What is customer win-back?

What counts as winning a customer back?

Reacquiring a customer who has left, and managing the relationship that follows. Thomas et al. (2004)² describe customer winback as the process of firms “revitalizing relationships with customers who have” defected, and their summary names both halves: restarting those relationships and managing them afterwards. Tokman et al. (2007)³ use the term customer reacquisition and take their definition from Stauss and Friege: “rebuilding the relationship with customers who explicitly” quit it. Kumar et al. (2015)¹ note that “This reacquisition of lost customers has been termed” customer win-back, and they draw the edge of the term on the other side. Customers who call to come back on their own are not part of it: “We do not consider this customer win-back, because” those customers “are initiating their own return”. In the firm they studied, such customers were fewer than 1 per cent of the number to whom it made win-back offers.

Kumar et al.¹ divide a relationship into a first lifetime, a period after defection and a second lifetime: once a customer accepts the offer, everything after reacquisition “comprises their second lifetime with the firm”. What a returning customer is worth is measured on that second lifetime. Thomas et al.², following Stauss and Friege, call it the second lifetime value: “This metric focuses only on the NPV generated after a customer has been reacquired.” Tokman et al.³ quote Stauss and Friege on the same point: for a customer won back, the lifetime value of the terminated relationship “is not appropriate. Only the value of the regained customer” in the future counts. Kumar et al. split that value into “all three components of second-lifetime value”: the likelihood of reacquisition, how long the second lifetime lasts and how profitable it is per month. Tokman et al. add the cost side: “However, customers do not come back for free.”

In this guide, customer win-back means a firm’s deliberate attempt to reacquire customers who have ended their relationship with it, together with the terms on which it serves them once they return. It is judged by second lifetime value: the chance of reacquiring the customer, how long the second relationship lasts and what it earns, net of what the reacquisition cost. Customers who return without an approach from the firm fall outside it. Win-back is distinct from retention, which works on current customers, from acquisition, which works on people with no experience of the firm, and from churn, which counts the loss; the customer churn guide sets out how that loss is counted.

How does a lost customer differ from a prospect or a current customer?

By the record the firm holds about them. Tokman et al. (2007)³ quote Stauss and Friege’s contrast: recruitment is aimed at prospects who “lack experience with the service offered”, and in retention management “current customers are the target group”; Tokman et al. complete it in their own words, “whereas reacquisition targets defected or lost customers”. Kumar et al. (2015)¹ put the practical consequence plainly: rather than treat lost customers as new ones, firms should “leverage data from their first lifetime to reacquire them”.

Their data show that those who came back had a different record. Of the lost customers who received an offer, those who came back had a longer first lifetime on average than the whole group, returners included (1,564 against 1,431 days), and had paid more per month before they left (US$81.70 against US$72.60). In their model, customers who had referred others during the first lifetime were more likely to accept the offer, and customers who had complained were less likely. A service recovery, the firm’s attempt to put a service failure right, was also associated with a higher chance of return, against the authors’ own hypothesis: “Surprisingly, a service recovery in the first lifetime was” positively associated with reacquisition. The reason for defection a customer gave on leaving was related to return too: in the model, those who had left over price were the most likely to come back, those who had left over service next, and those who had left over both were the “least likely to accept the firm’s win-back offer”. These are associations in the firm’s records; only the type of offer was assigned at random.

The relationship dissolution essay follows the path by which a business relationship weakens before the loss is recorded. Win-back starts on the other side of that record, with a customer who has already gone.

What did a newspaper’s lapsed subscribers do at different prices?

Thomas et al. (2004)² analysed a newspaper’s subscription records: 566 lapsed customers whom the firm targeted for reacquisition, of whom 416 came back. The firm made one reacquisition offer per customer during the observation period, and customers who had lapsed because they moved or went on vacation were removed. A customer was more likely to be reacquired if the lapse had been shorter, if the first tenure had been longer and if the reacquisition price was lower. The price relative to what the customer had paid before the lapse “has a much smaller effect” than the price itself, so the authors suggest simply offering “a low price, regardless of the price that the customer” used to pay.

After reacquisition the picture changed. The observed second tenure averaged 177.4 days, and 66.8 per cent of the returning subscribers ended the relationship again within the two-year observation horizon. Higher retention prices, the prices charged after reacquisition, went with longer second tenures, which the authors describe as “contrary to economic theory”. Price cuts relative to the old price went with significantly longer second tenures; price rises relative to it showed no significant effect. Across customers, the inclination to be reacquired and the inclination to stay were negatively correlated (a posterior mean of −0.09): customers who are “easiest to win back” “may not always be the best” customers to keep. The article does not describe the prices as randomly assigned, and the authors list price endogeneity among the problems it would “be useful to address”, so these are associations in one firm’s records.

Which lapsed customers were worth an offer?

In the newspaper data, it depended heavily on the profile. Thomas et al. (2004)² used the fitted model to forecast the expected second lifetime value of lapsed subscribers with five profiles built from what the firm knew before making an offer: the lapse duration, the length of the first tenure and the last price paid (Table 4, p. 41). For the profile the table labels the 10th percentile of these covariates, a lapse of 35 periods, a first tenure of 5 days and a last price of US$2.90, the predicted probability of reacquisition was 0.001 and the expected value US$0.01. For the 90th-percentile profile, a lapse of one period, 365 days and US$1.75, they were 0.968 and US$67.45. The figures rest on costs the authors assumed (US$1.00 per reacquisition and per period). For the average reacquired customer the model gave US$41.07, which the authors read as showing that, on average, the “firm targets attractive prospects among its lapsed customers”. Across the profiles, though, they conclude that “winback should be a selective process and that not all lapsed” customers should be pursued.

The same model gives a pricing rule for the median profile: within the range of prices the firm charged, “a reacquisition price of $1.75 and then raise the retention” price to US$3.00, with a reacquisition probability of about 0.68. It maximizes margins, not the length of the second tenure, and the authors are careful about it: “it is uncertain whether the anticipated customer response” and profit will occur, and “Price changes of this magnitude” “rarely occurred in the data”. The customer lifetime value guide sets out how a lifetime value is forecast and discounted; second lifetime value applies the same idea from the day a customer returns, with the cost of bringing them back on the other side of the account.

Does a discount bring back the customers who left over price?

In stated intentions, more readily than others: in the second of two scenario experiments, a lower price raised the intention to return more strongly when price had been the reason for leaving, while in the first, a reason for switching written into the scenario made no difference to the ratings overall, and the authors report no test of its interaction with price. Tokman et al. (2007)³ ran both studies with adults recruited through students given “extra credit to recruit” them. In the first, respondents imagined leaving a car servicing firm for a rival and rated the old firm’s offer to win them back; a “quasi-experimental design was employed using a scenario”, with 429 usable responses from people randomly assigned to one of eight versions. The offer varied in price (US$5 more or less for an oil change and 10 per cent for other repairs), in service benefits (a 30-day guarantee and renovated premises with more services) and in the personal attention customers remembered, which the authors treat as the old provider’s social capital. The outcome was win-back offer worth, the perceived overall value of the offer to a customer who has left, on a ten-item rating scale. Lower prices and better service benefits each raised it; “no interactions with price were significant”; and when personal attention had been high, customers “put less emphasis on service benefits”. A reason for switching written into the scenario, price or service, made no difference to the ratings.

In the second study, 288 respondents “were first asked to think of a personal experience where” they had switched provider, half of them in mobile phone, internet or long-distance phone services, and then rated their intention to return, their switch-back intention, in reply to a hypothetical offer. A lower price raised the intention, and the effect “was significantly stronger when the original switching reason” was price; better service benefits did not interact with the reason. Satisfaction with the old provider, regret about the switch and delight with the new provider were all significantly associated with the intention, and in the authors’ reading customers delighted with their “new service providers would be less likely to switch-back”. The time since the switch showed no significant association. Of the first study the authors write that “we cannot say with certainty that the same pattern” “of findings would emerge from a field study of subjects” who actually returned; they call the second study’s approach more appropriate and still name a field study of customers who switched back as the complement it lacks.

What did randomly assigned offers do at a telecommunications firm?

The type of offer changed how many lost customers came back, though the authors’ model and the shares the article prints rank the offers differently; what happened after the return differed by offer too, on weaker evidence. Kumar et al. (2015)¹ used records from a United States telecommunications firm that ran “a large-scale randomized field experiment”: 53,729 customers who had cancelled were sent a single offer two to six months after they left, typically valid for 30 days, and 14,384 (27 per cent) returned. The offer was a price discount for six months on the old package, a service upgrade (a better package) for six months at the old price, or both, and its type was assigned at random, “unrelated to the reason they left the firm”. “The focal firm did not vary the depth of” the discount or the upgrade. In the authors’ model, customers offered both were “the most likely to return compared with customers” offered either one, and customers offered only the upgrade were the least likely. The shares in the article’s own descriptive table do not show that order: each offer’s share of the 14,384 who returned and of the 53,729 who received an offer, turned into numbers of customers and set against each other, implies that about 30 per cent of the customers offered the discount came back, about 25 per cent of those offered the upgrade and about 24 per cent of those offered both (Table 3, p. 13 of the Ahead of Print version). With the type of offer assigned at random, those raw shares are themselves a fair comparison, and the article does not say why its model ranks the offers differently, so this guide reports the ranking as the authors’ estimate, not as a settled result. In the model, the gap between a single offer and the combined offer was also smaller where the single offer matched the reason for leaving: the terms for a discount to customers who had left over price and for an upgrade to customers who had left over service were both positive. The reason for leaving was recorded, not assigned, so the authors’ further comparison, that customers who had left over price and were offered a discount were more likely to return than customers who had left over service and were offered an upgrade, sets different customers side by side and is an association.

The comparisons after return are weaker, because each offer brought back different customers; the authors estimate reacquisition, duration and monthly profit jointly to allow for that. In their estimates, the upgrade was more positively associated with the length of the second lifetime than the discount, and the combined offer least of all, with the lowest monthly profit too. Their explanation is a conjecture: “It may be the case that the customers who” accepted so attractive an offer were “actually the most price sensitive”, and “a too-attractive win-back offer may” “have a negative effect relative to other win-back offers by” “encouraging customers to seek better deals”. Customers who had left over price stayed longest in their second lifetime, and those who had left over service earned the firm the most per month. Figure 1 shows the authors’ simulation of both outcomes for each pairing of reason and offer.

The simulation also ranks the offers by expected second-lifetime profitability, which the table defines as the product of reacquisition probability, duration in months and monthly profit. On that ranking, for customers who had left over price, “a price win-back offer yields the” highest value, and for the other two groups the combined offer does. The authors ask readers to be cautious with the simulation, since the firm did not vary how deep the offers were. A further reason: the article does not show the calculation, and this guide could not reproduce the printed amounts from the table’s own means. Multiplying the mean probability, mean duration in months and mean monthly profit for the first cell gives about US$460, against the US$1,117 printed. Even a reacquisition probability of one, at the table’s mean duration and monthly profit, would give only about US$920. The amounts are best read as a ranking inside the simulation, not as dollar forecasts.

Does the time since leaving change the odds?

The three studies give three answers. In the newspaper data, the longer the lapse, the lower the chance of reacquisition, but once a customer was back, the authors found no relationship between “the length of the lapse and the customer’s second tenure”; they add that it “is possible that this null result is due to the relationship” between the lapse and the way their model allows for differences between customers. In Tokman et al.³’s second scenario study, the time since the switch showed no significant association with the intention to return. In the telecommunications data, the relationship with reacquisition had the shape of an inverted U: return was less likely if “the firm contacts the customer too soon or” “too late after defection with a win-back offer”, and the time since defection was “negatively correlated with second-lifetime duration and” monthly profit. That firm made its offers within two to six months of the loss, and the article does not say that the timing was assigned at random, so none of the three results is an experiment on timing.

What should a reacquisition decision record?

Six things, one row per group of lost customers: the reason recorded when they left, their first-term record, the time since the loss, the offer and what it costs, the second term forecast net of that cost, and the customers who would have returned with no offer at all. The first five fields follow the variables the three studies measured; the last is this guide’s, because none of them reports results for a group that received no offer.

An illustration, with no company data: a B2B software firm reviews the annual subscriptions that lapsed at renewal over the past year. For each group the record notes the reason given in the cancellation record (price, missing features, or both), the length of the first term and the last price paid, the months since the lapse, the offer under consideration (a discount on the first year back, or an added module at the old price) and its cost, and a forecast of the second term’s margin net of that cost. It marks which inputs come from the firm’s own records and which are borrowed from studies of newspapers, car servicing and telecommunications. To fill the baseline field, the firm would leave a randomly chosen part of each group without an offer and follow both parts for at least the length of a typical second term, since only that comparison separates what an offer did from the customers who would have returned anyway. These are the author’s examples and have not been tested.

Which questions do these studies leave open?

The first is whether an offer brings back anyone who would not have returned without it. Thomas et al.² (2004) analysed customers who were all targeted, at prices the article does not describe as randomly assigned; Tokman et al. (2007) presented every respondent with an offer; and Kumar et al. (2015)¹ report results only for lost customers who received one, although their firm “randomly selected lost customers to send win-back offers”. The newspaper study assumed standard costs, since “A paucity of cost data prevents us” from doing more, and its firm made a “single reacquisition offer”. Tokman et al.³’s outcomes are ratings and stated intentions in scenarios. Kumar et al. could not see what customers did at competitors, “we observed only a single incidence” of churn per customer, and the held copy is the Ahead of Print version (pp. 1 to 22). Both earlier papers open with odds of selling to active, lost and new customers that Thomas et al. attribute to a practitioner book by Griffin and Lowenstein, and Tokman et al., citing the same book, to a named study (“A study by Marketing Metrics has found”); neither paper measured them, and this guide does not use them.

On this guide’s reading, two points carry over with some confidence, each from the design that shows it. Lost customers differ widely in what they are worth a second time, from almost nothing to about US$67 across the newspaper’s forecast profiles, and their first-term record is associated with whether they return: the newspaper’s records and the telecommunications firm’s records agree on that link, as associations, and in the telecommunications records so is the stated reason for leaving, which the newspaper data did not record. And getting customers back is not the same as keeping them: in the newspaper data, the inclination to be reacquired and the inclination to stay were slightly negatively correlated, and in the telecommunications simulation the combined offer reacquired the most and was forecast to keep them the shortest, by small margins and on a model whose ranking of the offers the article’s own shares do not reproduce. Less certain is timing, where the three studies disagree and none assigned it at random. Whether any of these patterns holds for B2B software contracts, with buying committees and annual terms, the sources do not say.

References

1. Kumar, V., Bhagwat, Y., & Zhang, X. (2015). Regaining “lost” customers: The predictive power of first-lifetime behavior, the reason for defection, and the nature of the win-back offer. Journal of Marketing, 79(4), 34–55. https://doi.org/10.1509/jm.14.0107
2. Thomas, J. S., Blattberg, R. C., & Fox, E. J. (2004). Recapturing lost customers. Journal of Marketing Research, 41(1), 31–45. https://doi.org/10.1509/jmkr.41.1.31.25086
3. Tokman, M., Davis, L. M., & Lemon, K. N. (2007). The WOW factor: Creating value through win-back offers to reacquire lost customers. Journal of Retailing, 83(1), 47–64. https://doi.org/10.1016/j.jretai.2006.10.005