Automated In-Flight Wifi Refund Processing for a UK Airline Group
- Client
- Airline group · United Kingdom · dedicated mailbox for in-flight wifi refunds
- Industry
- Retail Customer Service
- Department
- Customer Service & Contact Centre
- Region
- United Kingdom · UK & Ireland
A generic "the wifi was bad, refund me" email should not need a person to read it before the right form goes out, and every one that does is a person not handling the cases that need judgement. We built the flow that takes a free-text complaint to a checked, automatic refund, or to a ticket for the refunds team when something does not add up.
A working model of the approach on sample data. It plays on its own and pauses when scrolled away; nothing here touches a live system.
The results
- 1
- dedicated mailbox as the trigger
- Single-purpose, so no intent classification was needed. Scope, not a saving.
- 2
- stages from email to decision
- Collect the purchase details by form, then match and check eligibility. Scope, not a saving.
- 0
- manual triage steps in the flow
- Every request ends as a refund or a ticket for the refunds team. By design; no volumes were recorded.
The challenge
Our client is a UK-based airline group that sells wifi on its flights. Customers unhappy with the connection email a dedicated refund mailbox in their own words, something like "Wifi was terrible, I'd like a refund."
Every one of those emails had to be read, matched to the right transaction in the airline's custom transaction platform, checked for eligibility and then processed or escalated. By hand, at volume.
What it was costing them
- People reading generic emails. A free-text complaint gives nothing to act on until someone reads it and asks for the purchase details.
- Matching by hand. Each request had to be found in the transaction platform before anything else could happen.
- Eligibility checked by eye. Purchase date, refund window and the other rules were applied manually on every request.
- Judgement spent on routine cases. Every routine refund processed by hand is time not spent on the genuinely exceptional cases.
- Free-text complaint email
- Read by a person
- Transaction matched by hand
- Eligibility checked by eye
- Refund processed
The approach
We built a single-purpose, rules-based automation that runs from the complaint email to a processed refund or a ticket for the refunds team, with no manual triage step in between.
From complaint email to refund or ticket
- Single-purpose trigger. The automation watches the dedicated wifi refund mailbox. The mailbox itself is the signal, so no communications mining or intent classification was needed.
- Data collection. Every new email gets a reply with a form that captures what is needed to identify the customer's purchase.
- Transaction matching. When the completed form comes back, a second stage looks up the transaction in the airline's custom platform.
- Eligibility rules. The transaction is checked against the refund rules, including whether the purchase still falls within the refund window. No refund is issued before every rule passes.
- Refund and confirmation. Found and eligible: the refund is processed automatically and the customer receives a confirmation email.
- Exception ticket. Transaction not found, or any other issue: an internal ticket is created for the refunds team to review, instead of the automation failing silently or guessing.
- Refund mailbox
- Form sent to customer
- Transaction lookup
- Eligibility rules
- Refund or ticket
Why there is no AI layer
A mailbox used only for wifi refunds already says what every email is about. Adding intent classification would have added something to train, monitor and maintain, and something that could misclassify, without adding a decision the flow needed. The structured form gives the rules everything they need to decide.
Technology
| Layer | What we used |
|---|---|
| Trigger | Email automation and mailbox monitoring |
| Data capture | A web form sent to the customer |
| Transaction platform | Integration with the airline's custom transaction and refund platform |
| Decision | Rules-based eligibility checks, including the refund window |
| Exceptions | Internal ticket creation for the refunds team |
The outcome
Every request now ends in one of two places: a processed refund with a confirmation email, or a ticket for the refunds team with the case to review.
| Before | After | |
|---|---|---|
| Reading each complaint to work out what it is | A person | Not needed: the mailbox is the signal |
| Collecting the purchase details | Case by case | Form sent automatically |
| Finding the transaction | By hand | Automatic lookup |
| Refund window and eligibility rules | Checked by hand | Enforced before every refund |
| A request that does not add up | Handled by hand like the rest | Ticket for the refunds team |
- No manual triage step between the complaint email and the refund or ticket.
- Refunds issued only after the eligibility rules pass, including the refund window.
- No silent failures and no guesses. Anything the automation cannot complete becomes a ticket.
- The refunds team works the exceptions, not the routine eligible refunds.
- Not quantified. No volumes or timings survive for this engagement; the figures above describe scope, not savings.
The risk most teams miss
A refund robot that guesses is worse than no robot, and one that fails silently is worse still: the customer waits, nobody knows, and the complaint comes back angrier. When a transaction cannot be found, the safe answer is a ticket for a person with the case attached. Design that path before the happy path.
The multi-intent sibling of this build: email triage and one-click agent assist for a car rental group, where five request types arrive in one inbox and every email is classified with a confidence score.
Built with
The tools this ran on, by the layer each one carried.
- Apps & integration
- Automation
Related work
Retail Customer Service5request types automatedEmail Triage and Agent-Assist Automation for a Car Rental CompanyCar rental · Europe · multi-billion-euro rental group, hundreds of thousands of vehiclesRead the case
Energy & Utilities~120group companies covered every monthEnergyLink Invoice Reconciliation Automation for a US Oil ProducerOil and gas producer · United States · NYSE-listed, about 120 group companiesRead the case
Energy & Utilities20+automations in productionIntelligent Automation Program for an Italian Multi-UtilityListed multi-utility · Italy · five provinces, energy, water, waste and public servicesRead the case
Next step
Start the same conversation
Tell us the process that keeps breaking. We will come back with what it would take to fix it, what it would cost, and what we would measure.
Or write to us directly: •••••@•••••••.•••
- A reply from a partner within one business day
- Fixed scope and fixed fee, agreed before we start
- The maps and the backlog are yours either way