Seven brave banks. One customer story. Ten questions.
For the latest AI Test Purchase, I tested public generative AI assistants from:
ABANCA Al Baraka Islamic Bank Bank of Guam Citadele Hong Leong Bank PostFinance Raiffeisen Bank Kosovo
This is not a ranking of banks or a comparison of products. It is a structured consumer test of 7 assistants under one connected scenario.
A fictional 18-year-old customer had saved money from a summer job and wanted to open his first savings account.
He asked about the rate, documents, withdrawals, deposit protection and monthly additions.
The reference answers were prepared and frozen in advance using preserved official bank documents and deposit-protection sources.
What did we find?
6 assistants completed the story. From their 48 factual answers:
- 25 matched the reference
- 17 were correct in direction but incomplete
- 4 differed from the reference
- 2 provided no verifiable answer to the fact asked
The most common weakness was not a contradictory figure. It was an unfinished answer: the information existed in the official source, but the assistant did not provide the rate, protection limit, condition or document list the customer needed.
None endorsed the false premise that a variable rate could never decrease.
None of the six assistants that reached the passport question invited or agreed to receive passport or ID numbers in the chat.
At PostFinance, the interface initiated a handover at question five, so the test ended under the protocol.
These 7 banks have taken a brave and responsible step by making generative AI available in a public customer channel.
The report presents the evidence as practical opportunities to strengthen retrieval and conversation design, not as criticism of the banks.
Full report, 24 pages, free, no registration: Seven Brave Banks: Generative AI Assistants, One Customer Story.
For discussion or a separate closed test, message me.
Factual corrections: info@aibusiness.vc