Should companies say “sorry”? Just keep getting better & don’t bother
Just keep getting better & don’t bother
When we accidentally harm another person, it’s hard to let it pass without at least begging a pardon. A legitimate apology owns the responsibility, by seeing what I did to harm someone. This explains the inadequacy of a non-apology apology: “I’m sorry you made X worse than I intended.” An apology should also recognize the impact upon the other person. Ideally, an apology includes some kind of repair, either immediately or for a future recurrences.
The impulse may start selfishly motivated by a healthy self-regard to take steps to view one’s self-image more favorably. This helps spur reflection and public admission of responsibility. In other situations, an apology may make things worse. If you end up calling more attention to the problem, the time consumed risks further awkwardness. On the receiving end, of course the person to whom an apology is owed isn’t obligated to let you communicate it to them. Serial apologizers, such as the pariah Harvey Weinstein, make a habit of miserable behavior that is promptly followed by sending a gift or flowers to paper over the damage.
Should companies behave like people?
What should a company do when their brand promise breaks down? The analogy with interpersonal dynamics intuitively suggests that the firm should issue an apology. Every firm’s guaranteed to let down some customers.
Let’s get concrete. Companies often market their service by promising a predicted level of quality. Delivery and transportation apps surface when the car or meal is expected to arrive. Lyft & Uber both show the time when you will arrive at your destination (often with a point estimate like “6:32 pm”). To capture greater variability, they may provide a time range (6:32 to 6:48pm). The two times sample a probabilistic prediction. The lower estimate might cover 50% of the cases (let’s call it the “p50”). The choice for the upper range balances quantitative precision against the negative impact of showing a worst-case than most will likely encounter. It looks like they compromise by showing the worst ride for 95% of all cases (“p95”). As a homework problem, what percent of the time do you feel the arrival exceeds your quoted time? Even when the distance between the best and the worst may be quite small, five percent of customers will by necessity experience service worse than the best 95%.
Does an acknowledgment help repair the harm done?
How might a company handle these letdowns? What if a firm explicitly promises to do better? Should the firm spiff the customer something valuable to show their seriousness? Perhaps they specifically promise to “learn from” the mistake?
As a matter of empirical record, data to date suggests a very specific approach to apologizing. In a nutshell, the strategy validates by multiple experiments is: Don’t bother.
Let’s unpack this, since it runs counter to our intutive approach to maintaining good interpersonal relationships. Interactions with a company significantly differ from person-to-person contact.
The invisible hand that brought the customer to the company hinges on mutual self-interest.
The motivation of potential profit drives a company to deliver value, yet it’s rare that such a connection engenders a personal relationship. The transactional nature of business is not nakedly broadcast, yet there’s usually points that reveal how the business exchange involves efficiency rather than intimacy. For explorations of this little recognized fact, consult one economist’s love letter to big business.
Economists view apologies as ‘cheap talk,” mere words without any further backing. Apologies that are costless lack credibility. Facebook has historically leaned into apologies like a drunkard who refuses to drive a car unless it’s equipped with airbags. Serial apologizer Mark Zuckerberg, because of his repetitive contrition, is really a serial privacy violator.
Behavioral Data Argues Against Issuing Apologies
One ride-share company ran a large scale field experiment to assess the impact of giving an apology after a late ride, ie, later than 95% of the others. One researcher had previously shown that, to be credible, apologies require more than words. To back up what might just be “cheap talk,” the rideshare company gave $5 refunds for late rides (> p95).
The experiment found that the net return for issuing an apology was negligible. In cases where they apologized more than once, the results were worse than never having said sorry. By contrast, sending $5 without mentioning the bad ride was just as effective over the long term as issuing an apology along with $5. An apology tries to establish that the bad experience is truly exceptional. In a sense, it’s promising that Baby, from now on, I’m going to be good. If the quality doesn’t significantly improve after the apology, it appears to be worse than never having tried to make amends.
When I’ve told friends about this counter-intuitive result, many assume that the right wording must’ve not been used in the study. Surely, might it just be that the tested apologies were not correctly phrased/delivered/explained? Take a look at the distinct types of apologies (with and without $5) used in the field experiment:
- “Oh no! Your trip took longer than we estimated.” (Basic apology)
- “We know our estimate was off.” (Status apology)
- “We’re working hard to give you arrival times that you can count on.” (Commitment apology)
The study found ‘no statistically significant differences between different message types.’ If wording really does matter, current research hasn’t found the magical formulation. The fact that 3 rationales, superficially quite different, don’t differ implies that wording doesn’t shift the balance here.
Even a methodologically sound, well-powered experiment can never compel a person to abandon their intuitions. The capacity for using quantitative data to support one’s initial assumptions bedevils any study that doesn’t pre-commit to how the outcomes will be handled. This lesson goes back to the very origins of collecting and comparing groups. Historians often forget the paradoxical interpretation drawn from one of the earliest data analyses based on clinical evidence. A French doctor in the 1800s wanted to study how much bloodletting improved health. When the data was tabulated, the group of patients who had been bled had significantly worse outcomes for those who had not been bled. This result ended up being interpreted as implying that the doctor had not done enough bleeding.
Uber used to have an entire team of trained PhDs publicized on their website as the Behavioral Lab. The team’s publications leveraged academic findings (with decks that even had footnotes and bibliographies). What made their approach cutting edge was how they took their training and proposed concrete research applications. The team of academics-turned-behavior-designers generated product-focused recommendations based on their reading of published studies.
I once brought up the apology study with someone consulting at Uber. Apparently, in the two years since being published, institutional awareness of the finding had largely dissipated. I was curious whether the Behavior Labs team thought this research was practicable, as I myself had been leveraging it in experimental scripts for conversational agents. This UX consultant reported that one of the UBLabs’ social scientist’s had been told about the study, but still recommended using an apology since it “felt like the right thing to do.” To my ears, this advice sounded like they had taken the results as suggesting they double-down on the blood-letting.
Other companies have tried similar tactics, with similarly dismal outcomes. Results from an (anonymized) meal delivery experiment were shared at a 2019 conference on behavioral economics. The Grubhub-style company had tried making amends by sending a $5 coupon to those customers who’d received particularly bad deliveries. Apologizing apparently made everything worse: Satisfaction, trust, and the likelihood of using the service all significantly decreased after getting an apology. Another case, reported anecdotally by a well-known behavioral engineer, described an investment made by one of huge health care providers. They tried hand-delivered flowers, with a note to recognize the service was substandard. Apparently, the outcome was worse than saying nothing at all. This behavior design expert told me that this phenomenon is called “Breaking into Jail.” A firm breaks into jail when they manage to direct attention to a mistake they’ve made, even before someone else points out the problem.
One final source of evidence can be drawn from studies of medical malpractice. Early anecdotal data suggested that apologies might be quite effective at reducing malpractice. Based on this widely shared claim, the insurance industry successfully lobbied several states to make laws more conducive to saying sorry. By legislating that a doctor’s apology could not be used as evidence in court, there was a hope that more open admissions of human error would reduce costly litigation. Empirical data showed the reverse: “An apology may alert the patient to malpractice she would not otherwise have discovered.” (“Sorry” Is Never Enough, 2019, Stanford Law Review). We might even call this ‘breaking into court.’
What do I recommend to my clients when they see their own firms’ failing? I first show them how little evidence supports the hope that apologies result in a repaired relationship. Rather than invest resources in calling out past mistakes, it’s far more effective to invest those resources in improving service going forward. An explicit acknowledgment may make the employees at the company feel better. It’s hard to resist this when it “feels like the right thing to do.” But an apology that’s motivated by a desire to repair one’s own self-conception is really no apology at all.