Where every dollar your team spends shows up before the credit card statement.

All four, bidirectionally. QuickBooks Online + Xero are 1-click OAuth and ship from day one — sync is real-time, write-back of class/location/project supported. NetSuite is on the Growth plan and uses a SuiteApp we maintain; mapping setup is a one-time 30-minute call. Sage Intacct is also Growth, supports custom dimensions, and is the only integration that requires a Smart Rule for posting to the right entity if you're on a multi-entity GL. Desktop QuickBooks is supported via IIF export — not real-time, but a one-click batch.

Each virtual card is hard-locked to a single merchant identifier (MID) at the network level — not just at our policy engine. If someone tries to use the "Anthropic" card at any other merchant, the issuer (Lattice Bank) declines before we even see the auth. You also get amount caps (per-transaction and per-month), an auto-renewal calendar that warns 14 days before a charge, and the option to flip the card to "send 1 charge then auto-freeze" — useful for one-time vendor payments. When a vendor is offboarded, freezing the card stops future charges instantly.

Any combination, with full boolean logic. A typical rule looks like: `if amount > $200 AND category == "Travel" AND vendor NOT IN ('Delta','Marriott','Lyft') THEN require approval from L1+`. You can add employee role, department, week-of-month, even time-of-day. Every rule is testable: hit "preview on last 90 days" and see which transactions would have triggered, before turning the rule on. Rules can compose — a global "no purchases over $5k without CFO approval" sits atop your per-team rules.

Our internal accuracy on totals + tax + tip is 99.4% on US English receipts (we test against a labeled corpus of 14,408 receipts monthly). When it's wrong, the bookkeeper sees a soft-flag in the dashboard and can correct inline; the correction is fed back into our extractor as ground truth, so accuracy improves for everyone. For receipts in 12 supported languages (Spanish, French, German, Portuguese, Japanese, etc.) accuracy is 97.8%. Handwritten amounts and crumpled thermal receipts are still the hardest cases — we surface low-confidence extractions explicitly rather than guess.

Employees submit a reimbursement from the mobile app: photo of receipt, category, optional note. It runs through the same policy engine as card spend, then queues for the approval chain you set. Approved reimbursements pay out via ACH (1-2 business days) using the bank account the employee added during onboarding. International reimbursements go via Wise — 8 currencies supported, FX shown upfront. The reimbursement amount posts to your GL with the right account and class just like a card transaction.