Service apology statistics and the difference between withdrawal and resolution

Updated 11 source citationsSources checked 13 November 2026

On this page
  1. Quick answer
  2. Key Takeaways
  3. What exactly did the experiment measure?
  4. How different were the message outcomes?
  5. Does this show that apologies outperform remedies?
  6. Which support measures should stay separate?
  7. Which calculations can a reader reproduce?
  8. Which event should determine whether a remedy is complete?
  9. Methodology and Sources
Download study text

Quick answer

Service apology statistics can describe a changed customer record while leaving the actual support outcome unanswered. In a field experiment with 632 customers of one German eBay seller, 44.8% withdrew neutral or negative feedback after an apology message. Withdrawal was 21.1% across two monetary offer groups. The purchase period was November 2007 through April 2008. (Abeler, Calaki, Andree and Basek, Jun 2009)

The finding is useful for deciding what a support dashboard should count. A complaint can disappear because the customer received an explanation, accepted a request or changed a public response. That event is distinct from a delivered replacement, a completed refund or a retained customer. The study is historical evidence about response measurement, not a recommendation to pay shoppers to alter reviews.

Key Takeaways

What exactly did the experiment measure?

The seller asked customers to withdraw an existing evaluation through a historical eBay feature called mutual feedback withdrawal. Assignment to the messages was randomized. The apology message explained a delivery problem and asked for withdrawal; the other messages offered EUR 2.50 or EUR 5 in connection with that same requested action. The source states that the feature was no longer available by the paper's discussion. (Abeler, Calaki, Andree and Basek, Jun 2009)

Service apology statistics interpretation matrix: distinguish pending versus settled refund state from positive versus unhappy customer response, so a changed communication record does not become fabricated remedy completion.
Figure 1: A positive response and a completed refund are separate records. Source: Abeler and colleagues, historical outcome definition, Jun 2009. Original comparison from the cited study. Study period and calculation limits appear in the text.

This is a particularly clear example of an outcome that can move without answering the merchant's operational question. Withdrawal establishes that a record changed. It does not establish that the product arrived, the stated explanation was correct, a remedy was completed or the customer wanted to buy again.

For human handoff measurement, keep the resolution event tied to the actual request. A refund is complete when the relevant payment state confirms it. A replacement has a separate fulfillment state. A shopper's acknowledgement can supplement those records but should not overwrite them. An apology remains part of the conversation, rather than a substitute for the transaction event.

How different were the message outcomes?

The published withdrawal rates were 44.8% after the apology, 19.3% after the lower offer and 22.9% after the higher offer. The apology versus either monetary condition was significant at p below .001. The difference between the two monetary amounts was not significant, with p equal to .330. (Abeler, Calaki, Andree and Basek, Jun 2009)

Our arithmetic comparison gives a 23.7 percentage point difference between apology and the pooled compensation arms, and a rate ratio of 2.12. The larger monetary offer was twice the smaller one, yet its withdrawal rate was only 3.6 percentage points higher. Rates are rounded published inputs, so these calculations should not be converted into invented exact arm counts. (Abeler, Calaki, Andree and Basek, Jun 2009)

The intervention was a message bundle. The apology included an explanation and different language from the payment offers. It would be too narrow to attribute the whole contrast to the word sorry. A current store that wants to test explanations, ownership or remedy choices would have to vary those elements deliberately and measure the outcome it actually cares about.

Does this show that apologies outperform remedies?

It does not support that general claim. There was no no-contact comparison, so this design cannot estimate the effect of apologizing relative to leaving the customer alone. It also did not compare a fulfilled refund or replacement with an apology under otherwise identical wording. (Abeler, Calaki, Andree and Basek, Jun 2009)

The average transaction value was EUR 23.20 including shipping. The EUR 5 offer was about 21.6% of that amount, calculated from the published average. Even that comparison does not tell us the adequacy of compensation for any particular customer's loss. A fixed amount represents a different share of an inexpensive order than of an expensive one. (Abeler, Calaki, Andree and Basek, Jun 2009)

The merchant should therefore decide the remedy from the confirmed facts and applicable order terms. Assess the communication separately: did it accurately explain what happened, make the next step clear and route an exception to a person with authority? A message experiment cannot turn an unfulfilled remedy into a resolved case.

Which support measures should stay separate?

Keep four records distinct: the issue the shopper reported, the action the store completed, the shopper's response and subsequent behavior. For example, an accepted explanation is a response, while a refund settlement is an action. A later purchase is behavior, and it needs a defined follow-up window and a customer identity that can be joined reliably.

This separation is useful when automation hands a case to a human. The system may close a chat because it successfully transferred the conversation. The human may close a ticket because a decision was made. Neither event automatically means the shopper received the outcome promised. Use explicit statuses that match those events and audit cases that were reopened.

A test of apology wording should keep the actual remedy and eligibility consistent across assigned groups, then follow the operational result. It should also preserve complaints that remain open. The historical study demonstrates how strongly a conversational intervention can move a recorded response. That is precisely why the dashboard needs a measure of the customer's underlying problem.

Which calculations can a reader reproduce?

The download retains the published inputs, their units and their comparison populations. The following arithmetic uses those inputs. It does not add a new sample or a new causal comparison.

CalculationExpressionResult
Apology minus pooled compensation, percentage points44.8-21.123.700
Apology to pooled compensation rate ratio44.8/21.12.123
High minus low compensation, percentage points22.9-19.33.600
EUR 5 offer as share of average transaction, percent5/23.2*10021.552

Which event should determine whether a remedy is complete?

A support system needs separate records for the promised action and the customer's response. Consider a refund case. A positive acknowledgement while settlement remains pending is progress in the conversation, but it is not a settled refund. A settled refund with an unhappy response confirms the action while leaving the experience problem visible. Neither record should overwrite the other.

Verified refund stateCustomer responseWhat the record supports
Pending or not confirmedPositive acknowledgementCommunication progressed; remedy still needs verification
SettledPositive acknowledgementAction completed and response recorded
Pending or not confirmedStill unhappyBoth operational and experience work remain
SettledStill unhappyAction completed; dissatisfaction remains a separate outcome

These are worked case states, not data from the historical experiment. They show why feedback withdrawal cannot substitute for the underlying task. The public study randomized messages that requested a change to a feedback record; it did not observe these refund states or measure repeat purchasing. (Abeler and colleagues, outcome definition, Jun 2009).

Which comparison would isolate the communication component?

Keep the same verified incident, remedy eligibility and operational action across groups. Then vary the communication element of interest. For example, an explanation and an apology can be varied separately while the actual remedy remains constant. A crossed design can distinguish the effect of explanation, the apology element and their interaction more clearly than comparing a differently worded apology with a monetary offer.

That is a proposed evaluation design, not a claim that the source identified the separate effect of saying sorry. The source's messages bundled a delivery explanation, apology or payment wording with a withdrawal request. A significant difference between bundles does not isolate any one sentence. The absence of a no-contact arm also leaves the effect versus silence unidentified.

What should an automated handoff transmit?

Carry the exact issue, verified action state, promise already made and the customer's unresolved concern. If a payment system reports a refund initiated but not settled, the summary should preserve that state. If the customer accepts an explanation but still awaits a replacement, the handoff should not close the operational task because the tone improved.

A team can audit reopening reasons against those two records. Cases reopened for an uncompleted action expose a different defect from cases where the action was completed but poorly explained. A closing metric that treats either as simple success can conceal work the shopper still needs.

The response-and-remedy worksheet contains these explicit review cases and the evidence needed for each state. It is not a tested classifier, customer sample or accuracy estimate. A current merchant should evaluate truthful communication and completed remedies, with later behavior measured separately; the historical payment-for-withdrawal intervention is not a current operating recommendation.

Methodology and Sources

We read the original paper and decisive methods and results spans on 13 November 2026, retained the source PDF and readable text or page images, and transcribed only the aggregate figures used here. Calculations subtract reported percentages or divide their rates. We do not reconstruct individual records, pool unlike satisfaction scales or treat a current workflow recommendation as an observed source result. (Abeler, Calaki, Andree and Basek, Jun 2009).

The messages differed in explanation, apology, request and payment wording, not merely sentiment. Outcome was feedback withdrawal, not measured satisfaction, repeat purchase or resolution. There was no silent no-contact arm. The historical payment-for-withdrawal design is analyzed, not offered as a current merchant practice.

Glossary

Continuation: Reaching the next specified stage of a process.

Percentage point: The absolute difference between percentages.

Random assignment: Allocation to a comparison group through a randomized procedure.

Selection: A difference in who remains observed after an intervention.

Resolution: Completion of the action needed to address the underlying issue.

Sources

Cite this page

Use this citation when referring to the analysis. Retain the original study scope with any figures.

offload. "Service apology statistics and the difference between withdrawal and resolution." offload, November 2026. https://offloadrep.com/guides/service-apology-feedback-withdrawal-statistics/
<a href="https://offloadrep.com/guides/service-apology-feedback-withdrawal-statistics/">Service apology statistics and the difference between withdrawal and resolution</a> (offload, November 2026)