Spoiler Alert

Inspiration

Linda is a retired teacher. Her "grandson" calls from jail, and a "lawyer" tells her to wire $8,500 and tell no one. Sound did everything right. The teller asked if she was sure, and Sound held the wire for three hours. She sent it anyway. The next day, the scammer called back for $4,250 more. That second call is what finally tipped her off, one day too late.

When we dug into the challenge data, we found Linda's story isn't rare. In every case where a held wire was released, the scammer came back within two days for exactly half again. Scams follow a script, and every script has a sequel. That gave us our question: what if members knew the sequel before it happened?

What it does

Spoiler Alert works in three steps.

1. Spot it. It flags a "rushed first wire" using three fields Sound already has:

2. Spoil it. Before the wire can go, the member answers one required screen showing the scammer's next three moves: keep it secret, send more money, get cash. Any "Happened" keeps the wire held and brings a teller over with questions built for that scam. If every answer is "Not yet," the member can continue. It's their money and their choice. The screen also gives them a true line to buy time: "My credit union is holding this payment for review."

3. Save it. The predictions are texted to the member's phone. When the scammer calls back, one tap opens a case and starts the wire recall.

How we built it

These are synthetic results, not real fraud rates.

Challenges we ran into

• The member decides. For a payment the member authorizes, Sound can pause and advise but not block. A "be careful" warning loses to a scammer saying "don't tell the bank." We needed something the scammer can't coach around, and a prediction about the scammer's own next move fit.
• False positives. Our first triggers flagged home buyers, contractors, and members who really moved. Adding "payee added less than an hour ago" separated scams from home closings, where the title company was added days earlier.
• The data is cleaner than real life. Catching 24 of 24 won't hold in practice. We tested stricter settings to find the scams we'd miss (at 3x, a $5,000 wire slips through) and designed a one-tap recovery path for them.
• Knowing which questions to ask. The same rule flags a fake grandson, a fake federal agent, and a real car buyer. One open question, "What is this payment for?", lets the member's answer pick the teller's script.

What we learned

• Generic warnings and "Are you sure?" checks don't work. The right question at the right moment does.
• Prebunking research backs this up. In one study, teaching people a scam's tricks cut their willingness to fall for it by about 40%, but the effect faded without reminders. That's why we show the spoiler at the moment of payment and then text a reminder.
• The best protection keeps the member in control. Linda still decides. We just make sure she knows how the story ends.

What's next

For Phase 2, we'll build a clickable prototype of the Spoiler Screen and the teller card. Our 90-day pilot plan runs 30 days in silent mode to tune the thresholds, then 60 days live at a few branches.

Every scam has a sequel. We spoil it.

Built With

• excel