Rivian’s AI Agents Cut 15 Days From Every Close

Electric vehicle manufacturer Rivian automated one of the more tedious jobs in corporate finance: tracking how much to set aside for custom manufacturing tools it has ordered but not yet received a bill for.

Rivian built an artificial intelligence system on Amazon Bedrock and eliminated more than 15 days of manual work per close cycle, according to an Aug. 13 Amazon Web Services (AWS) blog post.

“The partnership with AWS and the implementation of AI-powered automation for our purchase order accruals represents a fundamental shift in how we approach finance operations,” Yogesh Yadav, senior director of global finance systems at Rivian, said in the post. “We’ve not only eliminated weeks of manual work each month, but we’ve created a scalable foundation that will support our growth.”

The accounting problem itself is hard. Custom tooling for building cars, like stamping dies and injection molds, can take one to two years to build, but Rivian often doesn’t get billed until 18 months or more after placing the order, the post said. Accounting rules still require the company to recognize that cost gradually over time, not all at once when the invoice finally shows up.

Why Rivian Stopped Hard-Coding Its Accounting Rules

Rivian and AWS skipped the usual fix, writing more rigid automation rules, because the actual rules kept changing depending on the size and type of each purchase. Instead, they stored Rivian’s real accounting procedures as plain-written instructions the AI reads directly, rather than hard-coding those rules into software that would need a developer to rewrite every time a policy changed, according to the post.

That setup is what lets Rivian change the process without touching any code. A finance manager who wants to update a rule just edits the written instructions; the AI picks up the change automatically the next time it runs, the post said. The system watches for new purchase orders that need attention, checks its own instructions for what to do, emails the right person if information is missing, calculates the amount owed and drafts an accounting entry, which a human still must approve before it becomes official.

A Human Still Signs Off on Every Entry

That human approval step isn’t a workaround. It’s intentional. Finance managers keep the final say before anything posts to the books, satisfying both Rivian’s internal controls and its outside auditors, who found the new process easier to trace than the spreadsheets it replaced, according to the post. When a manager corrects something that the AI got wrong, that correction gets fed back into the written instructions, so the same mistake shouldn’t happen again.

Rivian built the proof of concept in five weeks before putting the system into real use, and the company is already planning to apply the same approach to other finance tasks, the post said. That instinct matches where large companies are already putting AI to work.

In structured, rules-based finance areas like working capital monitoring and cash flow tracking, exactly the kind of work Rivian automated, 45% of chief financial officers reported using AI today, the PYMNTS Intelligence report “CFOs Are Letting AI Into the Finance Function, Carefully” found in December.

Another December PYMNTS Intelligence report, “Smarter Spend: AI-Powered AP for Data-Based Decision-Making,” found that 66% of accounts payable teams separately reported an increase in manual workload over the previous year, the exact pressure Rivian aims to relieve.

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.