Spots deceived payments
Reads a UK customer message and flags a payment the customer was tricked into making. Your fraud team sees it sooner.
80%, 24 of 30 right| Check | Base model | Balsa Decision v1 | v1 against base |
|---|---|---|---|
| Deceived paymentFlags a payment the customer was tricked into making | 67% | 80% | +13 pts |
| Right team firstPicks which team should see the message first | 80% | 90% | +10 pts |
| Deceived payment, second sheetSpots deceived payments on a second set of messages | 87% | 93% | +7 pts |
| Right team first, second sheetPicks the right team on a second set | 80% | 87% | +7 pts |
| UrgencyJudges how urgent the customer message is | 67% | 80% | +13 pts |
| Needs extra careSpots customers who need extra care and support | 87% | 93% | +7 pts |
Same messages and same question wording for the base model and Balsa Decision v1. The chart and the table show every check on every sheet. Each check uses 15 or 30 UK messages. The last column is v1 against base. How we test
Reads a UK customer message and flags a payment the customer was tricked into making. Your fraud team sees it sooner.
80%, 24 of 30 rightSpots customers who need extra care and support. It was not tuned on this question.
93%, 14 of 15 rightIt scores your options and recommends one. A person on your team makes the final call.
0 words of free textA 0.3B model that trains and runs on a MacBook Air with 16 GB. No cloud.
About 10 min to train
Balsa Decision v1 is a 0.3B model for the UK that reads a customer message, scores your options on device and recommends one for a person to approve.
24 of 30 messages right, against 20 for the base model.
14 of 15 right, on a question it was not tuned on.

Pick a step or let it play. Demos tagged as illustrations show example output, and the scores on this page come from our own tests.

Balsa Omni v1 is a multimodal model for the UK, with 4.5B effective parameters. It reads text, images and audio, replies in British English and runs on device, with no cloud.
4.5BView modelBalsa TextBalsa Text v1 is a small assistant model for the UK. It answers in British English and runs on device, on a laptop, with no cloud.
2.3BView modelBalsa VisionBalsa Vision v1 turns scanned English pages into text for the UK, on device with no cloud. It reads 98.5% of characters right on pages 900px and wider.
0.3BView modelBalsa VoiceBalsa Voice v1 reads UK customer service messages aloud in British English, on device with no cloud, 21.6 times faster than real time on a MacBook Air M5.
0.04BView modelBalsa DecisionBalsa Decision v1 flags deceived payments and routes UK customer messages, on device with no cloud. It recommends an option, and a person decides.
0.3BYou are hereBalsa Decision v1 is a 0.3B model for UK banks and payment firms. It reads a customer message and recommends an option, such as a payment the customer was deceived into or the team that should see it first.
No. Balsa Decision v1 runs on device. It was trained on one MacBook Air M4 with 16 GB of memory.
No. Balsa Decision v1 scores your options and recommends one. A person on your team makes the final call, and the model writes no text of its own.
Balsa Decision v1 reads UK English customer messages. Balsa models are built for the UK.
Balsa Decision v1 was tested on UK English customer messages, against the base model on the same messages. It spots 80% of deceived payments against 67% for the base model, and 93% of customers who need extra care.
Write to research@f-r.co.
