AI expense review troubleshooting

AI Expense Reviews collaborate with your company’s spending guidelines to automatically flag non-compliant transactions. See more on AI expense review and Set spending guidelines.

This guide focuses on resolving two specific scenarios:

• Missed Violations: Expenses marked as OK that should have been flagged
• Incorrect Flags: Compliant expenses flagged as NOT OK by the AI

Use the following steps to refine your policy settings and ensure accurate automated reviews.

Quick checks:

1. Confirm you’re looking at the right category:
   • Spending Guidelines are evaluated against the expense category the transaction/expense is in.
   • If the category is wrong (or too broad), the outcome will often be wrong too.
2. Check the guideline is published and up to date:
   • Ensure the category has a guideline written and published.
   • If you just edited the guideline, save/publish again (this helps ensure the latest version is used).
3. Check the available details:
   • The AI can only evaluate what it can “see” in the expense details. Missing information (attendees, context, location, etc.) can cause incorrect results.

Missed violations

Below you can see the most common cause and fix for missed violations:

• Guideline category does not prohibit the behaviour: Describes what’s allowed, but doesn’t define what isn't allowed
• Policy is descriptive and not prescriptive: No defined amount, limits or conditions
• Exceptions are too broad: Eg, "when necessary" without defining the necessity

All of the above missed violations can be be avoided or fixed with more precise spend guidelines. To ensure maximum precision and clarity, guidelines must explicitly define permitted and prohibited behaviours, specific spending limits, and clear usage conditions. See how to write effective spend guidelines.

Misleading or insufficient expense details can also sometimes cause a missed violation. To resolve this, update the guideline to request the details needed for the relevant category. Consider adding (or tightening) submission requirements for:

• Notes
• Attendees
• Receipts
• Tags / cost centres (if relevant)

Incorrect flags

Below you can see the most common cause and fix for incorrect flags:

• Guideline is unclear or open to interpretation: Lacks defining limits or conditions
• Guideline requires context not available in the expense: AI needs missing context that should be defined in the guideline
• AI assumes a rule: Explanation mentions a rule, such as an amount, that does not exist in your guideline

All of the above incorrect flags can be be avoided or fixed with more precise spend guidelines or additional expense requirements. To ensure maximum precision and clarity, guidelines must be explicit. See how to write effective spend guidelines.

Edge cases:

• Amount / currency interpretation issues: AI claims something is above a limit, references the wrong currency or compares mismatched currencies
  1. Write limits using one clear currency and be consistent (e.g., always “€”)
  2. If your policy differs by country, specify it explicitly (e.g., “UK: £X, EU: €Y”)
  3. If your company reimburses in one base currency but spends in multiple, consider adding a line like: “Limits apply in local currency at the time of purchase.”
• Category mismatch:The transaction is OK under another category’s policy, but is flagged here
  1. Move the expense to the correct category. 
  2. If this happens frequently, consider:
     • Renaming categories to be more distinct
     • Splitting broad categories (“Meals”) into more specific ones (“Client meals”, “Internal meals”, “Team events”)
     • Adding clearer “What this category includes” lines to the guideline