The story (for the sales conversation)
"Most invoices match their order and receipt perfectly — but a human still opens every one. This agent posts the clean ones, investigates the exceptions, and only asks you when something genuinely doesn't add up."
😣 Today, without the agent
Dana in AP has 400 invoices in the queue. She opens each one, pulls up the PO, checks the goods receipt, eyeballs the price. Most are fine — but the checking still takes all morning. One invoice is $2.55 over the PO price; she has to dig out the contract to see if that's within tolerance. By the time the exceptions are sorted, two invoices have slipped past their early-pay discount.
😌 The same queue, with the agent
The agent runs the three-way match on all 400. The clean ones within tolerance post for payment automatically. For the $2.55 variance it reads the ivalua contract, sees the tolerance is ±2% and this is 5.3%, and routes it to Dana with the contract clause and a recommendation attached. The partial-delivery one it holds and explains. Dana opens her queue to a handful of real exceptions, each pre-investigated.
"It never posts an invoice or releases a payment on its own beyond policy tolerance. Within tolerance it acts; outside it, it investigates and recommends, and a human approves. The audit trail is automatic."
The villain: the queue
Every invoice waits for manual matching, even the clean majority.
The hero: match + investigate
The agent posts clean matches and pre-investigates exceptions with contract context.
The reason to trust it
Posting and payment are reversible Action Tickets; beyond-tolerance needs a human.