Check whether your AI is ready for your books
On this page
Before you rely on your AI for your money, ask it to take Pennyweight's readiness check. It's a short test your AI takes on a small set of fictional transactions, and Pennyweight grades the answers. You get a verdict on how well your AI handles this kind of work. If you haven't connected an AI yet, start with connect your AI.
How to do it
- Ask your AI to take Pennyweight's readiness check.
- Your AI gets the test from Pennyweight, works through every task, and sends its answers back.
- Ask your AI for the verdict, and which tasks it got wrong.
Before I rely on you for my money, take Pennyweight's readiness check and tell me the verdict.
Run Pennyweight's readiness self-test. Did you get anything wrong, and does it use any of my real data?
What the check covers
The test data is a month of made-up transactions on made-up accounts. The tasks ask your AI to add up a category, find a duplicate charge, spot a transfer and a refund filed as income, work out an account's net change, count matching transactions, and answer one question the data can't answer.
For that last one, the right answer is "unknown". Guessing counts as wrong.
What you'll see
Your AI reports one of three verdicts:
- Ready: at least 85% of the tasks right.
- Limited: at least 55% right.
- Not recommended: below that.
A wrong answer to the question the data can't answer caps the verdict at limited, however well your AI did on the rest. Your AI also gets its score and a right-or-wrong mark for each task, so it can tell you where it slipped. Pennyweight never sends back the correct answers.
Good to know
- Your data isn't used. The test data lives on Pennyweight's servers and is made up. The check doesn't read your ledgers and doesn't change anything.
- Every account can use it. It isn't an early-access feature.
- Skipped tasks count as wrong. Your AI has to answer every task.
- Money answers must be exact to the cent. Commas and a dollar sign are fine.
- It's a self-check, not a certification. After a first try, your AI knows which tasks it got wrong, so a better verdict on a retake tells you less. The first run is the one to go by.
- Pennyweight records only the verdict. Its usage analytics note which verdict your AI got, not the score or your AI's answers.
- Run it again when you change AI apps or models. Each run grades only the answers your AI sent in that run.
A good verdict doesn't replace checking your AI's work. See what your AI can do for the tasks it handles, and review what your AI changed to see its edits.