A software firm says its AI checked 68,000 utility invoices against contract terms and flagged over $1.1 million in possible overbilling before payment.
A technology firm called TechWish says its AI tool found more than $1.1 million in possible overbilling and missed discounts in vendor invoices at one of the largest electric and natural gas delivery companies in the US. The company announced the project on October 26, 2026. The utility is not named.
The tool, called the Invoice Review Console, reads vendor contracts and checks each invoice against them. It covers rates, limits, surcharges and deadlines. TechWish says it looked at more than 68,000 invoices worth over $190 million from 2025. It describes that as only a fraction of the invoices the utility handled that year.
The $1.1 million covers potential overbilling and contract discounts that were never applied. TechWish gave two examples:
Those two examples add up to $571,000. The release does not itemize the rest of the $1.1 million.
Contract terms are usually spread across master agreements, amendments, statements of work and rate cards. Each vendor words them differently. Because the terms sit in documents and not in a billing system, people have traditionally reviewed and approved invoices by hand.
TechWish says its AI reads each contract clause by clause and turns the terms into structured data. It also records the clause and page each term came from. When checked against human reviewers, it pulled out contract terms with 97% accuracy, the company says. The utility can now spot overbilling before payments go out.
TechWish says the platform is not limited to utilities. It names energy, telecommunications, construction, healthcare and field services as fits. Companies can start with a small set of checks and vendors, then add more. The system runs on Amazon Web Services, with limits on AI spending, US data storage and a full audit trail.
The numbers come from TechWish's own announcement. The release does not say how much of the flagged money was actually recovered. It also does not say whether the vendors agreed the charges were wrong, or what the tool costs. The word used is "potential," so each flagged item still needs a person to confirm it.
If your business pays many vendors, checking invoices against contract terms before paying can catch errors that are otherwise easy to miss.
Visa tested 22 AI models on realistic shopping tasks. They passed just over half, and often bought the wrong thing anyway.
Read more: Visa tested 22 AI shoppers, and most still make costly mistakesDARPA wants AI to hunt for design flaws in stock, bond, energy and cloud markets, so enemies cannot use them to cause economic damage that reaches households.
Read more: Pentagon wants AI to find weak spots in stock and energy marketsTuring Award winner Richard Sutton says more computing power alone will not make AI truly smart. Big AI companies disagree, which matters for anyone betting on AI.
Read more: AI pioneer says piling on computing power won't make machines smart
