Understanding Confidence Scores

Understanding confidence scores

Every transaction Penny categorises comes with a confidence score — a percentage that tells you how certain the AI is about its categorisation. This system ensures accuracy whilst keeping your workload to a minimum.

What do the scores mean?

Confidence scores are divided into four tiers, each triggering a different workflow:

95% and above — Auto-categorised

Penny is highly confident and categorises the transaction automatically. You don't need to do anything, though you can always review and correct if needed.

This typically applies to:

• Recurring payments to well-known merchants (e.g., your monthly phone bill)
• Transactions you've previously confirmed or corrected
• Common business expenses with clear merchant names

80–94% — Suggested category

Penny is fairly confident and suggests a category, but wants you to confirm. You'll see the transaction highlighted in your dashboard with Penny's suggestion and a "Confirm" or "Change" option.

Via WhatsApp, Penny will send a message like: "I've categorised your £45.00 payment to Ryman as 'Office supplies' under Office costs. Does that look right?"

50–79% — Multiple options

Penny has identified several possible categories but isn't sure which is correct. You'll be presented with a shortlist of two to four options to choose from.

This often happens with:

• Payments to generic-sounding merchants
• Amounts that could be personal or business spending
• First-time payments to a new supplier

Below 50% — Manual review required

Penny doesn't have enough information to make a reliable suggestion. The transaction is flagged for manual review. If you have an accountant connected to your account, it can also be routed to them for classification.

Where to see confidence scores

Confidence scores are visible in several places:

• Transaction list — Each transaction shows a small confidence indicator (a coloured dot: green for auto, amber for suggested, red for review needed).
• Dashboard summary — Your dashboard shows a count of transactions awaiting review, grouped by confidence tier.
• WhatsApp — When Penny messages you about transactions, she includes the confidence level in natural language (e.g., "I'm pretty sure this is..." vs "I'm not sure about this one...").

Improving your scores over time

Confidence scores improve as Penny learns your patterns. Here's how to help:

• Review promptly. The sooner you confirm or correct transactions, the sooner Penny learns.
• Be consistent. If you categorise similar transactions the same way each time, Penny picks up on the pattern faster.
• Correct rather than ignore. If Penny gets something wrong, correct it rather than leaving it. Corrections are the most valuable learning signal.
• Add business details. Your industry selection and business type help Penny make better initial guesses, so make sure these are accurate in your settings.

Accuracy over time

New accounts typically see around 60–70% of transactions auto-categorised in the first month. By the third month, this usually rises to 85–90% as Penny learns your specific spending patterns. Long-term users often see auto-categorisation rates above 95%.