IslandRemit is a remittance operator — moving money across borders for customers who send cash home to family and friends.
The queue doesn't have to stay stuck.
“The people who need us most, the fraud cases, wait in the same line as someone asking a question we've answered four hundred times this month.” — Ishara Bandara, Founder & CEO
Founded 2018 · 42 staff · international money transfer
Source: Naveen Jayasena, Head of Support, internal email — confirmed against 6 months of ticket data (61.5%)
The 6-month log (3,463 tickets) validates the category mix, not total volume — it's used here to prove the clustering pattern, not to restate the monthly count.
A normal day in the queue — click a ticket
“My cases sit in the exact same queue as someone asking why their transfer fee changed by forty cents.” — Zara Hameed, Fraud & Risk Lead
Every question gets classified first — click a step to see what it does.
“I don't think we need more people. I think we need the routine 80 percent handled without a human at all, so my cases stop waiting behind them.” — Zara Hameed, Fraud & Risk Lead
This is the judgment call that used to run on gut feel — a person skimming each ticket to guess if it's routine or urgent. Claude does that classification in structured output, before a person ever sees the ticket.
Routine questions get an instant answer. Risk gets a human.
A fixed suite of real-shaped questions runs through the actual interface before anything ships — the security cases are zero-tolerance.
Types into the actual input, clicks the actual chips — not a mocked function call.
Any edit to the FAQ or the interface re-runs the full suite automatically.
Broken layouts, silent failures, and mis-routed fraud cases get caught pre-release.
Same pattern — answer the safe stuff, hand off the rest — at the companies IslandRemit actually competes with.
Runs a Messenger-based support bot for routine functions — sending money, checking transfer status — so human agents are freed up for what actually needs a person.
Partnered with Oscilar in Nov 2025 on AI risk decisioning — fast-tracking trusted transactions automatically while flagging suspicious ones for a person, in real time.
Built its own transfer bot on Messenger using its API — guiding customers through sends and rate alerts, then stepping aside the moment someone needs more help.
Rebuilt its help center with AI-enhanced search plus a virtual support assistant that resolves issues 4x faster than a live associate — cutting average support time by 75%, without removing the option to escalate.
Deployed an AI digital assistant (built on ServisBOT) that automates 60% of live chats — tracking transfers, checking refund status, routing anything unresolved straight to a live agent.
The more the assistant handles on its own, the fewer tickets ever need a person to look at them.
62% repeat-question share: confirmed against 6 months of real ticket data (61.5% of 3,463 logged tickets). Resolution-rate range: Intercom Fin production data (45–53%) and vendor benchmark (67%), and Anthropic's own real Fin deployment (50.8%). Applied only to the repeat-pattern share — not a promise, a way to size the opportunity. In the same ticket log, the repeat categories averaged 12 minutes to resolve versus 145 minutes for fraud & flagged-review cases — the gap this fix is meant to close.
Every question it can't answer gets logged. The team reviews, adds the answer to the FAQ — no retraining, just a text edit — and it's grounded from that moment on.
A question the FAQ can't cover gets flagged automatically, not guessed at.
The support team writes the real answer once, the way they already do today.
No model retraining. The next customer who asks it gets the answer immediately.
Illustrative trajectory, not a forecast — modeled as gradually approaching Intercom Fin's published resolution-rate ceiling (67%) as FAQ coverage grows. Actual pace depends on how often the team reviews escalated tickets. First candidate: "Cannot log in" — 10% of ticket volume in the real log, not yet covered by the FAQ at all.
Thank you.