Messy Travel Data Was Good Enough to Fix Policy

In fall 2026, Figma finance chief Praveer Melwani sat down for a review of travel and expense spending as part of planning for 2027. The question was where to change policy, and where existing rules were not being upheld.

His corporate finance team had combined records held in separate systems. They used Navan for travel bookings, Zip for purchase approvals, and Ramp for corporate cards and expenses.

The hard part was getting that fragmented data into the right shape. That meant matching records across systems and interpreting ambiguous trips, such as whether someone who lived in San Francisco was traveling for an offsite or simply visiting locally.

Getting the data in the right spot, Melwani said, was 95 percent of the challenge. The software was good at making those judgment calls for the team.

He told the team he did not need something 100 percent perfect. He wanted something directionally right, enough to show where to tighten or loosen the key policy controls, what he called cranking the dial.

The work came back with visualizations and suggested policy changes. After Melwani gave feedback that morning on how he wanted it constructed, a revised version was ready in the meeting within about 30 minutes.

Melwani called it a fantastic use case, saying he did not think that kind of work could have been done by hand. Melwani said using "the right tool for the right purpose" mattered more than demanding perfection for every figure.

A16Z partner Seema Amble, who spoke with Melwani, noted the shift from the older workflow, where finance staff ran a database query, waited to see the result, and then had to act as their own analyst to build charts. With AI, she said, the disparate sources could be consumed and matched much faster.

For travel policy, a roughly right answer delivered quickly was enough to guide decisions. Exact reconciliation could wait for work where errors are not allowed.