AI-assisted matching

How the AI helps match transactions to claims, and what you should know about privacy and improving results.

How it works

Matching uses a two-step process:

  1. Rule-based matching — Compares transactions to claims using description keywords, amount, date proximity, reimbursing entity names, and built-in patterns (e.g. medical, pharmacy, insurance). When the rule-based engine has high confidence, it suggests a match without using AI.

  2. AI-assisted matching — For uncertain cases (medium confidence), optional AI can enhance suggestions. If you enable AI in Settings → AI Features, anonymised data is sent to OpenAI to analyse whether a transaction is related to a claim. The AI result is combined with the rule-based score (about 60% rule-based, 40% AI) to produce the final suggestion.

You always review and confirm matches on the Review matches page — nothing is linked automatically. The Review page also shows a “Likely claimable” section; when AI is enabled, it supplements rule-based identification to surface more transactions that may need a claim.

Data privacy

Your data is protected before it leaves ClaimStation. Anonymisation runs automatically: account numbers, card details, policy numbers, member IDs, references, names, emails, phone numbers, and addresses are removed or replaced with generic placeholders (e.g. [ACCOUNT], [PHONE]). Reimbursing entity names are mapped to generic types. Only anonymised descriptions, amounts, dates, and generic category types are sent to the AI.

  • Optional and reversible — AI matching is off by default. You must explicitly enable it in Settings → AI Features and consent before any data is sent. You can disable it at any time.
  • Where your data goes — When enabled, anonymised data is sent to OpenAI (a US-based provider) for analysis. ClaimStation does not share personally identifiable information with AI services.

Does the AI learn from my matches?

The AI model itself does not learn from your matches; each analysis is independent. If you enable Matching Intelligence in Settings → AI Features, ClaimStation learns from your confirmations and rejections to improve rule-based suggestions. This learning is optional, stored in your account, and you can clear it at any time.

Matching Intelligence

Matching Intelligence learns from your confirm and reject actions on the Review matches page. When enabled, it stores structured feedback and extracts reusable patterns to improve future suggestions. Learned patterns are used as a sixth scoring factor. When both Matching Intelligence and AI matching are enabled, your confirmed matches can also be included as anonymised few-shot examples in the AI prompt.

  • Opt-in and separate from AI consent — Learning is stored in your ClaimStation account and is not sent to external services on its own. You can disable or clear it at any time in Settings → AI Features.
  • Household — If you belong to a household, learning data is shared with household members so patterns discovered by one member can improve suggestions for everyone.
  • Clear learning data — Use “Clear learning data” in Settings → AI Features to permanently delete all stored feedback and learned patterns.

How to improve matching

  1. Descriptive claim text — Use wording similar to what appears in your bank statement.
  2. Add a reimbursing entity — If your insurer or employer is set on the claim, the matcher can compare entity names to transaction descriptions.
  3. Set expected reimbursement — When known, expected amounts help the matcher compare transaction amounts more accurately.
  4. Enable AI for ambiguous cases — AI is most useful when rule-based matching is uncertain.
  5. Review and correct — Confirming or rejecting suggestions keeps your data accurate.