Article: Funnel Psychology and Retention
Reviewed: November 2026 · Method version 2 · How these reviews work
Each factual claim in the article was checked against the source it cites, then challenged by reviewers whose job was to find the strongest evidence against it: replication attempts, meta-analyses, later papers that cite the original work, and research from other fields that study the same mechanism. A separate reviewer weighed both sides, and a script re-checked every quoted source. Where a claim was overstated or misattributed, the article was corrected. The research below supports the article as it now reads.
Using a credit card company’s figures, they projected that cutting the defection rate from 20% to 10% would double the average customer relationship from five years to ten and more than double each customer’s value, from $134 to $300.
The credit card figures in ‘Zero Defections’ are a projection from defection rates, not an observed before-and-after result; the $134 and $300 values could not be checked against the paywalled original.
Reducing defections by 5 percentage points, they estimated, would raise profits by 85% in one bank’s branch system, 50% in an insurance brokerage and 30% in an auto-service chain.
Heskett et al. (1994), with Sasser as co-author, describe the 25% to 85% profit figures as estimates.
In the credit card example the further 5% drop runs from 10% to 5% defection (five percentage points, halving defection), and the figures describe what the change would be worth, not what it would cost to achieve.
Byron Sharp of the Ehrenberg-Bass Institute made the same two points: the 5% is percentage points, and the projection assumes the reduction is free.
Across brands, loyalty tends to track the number of buyers, so a brand rarely has unusually high loyalty without also reaching more people.
Double jeopardy is a cross-brand pattern; Wright and Riebe (2010) found brand defection also follows market share, while Fader and Schmittlein (1993) found high-share brands get somewhat more loyalty than the model predicts.
Retention rose with tenure in both customer segments, and the model reproduced the rise without anyone growing more loyal; in their words it is ‘simply a sorting effect in a heterogeneous population’.
Fader and Hardie’s 2018 extension found that individual churn is unlikely to decline even when aggregate retention rises.
Read this way, a rising curve often says more about whom you acquired than about what the product did to them, though promotions and product changes can still move it.
Schweidel, Fader and Bradlow (2008) found promotions improved retention forecasts in all seven telecom services studied, so firm actions also shape the curve.
Automaticity rose along a curve that flattened over time. Among the participants whose curves could be modelled, reaching 95% of their plateau was projected to take anywhere from 18 to 254 days.
Lally’s model fitted 62 of 82 analysed participants and values above 84 days are extrapolations; a 2024 systematic review reports medians of 59 to 66 days and an individual range of 4 to 335 days.
Lally’s participants tied each behaviour to a consistent context from the start, and other studies find that repeating an action after the same cue builds habit strength.
Implementation-intention and cue-repetition research supports anchoring actions to existing routines; Lally (2010) did not compare routine cues with reminder times.
The study found that a single miss barely dented habit formation, so the reset punishes a lapse the habit itself survives. Research on logged streaks finds that showing a broken streak lowers later engagement, so showing “five of the last six days” keeps the record honest without making one bad evening look like failure.
Silverman and Barasch (2022) found across seven studies that highlighting a broken streak reduces subsequent engagement, and the effect weakens when a streak can be repaired.
Moving acquisition budget to CRM on the belief that loyal customers drive sales is a mistake; reach and retention should be planned together.
Wright and Riebe (2010) found brand defection is mostly explained by category and market share, which supports planning reach and retention together.
These edits were made on November 12, 2026 after the review.
Claim reviewed: In Reichheld and Sasser’s 1990 HBR article ‘Zero Defections’, a credit card company that cut its defection rate from 20% to 10% doubled the average customer relationship from five years to ten and more than doubled each customer’s value, from $134 to $300.
Now reads: Using a credit card company’s figures, they projected that cutting the defection rate from 20% to 10% would double the average customer relationship from five years to ten and more than double each customer’s value, from $134 to $300.
Why: The credit card example is real and the arithmetic holds: 20% defection implies a five-year average life and 10% implies ten. But it was a worked projection, not an observed result, and the exact $134 and $300 figures could not be checked because the original is paywalled. A later study of a catalog retailer found long-tenure customers were not reliably more profitable. Now cites Sharp, ‘Loyalty Myths’, Ehrenberg-Bass Institute.
Claim reviewed: Reichheld and Sasser reported that reducing defections by 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.
Now reads: Reducing defections by 5 percentage points, they estimated, would raise profits by 85% in one bank’s branch system, 50% in an insurance brokerage and 30% in an auto-service chain.
Why: A later HBR article co-written by Sasser confirms the 25% to 85% range, but calls it an estimate. The word ‘generated’ makes these sound like observed profit gains, when they were modelled from customer value. The per-industry figures could not be checked against the paywalled original. Now cites Heskett, Jones, Loveman, Sasser and Schlesinger, ‘Putting the Service-Profit Chain to Work’, Harvard Business Review (1994).
Claim reviewed: Loyalty tends to rise and fall with the number of buyers, so a brand rarely gets unusual loyalty without also reaching more people; loyalty is hard to raise on its own.
Now reads: Across brands, loyalty tends to track the number of buyers, so a brand rarely has unusually high loyalty without also reaching more people.
Why: Double jeopardy compares brands at one point in time, so it shows loyalty tracking size across brands, not a brand’s loyalty rising and falling over time. The conclusion that loyalty is hard to raise on its own is the Ehrenberg-Bass reading and is backed by defection research, though large brands do show some loyalty above the benchmark. Now cites Wright and Riebe, ‘Double jeopardy in brand defection’, European Journal of Marketing (2010).
Claim reviewed: Adidas’s finding is ‘the same shape’ as double jeopardy, i.e. company data confirming that loyalty follows reach.
Now reads: Adidas saw a related pattern in its own data: its econometric model showed that 60% of revenue came from first-time buyers, not the loyal customers its CRM spending targeted.
Why: The Marketing Week article still says nothing about double jeopardy, Ehrenberg, penetration or brand size; the 60% figure is about where Adidas’s revenue came from, not about buyers per brand. The research that cites Ehrenberg’s 1990 paper treats double jeopardy (bigger brands have slightly more loyal buyers) and penetration-led growth (a brand grows mainly by adding buyers) as two different patterns, and the Adidas finding resembles the second. No commentary found ties the Adidas figure to double jeopardy, so ‘the same shape’ is the article’s own conflation. Now cites Modeling brand market share change in emerging markets, International Marketing Review (2018): market share change is mostly penetration change, the generalization the Adidas figure actually resembles.
Claim reviewed: A rising retention curve says more about whom you acquired than about what the product did to them; useful comparisons hold tenure constant (cohort vs cohort, channel vs channel).
Now reads: Read this way, a rising curve often says more about whom you acquired than about what the product did to them, though promotions and product changes can still move it.
Why: Fader and Hardie’s later work says sorting matters more than individual change, which backs the first half. But firm actions such as promotions do shift retention, and some cohorts dip before rising, so a rising curve is not only about who was acquired. Comparing cohorts at the same tenure is sound advice the article adds itself. Now cites Fader, Hardie, Liu, Davin and Steenburgh, ‘“How to Project Customer Retention” Revisited’, Journal of Interactive Marketing (2018).
Claim reviewed: Automaticity rose along a curve that flattened over time, and reaching 95% of each person’s plateau took anywhere from 18 to 254 days.
Now reads: Automaticity rose along a curve that flattened over time. Among the participants whose curves could be modelled, reaching 95% of their plateau was projected to take anywhere from 18 to 254 days.
Why: The 18 to 254 day range is in the paper, but the curve fitted for only 62 of 82 participants, and anything past day 84 is a projection beyond the study. ‘Each person’s plateau’ overstates who it covers. A 2024 systematic review found a similarly wide spread with medians around two months. Now cites Singh et al., ‘Time to Form a Habit’, Healthcare (2024).
Claim reviewed: Anchoring the core action to an existing routine the user names (rather than a reminder time) supports habit formation, as Lally’s participants tied each behaviour to a consistent context.
Now reads: Lally’s participants tied each behaviour to a consistent context from the start, and other studies find that repeating an action after the same cue builds habit strength.
Why: Lally’s participants all used a consistent context, so the study supports cue-based repetition but did not compare a routine-based cue with a clock-time reminder. Separate trial evidence shows repeating a behaviour after a cue builds habit strength, so the advice is reasonable but goes beyond the cited study. Now cites HabitWalk micro-randomized trial, Applied Psychology: Health and Well-Being (2024).
Claim reviewed: A streak that resets to zero after one miss punishes a lapse the habit itself survives; showing ‘five of the last six days’ is better.
Now reads: The study found that a single miss barely dented habit formation, so the reset punishes a lapse the habit itself survives. Research on logged streaks finds that showing a broken streak lowers later engagement, so showing “five of the last six days” keeps the record honest without making one bad evening look like failure.
Why: Lally supports the first half: one missed day barely affects habit formation. The display advice is not in Lally, but a seven-study consumer research paper found that showing a broken streak lowers later engagement, and that letting people repair a streak softens the drop. The advice is sound and would be stronger with that citation. Now cites Silverman and Barasch, ‘On or Off Track: How (Broken) Streaks Affect Consumer Decisions’, Journal of Consumer Research (2022).
Claim reviewed: A paywall on day seven keeps mainly the people who were already committed, Fader and Hardie’s sorting effect carried out by design.
Now reads: A week is short against an 18-to-254-day range. A paywall that early is likely to keep mainly the people who were already committed, which would be Fader and Hardie’s sorting effect carried out by design. The evidence on timing is mixed, though: in one large software experiment, a seven-day trial won more subscribers and kept them longer than 14- or 30-day trials. Tie the paid prompt to the routine having held (several sessions in the chosen slot) rather than to the calendar, and test the timing rather than assume it.
Why: Neither cited source tests paywall timing, and the nearest research points both ways. The largest trial-length experiment, at a software firm, found that a seven-day trial produced more subscribers and better retention than 14 or 30 days, which is hard to square with a day-seven paywall mainly filtering out everyone but the already committed. On the other side, a smoking-cessation trial shows that a costly commitment does select for people who then succeed, and the habit research confirms that a week is short. The article presents an open question as settled. Now cites Yoganarasimhan, Barzegary and Pani, Design and Evaluation of Optimal Free Trials, Management Science (2023).
Claim reviewed: Product changes usually show up first in month-one retention.
Now reads: Compare month-one retention across those cohorts. Since most churn happens early, product changes often show up here first, but so do changes in who you acquired, which the next step separates.
Why: Still uncited, and no study tests whether product changes appear first in month-one retention. The deeper search found consistent evidence across fields that attrition is heaviest early (statins, new jobs, eHealth, subscriptions), which makes a hedged version safe, and found the likely uncredited source: product-analytics guidance that the first week is where onboarding shows. It also found two reasons the claim can mislead: some cohorts dip before they rise, and changes in who was acquired show up at month one just as product changes do.