AI Automations
AI Expense Suggestions
Topkey's AI reviews your card and bank transactions and suggests an accounting category (GL code) for each one, based on the merchant name, transaction amount, and your historical coding patterns. The more transactions you code in Topkey, the more accurate the suggestions become over time.
Where Suggestions Appear
AI expense suggestions are available on the Accounting Category Suggestions page, found under Transactions in the left navigation menu. This page lists every transaction that has received a suggested category, along with the suggested GL code, any existing category already applied, and the AI agent version that generated the suggestion.
Each row in the list shows:
- Transaction — the merchant name and transaction amount
- Suggested Category — the GL code Topkey recommends
- Existing Category — the category currently applied to the transaction, if any
- AI Agent Version — the version of the AI model that generated the suggestion
Confidence Scores
Each suggestion includes a confidence score that indicates how certain the AI is about its recommendation. Scores are expressed as a percentage and broken down into contributing factors, including how well the merchant matches the category, whether the amount is typical for that expense type, and how often that category appears in your chart of accounts.
Use the confidence level as a guide for how much scrutiny to apply:
- High confidence (90–100) — clear, unambiguous match; likely safe to accept without detailed review
- Good confidence (70–89) — strong indicators support the category; worth a quick check
- Moderate confidence (50–69) — reasonable match but some uncertainty; review before accepting
- Low confidence (below 50) — the AI is uncertain; manually verify before accepting
Focus your review time on lower-confidence suggestions. High-confidence suggestions from well-known merchants are usually accurate.
Accepting a Suggestion
When you agree with a suggested category, click the green checkmark next to the transaction. Topkey applies the suggested GL code to the transaction immediately. Accepted suggestions also feed back into the AI model, helping it make better recommendations for similar transactions in the future.
Rejecting a Suggestion
If a suggestion is incorrect, click the X to dismiss it. The transaction will remain uncategorized so you can apply the correct GL code manually. Rejected suggestions are also used to improve the model — the AI learns from both acceptances and rejections.
Rejecting a suggestion does not delete the transaction. You can still categorize it at any time from the Transactions list.
Bulk Accepting Suggestions
If several suggestions look correct, you can accept them all at once instead of reviewing each one individually. Select the transactions you want to approve using the checkboxes in the list, then use the bulk actions bar that appears to accept all selected suggestions in a single action.
Bulk accepting works well for high-confidence suggestions from merchants you recognize and regularly code to the same category.
Filtering the Suggestions List
You can narrow down the suggestions list by typing a plain-English description of what you're looking for into the search field — for example, "hardware store transactions over $50" or "office supply purchases this month." Topkey's AI converts your description into the appropriate filters automatically, so you don't need to set each filter manually.
You can also filter by AI agent version using the version selector at the top of the page, then click Filter to apply it. This is useful if you want to review suggestions generated by a specific model version.
How Topkey Learns Over Time
Every time you accept or reject a suggestion, that decision is recorded and used to refine the underlying AI model for your account. Factors like your chart of accounts, the merchants you frequently code to specific categories, and your overall categorization style all inform future suggestions. Accuracy typically improves noticeably after you have reviewed a few hundred transactions.
The AI model considers merchant name and location, transaction amount, attached receipts, property context, and patterns from similar past transactions when generating a suggestion.
Frequently Asked Questions
When do suggestions appear for a transaction? Suggestions are generated automatically when a new card or bank transaction is imported into Topkey, provided your account has AI expense suggestions enabled. There may be a short delay between when a transaction appears and when its suggestion is ready, as the AI model needs time to analyze the transaction details.
Why doesn't a transaction have a suggestion? A transaction may not receive a suggestion if the merchant information is limited, if the transaction type is one the model hasn't encountered before, or if there isn't enough historical coding data in your account yet to make a confident recommendation. As you code more transactions, coverage will improve.
How can I improve the quality of suggestions? The most effective way to improve suggestion quality is to consistently accept correct suggestions and reject incorrect ones rather than manually overriding without responding to the suggestion. Keeping your chart of accounts up to date and attaching receipts to transactions also gives the AI more context to work with, which leads to more accurate recommendations.
Questions? Contact Support