://brucehardie.com/notes/024/reconciling_clv_formulas.pdfNot peer-reviewed, but the standard reference on this specific discrepancy.
7. Schmittlein, D. C., Morrison, D. G., & Colombo, R. (1987). Counting your customers: Who are they and what will they do next?. Management Science, 33(1), 1–24. https://doi.org/10.1287/mnsc.33.1.1
8. Fader, P. S., Hardie, B. G. S., & Lee, K. L. (2005). 'Counting your customers' the easy way: An alternative to the Pareto/NBD model. Marketing Science, 24(2), 275–284. https://doi.org/10.1287/mksc.1040.0098
9. Berger, P. D., & Nasr, N. I. (1998). Customer lifetime value: Marketing models and applications. Journal of Interactive Marketing, 12(1), 17–30. https://doi.org/10.1002/(SICI)1520-6653(199824)12:1%3C17::AID-DIR3%3E3.0.CO;2-K