Flat business complexity means smaller finance teams within a year
Reconciliation, M&A due-diligence prototypes, and investment memo generation have been compressed to minutes by AI. Figma CFO Praveer Melwani argues that once business complexity stops growing, finance team headcount needs will shrink as a result.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Automation once meant hand-writing SQL; AI scales what took specialists
Praveer recalls that when he joined Figma, automation meant hand-writing SQL himself—merging product usage data with Stripe financial data, writing 400-500 lines of queries at a time, waiting hours per run, often needing multiple attempts. Once stable, he'd need to bring in a data scientist to integrate the logic into an ETL pipeline for continuous delivery to Salesforce, Marketo, and other systems. His point: the productivity multiplier once built by hand is now within reach for anyone using AI tools.
— Praveer MelwaniYou hire for judgment—AI can't replace your thinking
Praveer is hiring a VP Finance and deliberately gives candidates minimal information (the company is public with abundant data). What he's really testing isn't whether they build the model, but whether they truly understand the model structure, know which assumptions most affect results, and can explain why the model is built that way. Anyone can prompt a large language model to spit out a model. That's not the same as understanding the business. AI is a tool, but thinking cannot be outsourced.
— Praveer MelwaniM&A due diligence now includes: have AI rebuild the target's product
Figma's corp dev team has added a standard due-diligence checkpoint: use vibe coding to reverse-engineer the acquisition target's product and immediately assess how deep the technical moat runs. Further, they've built an AI personality model for each executive, then use these personas to simulate what questions Praveer, CRO Sean, and others would raise when drafting the investment memo. They resolve weak points before team meetings, cutting the back-and-forth discussion time.
— Praveer MelwaniNinety percent of your new agents might never get used—build them anyway
Ivan mentions a growth-stage finance team that built 40-50 agents in a month or two, but roughly 90% ended up with no users. Praveer sees this "waste" as worth celebrating—the point isn't agent retention but shifting the team's default from "let's do this manually" to "could we automate this?" The real risk comes later: when similar tools get reinvented across audit, GRC, AP, and other teams. That's when someone needs to step in, consolidate the scattered tools, and make them reusable.
— Ivan MakarovFinance engineers stay distributed; the CIO owns consolidation and reuse
Figma didn't centralize automation into one team. Instead, business experts build their own prototypes first—they know best what the workflow should look like. But Praveer won't let the same person both build systems and own upstream dependencies, maintenance, and handoff risk. That belongs with a structured CIO organization: identifying which bottom-up tools deserve to be productized and reused across everyone, rather than letting each team maintain its own version of the same thing. He describes new CIO Casper Sorensen's role as building centralized PM function.
— Praveer MelwaniAccess to AI tools is now part of your recruiting pitch
Praveer noticed team members voluntarily saying they enjoy their work partly because the company gives them freedom and policy to explore AI tools. He's started pitching this as a selling point in hiring conversations, especially to non-technical candidates—it's their first chance to touch an entirely new world. More compelling than just saying you can use these models is letting candidates actually feel it during the interview: the company isn't just granting permission, it's actually using and building with them.
— Praveer MelwaniFlat business complexity means smaller finance teams within a year
Asked whether efficiency gains will plateau, Praveer is direct: if Figma's business complexity stays the same, it probably won't need the same number of people a year from now. But he's precise about the limits—in reality Figma keeps adding product lines and acquiring teams; complexity is always rising, so the organization is still racing that growth rather than simply cutting. He's clear, though, that efficiency gains will persistently reduce the headcount required per unit of complexity.
— Praveer MelwaniExcel doesn't get replaced; it's how you build trust in finance
When asked if Excel is due for retirement, Praveer flatly disagrees: it's how he understands the business and builds confidence. He compares it to Figma's own product—you can generate a design draft from natural language, but to tweak colors you're better off manipulating the color wheel directly. Models aren't yet good enough to output a trustworthy forecast model straight away. That hands-on adjustment, organizing your thinking through spatial layout, won't be easy to replace in the near term.
— Praveer MelwaniIn their own words · checked verbatim
If our business complexity was staying the same, I think we probably wouldn't need the same number of people a year from now.
Praveer Melwani0:00
We're in the era right now where a lot of slop is being called out.
Praveer Melwani0:00
I do think that's the magic of shifting this role from what can be viewed as like a traffic cop sort of financing type function to one that's a business enabler and a true change agent within the organization.
Praveer Melwani3:05
And I think you don't want to outsource your thinking today. Like I think you want to use AI as a tool to help you and empower you. But it's really, really critical to still be that thinker.
Praveer Melwani9:12
can you vibe code a product is one of the one of the checks that we use in diligence now, which is actually really fun to see.
Praveer Melwani12:20
I was talking to one of our growth stage CFOs who was saying that in, I think, a month or two, there have been, you know, 40 or 50 agents created by their team, which obviously are not all being used.
Ivan Makarov19:32
No, it's like my love language.
Praveer Melwani40:25
Figures
| Early SQL automation: lines of code written | 400-500 | 4:09 |
| T&E data reconciliation: proportion of issues from data discovery | 95% | 18:32 |
| Growth-stage company agents built in 1-2 months | 40-50 | 19:32 |
| Agents ultimately unused: proportion | ~90% | 19:32 |
Glossary
- vibe coding
- Development approach using natural language prompts to have AI generate runnable product prototypes.
- ETL (Extract, Transform, Load)
- Automated pipeline that extracts data from one system, transforms and cleans it, and loads it into another.
How to listen
CFOs and finance leaders building AI-native finance and operations teams, plus founders and investors wondering how AI will reshape organizational headcount.
The opening (0:00-2:05) is a preview montage; everything covered later in full, so you can skip it.