Signals
Most small-business money leaves a trail before it reaches the books: a mobile money SMS, a printed receipt, a supplier's delivery note. Signals turns that trail into draft transactions so a day of trading becomes a review job rather than a typing job.
What it captures
Mobile money messages. Import the confirmation messages your phone already receives — MTN MoMo, Airtel Money, M-Pesa. Each is parsed into amount, counterparty and date.
Receipts. Photograph or upload a receipt and it is read into a draft transaction.
The flow
- Import messages or upload receipts.
- SBB parses each into a proposed transaction.
- High-confidence items may be auto-cleared; the rest wait for you.
- You confirm, correct or reject each one.
- Confirmed items become real transactions in your books.
Reading the counters
| Counter | Meaning |
|---|---|
| Signals captured | Everything brought in |
| Messages processed / Receipts processed | How much has been read |
| Successfully parsed | Understood well enough to propose a transaction |
| Transactions created | Confirmed and in your books |
| Auto-cleared captures | Confident enough to post without you |
| Unposted signal | Parsed but still waiting on you |
| Signal parse failures | Could not be read — see below |
| Receipts still processing | In progress |
| Failed — retry | Something went wrong; retry it |
| Expenses still waiting on receipt support | Spending with no proof attached |
| Missing proof | The gap that matters for tax and audits |
When parsing fails
Analyze again retries. Common causes:
- A blurred or badly-lit receipt photo — retake it in better light, flat
- An unusual message format from a provider
- A faded thermal receipt where the ink has already gone
Anything that cannot be read can still be entered by hand. Do not leave it — an unrecorded expense is worse than a slowly-typed one.
Missing proof
Expenses still waiting on receipt support is the number to watch. Spending without a receipt is real money you may not be able to claim, and is exactly what gets disallowed in a tax review.
The needs evidence lane in AI workflows gathers these so you can chase them in one sitting, or hand them to whoever spent the money as an evidence pack.
Getting the most out of it
Import regularly. Weekly is fine; daily is better. Old messages are harder to identify because nobody remembers what "payment to J. Okello" was for.
Correct rather than reject. A correction teaches the AI your preferences — see trusted patterns — so the same counterparty is categorised correctly next time.
Photograph receipts at the moment you get them. The paper fades, and the memory fades faster.
Watch for duplicates. A MoMo message and a manual entry for the same payment is the most common double-count. The duplicate lane catches these.