October 2026

In short

3,091 contacts reached Customer Support in October, from 2,610 distinct people. That is 5% fewer than September and 8% fewer than last October. Customer satisfaction rose to 87%, up from 85.0% in September and 81% a year ago.

The month was shaped by Storms Austen and Boelo (22–26 October), which cancelled and downgraded trains across the South West. Upgrade refund requests through the refund form hit 88–104 a day, against about 35 normally, and the refund form's monthly total was the highest since August.

New this month, as asked: what each contact costs, how much of the contact was avoidable, who the people contacting us are, and how many are helped before their train leaves.

The headline numbers

All figures in this section come from Zendesk, our customer support system. They count people who contacted us, not bookings or app users. Exceptions are labelled where they appear.

Once merged duplicates and test tickets are removed, the 3,091 contacts come from 2,610 distinct people. 268 of them (10%) contacted us more than once in the month.

We sent a first reply within a median of 21 hours (4.2 working hours), the same as September. This is the wait for a reply from a person: replies from our AI first responder aren't counted, because Zendesk doesn't log them as a ticket update. Cases took a median of 40.5 hours to close, down from 44 hours. 3.9% of cases came back to us after we had closed them, level with September.

Contacts ran at about 140 a day during the storms (22–27 October) against about 85 a day for the rest of the month.

Booking database cross-check (booking data, not Zendesk)

The refund rate rose from September and is double last October's, on slightly lower sales. October had two storms and an overhead-line failure at Stafford (10 October), but the year-on-year climb reported in August and September continues underneath.

Satisfaction trend

July to September are taken from earlier reports. October is the dashboard figure.

What a contact costs

The department figure divides everything Customer Support costs by every contact received (3,091), including refund forms, because an agent still handles each one. The handling figure shows why the department figure is higher: three people are on the books, about one and a half are on tickets. People are about 91% of the cost. Recovering Statutory Maternity Pay would lower the department figure slightly; Finance to confirm.

How much of the contact was avoidable

"Avoidable" is a judgement, so the contacts are shown in groups rather than as one figure. Out of 2,942 contacts after removing duplicates:

About 1 in 9 contacts came from a customer misunderstanding what they bought, or when it would arrive. The upgrade refund form and account deletion requests are routine form-based processes: 1,616 contacts between them, or roughly £5,350 of the month's cost at £3.31 each.

Rail ticket change requests (254 contacts) are not counted as avoidable. Each one is created when a customer presses the change button in the app, which is how customers ask to change a rail ticket. Where a request has no reply from us, the customer closed it themselves, usually because they rebooked and claimed a refund on the original ticket, or decided not to change.

Who contacts us

Spend and purchase history come from the booking database, not Zendesk. They are matched only where a ticket carries the customer's Seatfrog account number, which the upgrade refund form collects, so these are samples that lean towards upgrade customers whose train was cancelled.

September's 2,645

Of the 2,645 people, 251 (9%) contacted us more than once in September. 345 left their Seatfrog account number on a ticket; this is what the booking database shows for them:

• On average each had bought 8 upgrades worth about £230 before getting in touch. 80 of the 345 (23%) first bought an upgrade in August or later.
• After contacting us, 126 of the 345 (37%) have bought another upgrade by 12 November, six to ten weeks later. Among those who had bought before, it is 106 of 244 (43%).
• One-and-done: of the 48 who had bought exactly one upgrade before contacting us, 35 (73%) have not bought again so far.

October's 2,610

A random 300 of October's contacts with an account number were matched in the same way. 15% had only ever bought one upgrade, 36% had bought 2–5, 27% 6–20 and 21% more than 20. On average they had bought 17 upgrades worth about £430 in their lifetime. 27% were new customers since September; 67% had bought in the twelve months before October.

There is no comparison figure yet for customers who did not contact us, and the standard measure (no purchase within 90 days) can't be read for September until the end of December. An analysis in July found that being refunded does not make customers less likely to buy again, while an unhappy satisfaction rating does (about 10 points more likely to stop buying).

Helped before the train leaves

Departure times come from the booking database, not Zendesk. This is a random sample of 100 contacts that carried a sale reference.

• 78 of 100 arrived after the train had already left. Most were refund requests after a cancellation, where "before departure" doesn't apply.
• 22 of 100 arrived before departure, typically just 42 minutes before the train left (14 of them within two hours).
• None of the 22 was resolved before departure. 3 got a first reply in time, all of them sent a day or more ahead.

With a median first reply of 21 hours, a contact sent less than an hour before departure is not answered before the train leaves. Customer Support went from four people to two on tickets this year, so reply times are not expected to improve.

Why people asked for a refund

The refund form (1,379 requests) records one of two reasons:

• Train cancelled: 919 (September 720)
• Downgraded from First Class on the day: 491 (September 336)

Both rose with the storms. Every one is a contact an agent handles, and together they are almost half of everything Customer Support received.

Everything else (264 requests, about 9% of contacts) is triaged by an agent, who records a reason:

User error: is the app clear enough?

105 of 264 hand-triaged refund requests were people who bought something other than what they meant to. 80 of them were upgrades. Wrong route and wrong time are now level, and wrong date has fallen back from September's peak. The Customer Support team's reading of the booking journey still stands: the date is easy to miss on the search results, when adding a trip, on the auctions screen, and on the place-a-bid screen.

180 contacts (5.8%) ended with the customer not eligible for a refund or change, or with the booking found to be their own mistake (September 236, 7.3%). Each contact is counted once:

The 65 other own-mistake contacts are almost all customers who bought an upgrade for the wrong train:

• Product: 44 upgrades, 2 rail tickets, 2 Secret Fare; 17 not recorded.
• What went wrong: where the agent recorded a reason (34), it was the wrong time (10), wrong date (10), wrong route or direction (9), wrong number of seats (1), and 4 others (change of plans, seated apart, health). Where no reason was recorded (31), the customers' own subject lines mostly describe the same thing: "wrong train", "wrong train time booked", "upgraded the incorrect train", "purchased in error". A handful were about something else, such as a second payment or a bid they wanted to cancel.
• Outcome: 44 of the 65 were given a goodwill credit to their Seatfrog account instead of a refund.

The 34 with a recorded reason are also counted in the user-error table above.

How each product performed

Products are attributed from the product field (set on 26% of tickets) and from tag families, which together cover about 92%.

Upgrades

1,612 people, 62% of everyone who got in touch (1,535 in September). 40% wanted their money back, 27% needed help because their train was cancelled, and smaller groups complained about the journey on board (5%), had no First Class available or were downgraded (4%), wanted to change the class of their upgrade, mostly Standard Premium against First (4%), or couldn't use their upgrade (4%). Upgrade customers were the most likely of any product to be unhappy in their first message (36%). Satisfaction on rated upgrade contacts was still the highest of any product, at 93%.

Superfare

248 people, down from 557 in September, when the Microsoft email outage nearly doubled Superfare contact. 28% asked for a refund, 12% to change the date, 9% about a cancelled train and 7% to change train. 37 exchange requests and 11 refund requests weren't eligible under the fare rules (September 44 and 34). Of the people chasing a "missing" Superfare ticket in the last week of October, none had a ticket that was actually missing; tickets are issued the day before travel.

Rail ticket bookings

278 people. 43% wanted to change the date and 16% the train, mostly raised with the change button in the app; 15% asked for a refund and 15% about a cancelled train. Only 4% were unhappy in their first message, but four customers told us that cancelling a change request isn't clear (see Feedback). Two customers were charged for a booking that failed (September 3).

Secret Fare (28) and Train Swap (15)

Both small. Secret Fare window confusion: 0. Customer Support raised Secret Fare tickets being rejected at the Cambridge ticket gates on 12 October; the tickets were valid on our side.

Account deletion (258)

Handled cleanly again: none came back.

Where customers were least happy

Upgrades had the most customers unhappy when they first wrote to us. Secret Fare and Superfare had the most cases reopened after we closed them, and the lowest satisfaction among rated contacts (Superfare 77%, rail tickets 78%).

Standard Premium confusion

Customers buying an upgrade believing it is First Class and receiving Standard Premium.

Down from September's 65, still above August. 39% were unhappy in their first message, 35 of the 46 were logged as customer error, and 10 were not eligible for a refund. The weekly round-ups noted they were almost all on Avanti. A formal complaint at the end of the month argued that the partnership is marketed as First Class; the Customer Support Manager replied that both classes are offered and the class is shown before bidding.

Early November is running lower: 12 contacts in 1–12 November against 21 in 1–12 October. It's early days, but the new screens from the First Class Value experiment look to be helping.

Recurring themes

Feedback: what customers told us about the app and the service

17 contacts from 16 people carried direct feedback (20 in September). Every one was read in full, including agents' internal notes; agents left a feedback note on 15 of the 17. Another 103 contacts were negative feedback, mostly about the service on board (no First Class carriage, catering, overcrowding). Those are the train operator's service, so they are counted here but not themed.

Changing a rail ticket

Customers who started a change and then wanted to keep their original ticket couldn't find how to cancel the request, and agents cancelled them by hand. Related: 13 customers received a false "no ticket" email after an exchange that had worked, and Android briefly showed "No tickets here" after an exchange, reportedly fixed in an app update on 20 October with no reports since.

Bids

"Retracting a bid is extremely unintuitive."

"Not being able to pull out of an auction will make me avoid your service in future." From an unhappy rating

Refunds in the app

"It would be better to be able to do it through the app, rather than try and navigate the website to find the correct area to submit the claim form." From a five-star rating

That's now four months running of customers saying the refund route is harder than it should be. An agent replied that it is something the team are already working on.

Seats and accessibility

Three customers asked to see or choose a seat before paying: one travelling with a dog, one anxious traveller who needs to face forward next to their partner. Two accessibility signals stood out: a parent whose autistic child needs to sit with them, and a customer with a brain injury asking for calendar and wallet trip files, who wrote "I think your app may not be useable by me."

What customers told us in their ratings

Rating counts and comments come from the ratings themselves. The headline 87% is the dashboard figure.

365 people rated us (12% of closed cases): 321 happy, 44 unhappy. 24 of the unhappy ones explained why:

Slow replies fell from 8 in September to 4. Disagreement with a fare-rule outcome is again the largest group.

"The response came after my planned journey so really not any help."

"Your refund process is over complicated. It appears to be designed to present barriers to proceeding with the process."

When people were happy, it was speed, straightforward refunds after cancellations, and the people they dealt with, with Richard and Ella named most often.

"Many trains were cancelled due to the weather, I appreciated the prompt decision."

Problems inside five-star ratings:

"It might be worth reviewing the words used on the app to say maximum bid rather than opening bid?"

"Took me a little time to find where I applied for refund for an upgrade rather than a full ticket."

"Initially I got a standard reply from an AI chat bot which I found quite disappointing."

Trustpilot

Holds at 4.6 from 1,632 reviews (1,559 at the start of October), 502 of them in the last twelve months. 75% five-star, 17% one-star, 8% in between: as polarised as ever. The latest one-star review was about a two-seat bid that dropped to one seat after the auction started, and about receiving an AI reply.

App stores

iOS: 4.8 from 48,730 ratings (47,900 in September). Of nine written reviews in October, seven were one-star: bids that can't be withdrawn or trains that can't be changed, not finding the bidding section, being sold the return leg first, and too many marketing emails. Android: 4.8 from about 8,850 reviews (about 8,580 in September).

Our AI first responder

It opened 553 contacts (18%, down from 26%). 70% of those were passed to a person and 7% were fully resolved by the AI. Satisfaction on contacts it opened was 69.5% against 91.5% elsewhere (rating data; 59 and 306 rated). The gap is statistically real but is not a fair measure of the AI: it opens the hardest conversations (refund requests, date changes, class changes), and most of them end with a human. Two unhappy ratings and one Trustpilot review objected to receiving an AI reply.

What Customer Support raised with the rest of the business

• Android showing "No tickets here" after a rail purchase paid with credit, and after exchanges. Reportedly fixed in app updates on 14 and 20 October; no reports since 20 October.
• False "Customer has no ticket" emails after successful exchanges: 13 customers in the month. Still open with engineering.
• Payments: a duplicate charge, an upgrade charged but not issued, and two exchanges that needed replaying.
• Superfare terms: no link to Seatfrog's terms in the Superfare flow, and no fallback when the ticket email fails (a customer claimed about £100 in losses).
• Secret Fare tickets rejected at the Cambridge gates (12 October).
• Train Swap: a service shown on the website but not in the app; no LNER swap stock before 09:00.
• "No seats available" shown on an open auction: the message needs to explain what is happening.
• Buy Upgrade Now after departure missing since about 21 September (raised with Commercial).
• Account deletion failing for re-registered emails: engineering's fix accepted on 20 October.

For context: Stafford overhead-line damage (10 October), Storms Austen and Boelo (22–26 October), a TransPennine strike called off (19 October), savings now worked out from walk-up prices (28 October), and app releases iOS 4.5.0–4.5.1 and Android 4.15.3–4.16.0.

Main issues this month

1. The storms pushed refund contact to the busiest days of the autumn. 919 cancellation and 491 downgrade refund forms, 88–104 a day at the peak. Booking data shows 2.2 refunds per 100 upgrade sales, double last October. Each one is a contact an agent handles, at £3.31 each.
2. About 1 in 9 contacts (325) came from customers who bought the wrong thing or misread what they had bought. Wrong route and wrong time are level at 32 each among triaged refunds. Standard Premium mistaken for First accounts for 46 more; Product are working on a solution.

The chain runs through the report: someone books the wrong thing, contacts us, isn't eligible under the fare rules, and leaves an unhappy rating about the decision. That was 9 of the 24 explained unhappy ratings this month.

Method and caveats

• Windows. Period 1–31 Oct 2026 (31 days); previous 1–30 Sep 2026 (30 days); year-ago 1–31 Oct 2025. UK local time. Per-day figures are given because the months differ in length.
• Satisfaction for October (87%) and October 2025 (81%) are Zendesk Explore dashboard figures supplied by the operator; September (85.0%) is the published figure. Ticket-level rating figures (365 rated, 87.9%) are used only for counts, reasons and comments.
• Contact counts come from a Zendesk export of 3,091 tickets. 147 merged duplicates and 2 test tickets were removed for the distinct-people figure (2,610) and the avoidable-contact groups (2,942). September's published 3,255 now counts 3,245 after deletions.
• Cost per contact uses all three Customer Support staff fully loaded (employer National Insurance and pension), Zendesk's annual licence spread monthly, and monthly AI tool costs, divided by all 3,091 contacts. Individual salaries are not shown anywhere.
• Avoidable-contact groups use the agent-recorded refund reason, the not-eligible and wrong-booking markers, the Standard Premium confusion marker, and the "where is my ticket or booking" requests. Groups overlap; totals are distinct contacts.
• Who contacts us (booking database). September: 345 of 2,644 people had a Seatfrog account number on a ticket; purchase history is successful upgrade purchases only, counted before the first September contact and from the following day to 12 November. October: a random 300 of 1,501 people with an account number. Both lean towards refund-form users. Rail tickets, Superfare and Secret Fare spend are not included.
• Helped before the train leaves (booking database): a random 100 of 1,520 October contacts with a sale reference, joined to the auction's departure time. Small sample, and it leans towards the refund form; contacts without a sale reference couldn't be timed.
• Products are attributed from the product field (26% coverage) and tag families (about 92% together); 250 contacts couldn't be attributed.
• "Unhappy in their first message" is our system's automatic reading of the opening message's tone. "Got back in touch" means the case was reopened after we closed it, which includes customers replying "thanks".
• Reply and closing medians exclude 649 tickets (21%) with no reply metric (forms or automated replies only). Treat as directional. First reply time is Zendesk's measure of the first public reply from the team; replies from the AI first responder are not logged as a ticket update, so they don't count. 21 hours is clock time (nights and weekends included); 4.2 hours counts only business hours.
• Standard Premium early-November comparison counts distinct tickets carrying the confusion markers, 1–12 Nov against 1–12 Oct.
• Trustpilot read 12 November; iOS from Apple's public feed and Android from Google Play, read 10 November.
• Not done: a 90-day "never came back" measure for September's contacters (not possible until the end of December), a comparison group of customers who didn't contact us, and Linear issue states beyond what was posted in Slack.
• Customer identities have been removed from all quoted material.