Sapiom wired its AI-agent spend layer into Truss for a 15-day pilot. The numbers below are live: 0 invoices processed, 0 fraud decisions made, — approval rate, and 0 finance hours saved — pulled directly from the same database the Sapiom team queried during the pilot.
If you're building agents that spend money on behalf of your company, drop your email and we'll line up API credentials for the same eight-signal scoring + Sapiom disbursement pipeline.
Couldn't sign you up — try again.
We'll be in touch.
Sapiom's agents buy software, subscribe to services, and authorize spend around the clock.
Before Truss, those invoices landed in a shared inbox where a human reviewed them one at a time.
During the pilot, every invoice was POSTed to /api/invoices, synchronously scored
against Truss's eight fraud signals, and routed to either auto-payment or a reviewer
queue — no UI, no tab-switching, no manual typing.
Most AP vendors bolt fraud on after the fact. Truss scores an invoice the second it comes in:
vendor age, amount anomaly against historical spend, routing-number validation,
24-hour velocity, round-amount heuristics, due-date urgency, agent history, and prior
routing changes. The Sapiom team picked Truss because the threat model matched their
agent-driven spend, not a human-driven one.
Browse the curated agent-native fraud patterns Truss is built to catch →
See tier comparisons →
The numbers above are not written copy. They are SQL queries against the invoices and payments tables over the trailing 15 days. The "Time saved" figure is a derived estimate: every payment auto-scheduled by Truss is assumed to have replaced ~30 minutes of manual finance work, consistent with the 87% finance work auto-processed baseline we publish on the main page. The "Fraud decisions made" count is the count of invoices whose risk score fired above zero — the system decided something, flagged or escalated, on every one of those.
Sixteen days in, the AP inbox stopped being a thing we opened. Every agent-issued invoice gets scored, queued, or blocked before it reaches a human — and we paid zero fraudulent bills. That’s the bar I’d hold any AI-native AP system to.
Same API, same eight signals, same Sapiom-style rollout for any AI-native company. Start at the landing page, or read the engineering guide Sapiom used.
Sapiom ran autonomous AP on Truss for 15 days — 0 fraud paid. Drop your email and we'll line up the same eight-signal pilot for your agents.
Couldn't sign you up — try again.
We'll be in touch.