agent fico scores 127 transaction patterns so your AI doesn't look like fraud

stripe flags 18% of first-time agent purchases as fraud. not because the agent is malicious — because it buys like a bot.

i built agent fico to fix this. it's a credit score for AI agents transacting on behalf of humans.

what it tracks

• payment velocity (purchases per hour)
• refund rate (disputes / total transactions)
• merchant diversity (unique sellers)
• session consistency (same user, same agent)
• receipt audit trail (merkleaudit chain)

127 signals total. the score updates in real time as the agent transacts.

why this matters for ecommerce

if you're building a shopping agent, you need the agent to:

1. not trigger fraud detection on checkout
2. qualify for net-30 terms with merchants
3. dispute charges with provable history

agent fico gives merchants a number they can underwrite against. score above 680 — approve the transaction. below 580 — require human confirmation.

how it works with mnemopay

the sdk writes every payment to a tamper-evident chain. merchants query the agent's score via API before finalizing checkout. the agent carries its reputation across sessions, across merchants, across platforms.

this isn't theoretical — 3 ecommerce platforms are piloting it in production. agents with scores above 700 see 12% fewer declines than agents without scores.

credit scores for humans took decades. agents need them now.