Why this matters
One complaint handled well is customer service. Thousands arriving in public at once is a test of whether the business decided anything in advance. Replies slow down, remedies vary by agent, the press quotes the worst thread, and the monthly report later shows a jump in mentions that someone calls awareness. The service recovery guide sets out the sequence for one customer; at this level the work is running it at volume, in public, with the evidence kept for afterwards.
The app and its outage weekend
Swiftrail Tickets is an invented app that sells train tickets on behalf of rail operators. On a busy event weekend a payment release failed: customers were charged, saw "booking confirmed" and received no ticket.
Swiftrail Tickets: the outage in figures
| Day | Public complaints | Replies sent | Median first reply | Press articles |
|---|---|---|---|---|
| Saturday | 2,400 | 310 | 3 h 40 min | 1 |
| Sunday | 3,100 | 820 | 2 h 10 min | 4 |
| Monday | 1,200 | 900 | 55 min | 6 |
| Tuesday | 450 | 410 | 40 min | 2 |
- Failed bookings: about 11,800; double charges: about 2,050.
- Complaint types: charged with no ticket 48%; bought a second ticket at the station 22%; charged twice 17%; penalty fare from a train operator 6%; other 7%.
- Share of mentions among three ticket apps in a normal week: Swiftrail 28%. Outage week: 71%, of which 83% negative.
- Weekend cover: four social agents per shift, no on-call communications lead. Agents could refund only after a supervisor's approval.
From first replies to recovery
At volume the sequence still runs acknowledge, apologise, understand, fix, make amends, follow up and learn, but most of it has to be decided before the next complaint arrives. A pinned public statement carries the facts once; individual replies point to it and move each case into a private channel. Remedies that are rights, such as refunding a charge for a ticket never issued, are not goodwill and should never wait for approval. The community management guide separates a complaint from a crisis: this weekend was a crisis from the first hour, and the plan has to say who decides that.
Task
- Write the recovery sequence and the reply guidelines for public channels.
- Set who may offer which remedy, including the limits and what Swiftrail cannot remedy itself.
- Write the escalation path, with triggers and owners.
- Set what to monitor and what to log for every complaint and every article.
- Plan how sentiment and share of voice will be reported over the following weeks.
- Describe the root-cause loop back to the product team.
Worked answer
Sequence and replies. Within the first hour, pin a statement on every channel: what failed, that no one will be left out of pocket, how refunds will happen and when the next update comes. Pause scheduled posts. Every reply is short, specific and calm, names the problem in the customer's words, links the statement and offers a private route with a reference. No booking details in public, no arguing, no promises without a date. Post updates at the times promised, then a closing note once refunds are complete.
Remedy authority
| Who | May offer without approval | Limit |
|---|---|---|
| Any agent | Refund of a failed booking; refund of a duplicate charge; refund of a replacement ticket with proof | Up to the original fare plus £10 goodwill credit |
| Shift lead | Extra costs from missed connections, such as a taxi, with proof | Up to £150 per booking |
| Head of customer operations | Larger claims, vulnerable customers, legal threats, press-linked cases | Case by case |
Penalty fares are issued by operators: agents send a confirmation of the failed booking for the customer's appeal, and refund the fare if the appeal fails.
Escalation. Agent to shift lead for anything outside the matrix. Shift lead to the incident owner, the head of customer operations, when complaints about one issue keep rising hour on hour, a journalist asks a question, or safety, vulnerability or legal threats appear. The incident owner calls in the communications lead, product, payments and legal, decides public statements and updates the agents' guidance. Weekend on-call cover for that list becomes permanent.
Monitoring and logging. Queries for the brand, common misspellings, the app's name with "refund", "charged" and "ticket", the rival apps for comparison, and the main rail news outlets. Each complaint is logged with time, platform, type, private reference, remedy, status and a root-cause tag. Each article is logged as the media monitoring guide describes: title, tier, prominence, sentiment, whether Swiftrail's statement appeared and which rivals were mentioned.
Reporting. Daily during the incident, then weekly for two months. A huge share of mentions in the outage week is notoriety, so report Share of VoiceYour share of market visibility. split by sentiment against a fixed set of rivals and titles, alongside complaint volume by type, response time against standard, refunds completed and app store rating. The recovery KPIA metric tied directly to an objective. is time from complaint to refund, and the readout ends when negative share returns towards the pre-outage pattern. Report press audience figures as context, not as ReachUnique people who saw an ad. achieved. The share of voice guide explains why the denominator must stay fixed.
Root-cause loop. Tag by cause, not symptom: a release without rollback, confirmation shown before a ticket existed, no status page, tickets not stored for offline display. Take counts and customers' own words to the product review each week until each cause has an owner and a date, and tell agents when a cause is removed.
Common mistakes
- Reading the spike in mentions as awareness. Split share of voice by sentiment.
- Treating refunds as goodwill. Charges for tickets never issued are owed, not granted.
- Escalation as a queue. Triggers and owners decide, not whoever is senior and awake.
- Closing cases without tags. The same release fails again.
Bring the whole level together in the social programme strategy capstone, or revisit community and reporting at the intermediate social media level.
