VoxEU Column Frontiers of economic research Industrial organisation

How firms should apologise

Even economies built on market capitalism are built on relationships. And when trust within relationships fray, apologies can help to restore them. This column describes the first large-scale apology experiment done in the field. Using the Uber platform to better understand the costs of apologising, the study asked why and in what cases apologies helped restore relationships. It finds that apologies can indeed work but are sometimes costly.

Why do we apologise?

That’s easy. Society, even one built on market capitalism, is built on relationships. Amartya Sen famously proclaimed in 1977 that the classic homo economicus populating economic models are “rational fools” because they care only about themselves (Sen 1977). But as Kenneth Arrow noted in 1969, “it is useful for individuals to have some trust in each other's word. In the absence of trust it would become very costly to arrange for alternative sanctions and guarantees, and many opportunities for mutually beneficial cooperation would have to be, foregone” (Arrow 1969). Sometimes trust within relationships frays; when that happens, apologies help to restore relationships.

The more difficult question, then, is why do apologies restore relationships? If apologies are just empty words, or cheap talk, why do we let apologies influence our decisions? Moreover, since apologies do seem to reduce anger and lead to forgiveness, and since firms frequently make mistakes that anger their customers, why don’t firms apologise more often?

Our new study suggests that the answer to both questions is because apologies are hard (Halperin et al. 2019).

In previous work (Ho 2012), we showed in lab experiments and in theory that if apologies are costly, then they can act as a signal of trustworthiness and increase the likelihood of future repeat interactions by, say, a customer with a firm. Other lab experiments showed that apologies can sometimes backfire if they are seen as insincere (Fischbacher and Utkial 2013, Gilbert at al. 2017) and that apologies work best as promises (Schniter et al. 2013).

In the first large-scale apology experiment done in the field, we used the Uber platform to run an experiment designed around the theory developed in a previous study (Ho 2012) to better understand the costs of apologizing.

Uber faced a problem. When a consumer requests an Uber ride and is delivered later than promised – because the driver took a wrong turn, for instance – we found that the consumer spent 5-10% less in the future, following a single bad experience, compared to a counterfactually equivalent rider.

We wanted to ask whether an apology make up for that mistake — and if so, how? Theory suggests that costs are a necessary condition for apologies to work but aren’t sufficient. Our goal was to see if apologies work, and if they do, to identify the nature of the costs that allow them to do so.

We tracked 1.5 million Uber customers after they experienced a late ride with Uber, and randomly assigned them to one of eight groups:

  • the control group, which received nothing,
  • a baseline treatment, which received a $5 coupon, and
  • six other treatment groups, which received one of three types of apologies. Half of these treatments got a $5 coupon and the others got only an email.

The types of apology costs that we tested were motivated by theory (Ho 2012):

1) A monetary (promo coupon) cost

This could be a box of chocolates or a litigation settlement. In this case, it is the $5 coupon.

2)  A commitment (promise) cost

We can imagine the consumer offering Uber a menu contract: if you apologise, I will forgive you but hold you to a higher standard; if you don’t apologise, I won’t forgive you, but I will lower my expectations of your performance in the future. In this setting, an incentive-compatible contract would find that an apology followed by good behaviour is ideal, while an apology followed by future bad behaviour is the worst possible outcome for the company.

3) A status (reputation) cost

An apology may restore Uber’s reputation along one dimension but make it worse along another. Incentive compatibility is satisfied because an apology entails loss of status.

The figure below shows the change in spending (net of coupon costs) by customers from each treatment group in the 84 days following the late ride.

Figure 1 Percent change in spending over time

We find that apologies work, but they work because they are costly. A $5 promo coupon was generally effective across treatment conditions and yielded positive returns for Uber compared to the control even net of the coupon cost.

However, we also find that across conditions, apologies can backfire. Consistent with the ‘apology as a promise’ theory, apologies followed by further bad rides had a more negative impact than not apologizing at all. The worst outcomes for Uber occurred in the conditions where we made that promise explicit in the emails sent to customers. While such apologies yielded positive returns in the short run, by the end of our 84-day experiment, customers who received these promises would have spent more had they received no apology at all. This was true even for the customers who received a $5 promo coupon. On the other hand, the emails emphasizing a reputation cost had effects indistinguishable from the baseline treatment.

What does this all mean? It means that apologies work, on average. But apologies work because they are sometimes costly. We identify three costs to the firm for apologizing: 

  1. The cost can be monetary, as in the $5 coupon. 
  2. The cost can come from the implicit promise of being held to a higher standard. Sometimes an apology is worse for the firm than no apology at all. 
  3. The cost can be informational. A good apology requires the empathy and/or expertise to know when an apology will work to restore a relationship and when an apology will only make things worse.

Understanding these costs are important for firms deciding how and when to apologise; for consumers deciding how to interpret an apology; and for policy makers deciding the potential regulations that affect how firms deal with disputes and how the guarantee of consumer rights will affect the functioning of apologies.


Sen, A K (1977), “Rational fools: A critique of the behavioral foundations of economic theory”, Philosophy & Public Affairs: 317-344.

Arrow, K J (1969), “The organization of economic activity: issues pertinent to the choice of market versus nonmarket allocation”, The analysis and evaluation of public expenditure: the PPB system, 1: 59-73.

Fischbacher, U and V Utikal (2013), “On the acceptance of apologies”, Games and Economic Behavior, 82: 592-608.

Gilbert, B, A James and J F Shogren (2018), “Corporate apology for environmental damage”, Journal of Risk and Uncertainty, 56(1): 51-81.

Schniter, E, R M Sheremeta and D Sznycer (2013), “Building and rebuilding trust with promises and apologies”, Journal of Economic Behavior & Organization, 94: 242-256.

Ho, B (2012), “Apologies as signals: with evidence from a trust game”, Management Science, 58(1): 141-158.

Halperin, B, B Ho, J A List and I Muir (2019), “Toward an understanding of the economics of apologies: evidence from a large-scale natural field experiment”, NBER Working Paper 25676.

1,995 Reads