How AI Is Turning Figma’s Finance Team Into Builders
Praveer MelwaniSeema AmbleIvan Makarov
- Figma’s finance team is using AI to automate collections and help test products during diligence, extending finance into building. A collections specialist rebuilt invoice tracking, cash collection and NetSuite reconciliation. Corporate development recreates products as a diligence check on technical difficulty and moat depth, while continuing to partner with the product team.
- Finance headcount could fall if business complexity stopped growing, but that is a conditional forecast. CFO Praveer Melwani says that under unchanged complexity, “I think we probably wouldn’t need the same number of people a year from now.” Figma’s new products and acquisitions keep increasing the workload, even as efficiency improves period over period.
- The build-versus-buy boundary favors purchased systems of record and internally built workflow automation. Melwani buys systems “we just can’t mess up,” while specialists prototype around repetitive tasks. The organizational challenge is turning successful local tools into shared infrastructure that can “live on beyond the creator.”
- Data preparation remains the bottleneck, while acceptable precision depends on the decision. Reviewing travel and expenses for 2027 planning required matching Navan, Zip and Ramp records; Melwani says “ninety-five percent of the challenge” was getting the data together. Directionally correct analysis was sufficient for policy decisions, and revised visualizations followed his feedback within 30 minutes.
- AI makes judgment, accountability and consistent definitions more valuable. Melwani’s hiring process tests whether candidates understand their model’s sensitive assumptions, while Ivan Makarov raises concern that companies could forget what world-class work looks like. “You don’t wanna outsource your thinking”: authors remain responsible for understanding and simplifying what others must read.
- Finance software is moving toward completed results with an auditable explanation. Seema Amble sees products shifting from workflow tracking to delivered reconciliations plus an audit trail, and dates a substantial adoption change to roughly nine months before the conversation. Melwani hopes for more complete automation, starting with deterministic work, but audit-ready outputs still face an “extremely high” bar.
- Access to tools and a visible building culture are becoming recruiting and retention assets. Team members tell Melwani they are having “a ton of fun,” while automated reporting frees capacity for experiments, spending decisions and investigation. His recruiting argument combines technology access with colleagues who actively build and teach one another through example.
1. Finance’s operating advantage starts with trusted data
- Before AI, Melwani connected Figma’s product usage with Stripe data through queries running 400 or 500 lines of SQL. Data scientists later put his “golden tables” into recurring ETL jobs feeding Salesforce and Marketo. Consistent definitions stopped arguments over what was true and let finance become “a business enabler” rather than a “traffic cop.”
- He once reviewed about 30 new paid-plan signups line by line, distinguishing personal from business addresses to inform Figma’s product for organizations. Today, he checks daily metrics and hourly dashboards, expecting colleagues to act rather than wait until next week. Ivan’s comparison: pandemic usage spikes once took weeks to validate; today, he thinks similar questions could be resolved in an hour.
- On metric definitions, Melwani is categorical: “my team owns the outputs there.” Public-company controls favor consistency across board materials, earnings disclosures and historical reporting. Different ways of understanding the business are welcome, but reported measures should have longevity: “the historicals never change.”
- Ivan’s advice to a first finance hire: establish data integrity and tracking before rushing to close books or send board updates, then build reporting rigor, metrics and the story. He has seen hires discover within several weeks that revenue was calculated incorrectly. Seema flags more “creative” early-stage ARR; Ivan says numerical storytelling once associated with IPO preparation is now expected in Series A and Series B fundraising.
2. AI strengthens the hiring case for independent judgment
- Intellectual curiosity, data fluency, structured questions and clear communication still define Melwani’s ideal hire. His VP-finance case study supplies little information beyond available public disclosures and expects candidates to use AI. The examination is whether they understand their model, its most sensitive assumptions and the business underneath it: “you don’t wanna outsource your thinking.”
- Ivan’s pushback — worth keeping: instant document generation can produce an unsearchable pile of work, while the shift in hiring toward tool fluency raises his concern that people could forget “what world-class looked like.” Melwani concedes he sometimes encounters half-baked output. Figma has policies for branding AI-written work, but authors still owe readers review, simplification and understanding: consuming anything takes “time, effort, and energy.”
- Melwani’s own correction concerns influence: he initially thought he could “be the smartest person in the room and figure it all out,” then learned finance is a team sport. He had underplayed communication, relationships and persuasion. The hiring signal he now prizes is the ability to build trust and “really solid followership,” because analysis changes a business through other people.
3. Collections and acquisition diligence already have finance builders
- In credit and collections, Kajal rebuilt invoice tracking, cash collection and NetSuite reconciliation around Figma’s existing systems. That frees time to reduce days receivable and tackle further bottlenecks, including redesigned credit checks and connected datasets. Melwani says some collections automation uses ChatGPT agents and skills; the resulting tools can also be used by colleagues.
- Corporate development recreates products during diligence: “can you vibe code a product” helps test technical difficulty and moat depth. Melwani thinks the corporate development team is using Codex for this work. Ivan underscores that these are finance-background corp-dev staff. Melwani adds the limit: replication provides “a solid intuition,” with nuances still requiring judgment. Product remains a partner, but finance arrives with something it has already built.
- The team also creates skills reflecting recurring leadership perspectives: “what Praveer would say,” alongside Dylan and CRO Sean. Seema calls it an LLM investment committee; Melwani agrees. Running investment memos through those personalities highlights weak sections, missing diligence and predictable questions before leadership reads them, reducing the time spent cycling through feedback together.
- Figma Make offers another route: natural-language prompts can produce an application or website with hosting and a backend. Melwani is seeing internal use cases emerge there, which also turns finance into a source of product feedback. Building becomes a way to automate work and learn.
4. Useful analysis does not always require perfect reconciliation
- During 2027 planning, a travel-and-expense review combined Navan, Zip and Ramp data. Matching records required judgment: was someone traveling for an offsite, and how should someone living in San Francisco who was just visiting be classified? Melwani’s diagnosis: “ninety-five percent of the challenge there is just, like, getting your data in the right spot.” For this decision, directional accuracy was sufficient.
- The team produced suggested policy changes, then had revised visualizations ready 30 minutes after his morning feedback. Melwani doubts they could have done the work by hand. His counterweight to the slop critique is purpose-specific precision: understand “what is good enough and what needs to be, like, really, really picture perfect.”
- Seema describes another growth-stage finance team creating 40 or 50 agents in a month or two. She guesses “maybe ninety percent” are no longer used, but values the habit the exercise developed: when work is manual and tedious, “let me build.” Melwani agrees, particularly when teams share what others might reuse.
5. Local prototypes need shared ownership to survive
- A speaker offers the failure case: a customer-success leader found every rep had vibe-coded a different dashboard, with different numbers. Melwani sees a systems opportunity in repeated audit-evidence requests, GRC work across ISO and SOC 2, and accounts-payable approvals. These share a pattern of chasing information, triaging and reconciling; a systems team can turn scattered tools into a shared platform.
- Specialists should invent and iterate because they know the process and “what good looks like.” Maintaining upstream dependencies, arranging durable hosting and ensuring tools survive their creators’ departure need different ownership. Newly arrived CIO Kasper Sørensen, from Coinbase, is exploring centralized product-management-style functions that identify useful local tools and build foundations others can extend.
- This remains unfinished: “we’re in the messy middle right now.” Some teams have centralized better than others; Melwani has prioritized letting people build and is catching up with infrastructure. Seema finds similarly unsettled finance-engineer roles elsewhere, filled variously by engineers inside finance or product engineers borrowed from another team.
- Domain context matters. Figma’s finance-supporting data scientists include people with accounting backgrounds who understand refunds, uncollectibles and the requirement for correctness. Builders move faster when they understand dataset architecture, distinguish deterministic decisions from inference and know which tasks can tolerate each. FigNation demonstrations, hackathons and conversations with outside finance leaders help spread that understanding.
6. Buying core systems leaves room to build better working days
- Ivan presses the alternative: a hypergrowth company told him it had to buy everything off the shelf because it was moving too quickly. Melwani agrees for core systems of record and things “we just can’t mess up.” His team builds around repetitive workflow steps, finding tasks where someone is “almost, like, not thinking” and redesigning how that time gets spent.
- The payoff appears in retention conversations. Team members tell Melwani they are having “a ton of fun” and want to stay because they can explore and learn. He has begun selling technology access in recruiting, particularly to candidates outside technology who see it as an entrance into a new professional world.
- Ivan suggests treating models and tools like perks and benefits. Melwani adds that candidates should see colleagues actually building, including during interviews: learning happens through “osmosis of what others have built too.” He thinks developing one’s craft among similarly motivated peers can help Figma attract and build “the world’s best finance organization.”
7. Freed capacity is moving toward experiments and business decisions
- Seema recalls her 2020 framework: finance spent most of its time reconciling data and the smallest share on strategy. Melwani’s self-serve finance team can now focus more on experiments, trends and additional spending. Automating recurring materials shifts effort toward “the annotation, the discovery, the understanding.” He recalls a team discussion of Figma plug-ins to accelerate ideation, although growing complexity still competes for the saved time.
- Seema sees CFO software moving from workflow tracking to a result plus the audit trail explaining it. Melwani’s accounting example is flux analysis: tools can offer an initial explanation of period-over-period general-ledger changes, giving accountants threads to pursue. He does not claim the first explanation is always right; it provides a better starting point for investigation.
- The CFO Melwani wants shares the CEO’s business ownership and desire to win, using P&L knowledge to drive change and simplify the business model for others. Accounting fundamentals and planning remain prerequisites. Creativity means finding another business angle, not inventing “a fun, funky set of metrics.” He tries to generate those ideas but does not know whether he does it as much as he could.
- Seema dates a marked adoption shift to roughly nine months earlier: CFO dinner conversations moved from AI-generated tax memos and onboarding documents to core reconciliation and automated workflows. At board level, she sees deeper analysis and faster iteration. Leaders increasingly feel they must adopt the tools or be perceived as behind; finance now participates in automation hackathons.
8. Efficiency can reduce staffing needs while Excel retains its role
- Asked about the next 12–18 months, Melwani gives a conditional staffing call: “if our business complexity was, like, staying the same, I think we probably wouldn’t need the same number of people a year from now.” Figma keeps adding products and acquiring teams and businesses, so complexity still outpaces complete efficiency. Nevertheless, he sees improvement period over period.
- He hopes more processes become fully automated, beginning with deterministic work supplied with enough information to make outputs completely trustworthy. Business-critical, audit-ready results face an extremely high bar, including what auditors will accept. His expectation remains: “I do think we’ll get there over time.”
- Is Excel under threat? Melwani’s answer is categorical: “No, it’s like my love language, dude.” His analogy is Figma’s color wheel: prompting a change from red to green is useful, but moving directly through shades builds understanding differently. Spreadsheets similarly let him refine intuition by hand. Something better may emerge, but he would find Excel hard to surrender.
- Seema adds a limit: models are not “quite good enough” to create a forecast or model yet, while Excel’s spatial organization helps her think. Melwani uses AI to review a team’s 20-tab model, finding sensitive assumptions and important variables before investigating himself. His durable criterion is adaptability: “a great finance person” has longevity “if they continue to adapt.”
Full transcript
I think we're going to be able to attract and build the world’s best finance organization because so much of what people want is the ability to learn, feel like they’re being invested in, and be surrounded by people who want to become the best at their craft.
We’d love to hear a little bit about what makes someone a great hire on your team, a great person.
I want folks who feel extremely intellectually curious. You’re coming in and trying to solve problems and understand the business really well. You want to use AI as a tool to help and empower you, but it’s really critical to still be a thinker.
We’re in an era where a lot of slop is being called out. I’ve heard one CFO talk about how using AI just led to an unlimited number of docs being created every day.
Anything you’re putting out there and asking someone to consume is going to take them time, effort, and energy. Ninety-five percent of the challenge is just getting your data in the right spot.
What do we think changes in the next 12 to 18 months?
If our business complexity were staying the same, I think we probably wouldn’t need the same number of people a year from now.
Welcome to the a16z Podcast. I’m Seema Amble. I’m joined here by my partner, Ivan Makarov. We recently wrote a piece called “You Need a New CFO.” Our argument was that AI takes finance to the next level, and the CFO’s job increasingly includes building and automation, and really designing the systems and the team to do that. Today, we’re joined by Praveer Melwani, the CFO at Figma. We’d love to explore what you’ve seen over the course of your career and what you and your team are doing now. Praveer, thanks for joining us.
Yeah, thanks for having me. Excited.
Maybe to kick things off, Ivan and Praveer, we’d love to hear a little bit about how automation has changed the role in finance and what the team does over the course of your careers.
1. Finance as an Operational Leader
I’ve always felt that the role of a finance individual should be that of an operational leader. When I joined Figma, none of the automation pieces were in place at all. In order for me to be effective when I joined, a lot of my work needed to live in SQL: building the data tables and really understanding how I could orchestrate the financial information to drive the conversations I wanted to have across the company.
I continue to go back to those initial days in that seat and what I was doing as the force multiplier—the piece that actually gave me the ability to have a scaled impact across the organization. If anything, I think what I was building way back then is that much more possible, plausible, and accessible to individuals with the tools available to them today.
I hope more and more folks feel empowered to take that operational lens to the way they position themselves and work with their counterparts across the business. I think that’s the magic of shifting this role from what can be viewed as a traffic-cop sort of finance-y function to one that’s a business enabler and a true change agent within the organization.
To double-click on what you were using for automation before AI, maybe explain to people what exactly you were doing. Was it macros in Excel?
Yeah.
Was it using OCR for documents?
Yeah, honestly, it was everything under the sun. It started with living in SQL, looking at all of our product metric and utilization data, and mapping it to all the financial information we had in Stripe. I’d create 400 or 500 lines of SQL that I’d then use to create the golden tables, which I’d pass off to someone or pass to different go-to-market systems we’d be using.
I spent my day living in that query editor. I’d then go and tap—
And, of course, wait for the query to run for—
Totally, and then—
…however many hours.
Yeah, and you hope you got it right the first time.
Exactly.
Eventually, we hired data scientists, and I’d tap them and say, “Hey, I built this thing. Can you add it to our ETL process so there’s a recurring job that I can use to pass that same information to Salesforce or Marketo,” or whatever systems we were using at the time.
That was the initial unlock for me: If I could get consistency in our definitions and in the way we were talking about things, we’d no longer debate what was true and what was fact. We could get past that and really focus on how we were going to drive the business.
I was talking to someone in their early 20s who was complaining about how long a prompt takes to run. I was explaining to them, “You haven’t lived through hours and hours of waiting for a SQL query to run, only for it to come back and you’re like, ‘Shit, it didn’t really work.’” It’s an order-of-magnitude change.
Totally.
…to that.
Yep.
The timing is really interesting, especially the era we’ve lived through. When I was at Webflow and Figma, we saw a huge spike in usage during COVID that I couldn’t believe at the beginning. I don’t know if you experienced it this way too, where I thought our data was broken or something else was broken, but it took us weeks to figure it out.
Yeah.
Today, we live in a world where you could figure it out within an hour for some of those spikes in usage and revenue. It’s a very different world we live in. Did you experience that as well?
Yeah, our business took off during the early days of the pandemic. I think that was the time we started to partner with the folks at Andreessen, right at the early stages.
2. Running Finance on Daily Data
I wake up every day with the ability to rebuild my context on what’s happening. At any moment during the day, the team knows that I’m constantly in there looking at hourly dashboards or daily metrics to make sure I’m sourcing things that might look like anomalies or asking the right questions. I think they view me as a counterpart who has ideas and wants to build the best business possible, using and having access to the same data available to each other.
What’s interesting is that when you and I went to college around that time, it was more of a monthly or quarterly cadence. Then it became a weekly cadence. I remember, especially as data moved more into the cloud, it became, “Let’s have a big weekly dashboard,” whereas now what’s possible is daily. Has that changed your habits, the way you look at and think about data, or how often you think about it?
Honestly, I view data and having access to it as the key unlock to how I can run my day effectively. Throughout my time at Figma, it was the thing I’d wake up to every day. I remember there would be about 30 people signing up for a paid plan in those first few days after I joined. I’d go through them line by line: Who are they? Are they personal addresses or business addresses? Are they people we can reach out to? What can I learn from this to help inform what we eventually built, which was our organization’s product?
I think in the early days it’s extremely important. Now I understand the repercussions of movement in any one direction. If you catch something before it becomes a big deal, it can save the business hundreds or thousands of dollars. If you catch it going the other direction, you can put more juice behind it and really seek to unlock things the other way.
I view this as a daily, if not more frequent, rhythm. My goal is to empower other folks on my team and across the business to feel the same accountability to those metrics, truly act as business owners, and want to triage things rather than say, “Hey, I can wait till next week.” You do need to solve it sooner rather than later because I think the rhythms here are moving really rapidly.
Before we get into what you’ve automated, we’d love to hear a little bit about what made someone a great hire on your team, a great person in a leadership role, maybe before AI.
3. Hiring Curious Finance Thinkers
I think a lot of the same things apply today, but you’d apply them slightly differently. I want people who feel extremely intellectually curious. They’re coming in and trying to solve problems, understand the business really well, and then problem-solve in a way that drives the right outcome. I don’t think that’s necessarily changed today. I still very much value that.
You can make a lot of the tedious tasks, or the repetitive nature of things, take less and less of your time. That way, you can spend more of your time finding the places of magic within the business that you can supercharge. So: intellectual curiosity, fluency with data, the ability to ask really solid questions, and then structure your thoughts so you can make sense of the data you’re given and communicate it thoughtfully.
How do you interview for that? How has it—
Yeah.
—developed? What are some of the ways?
It’s interesting. Right now, I’m hiring for a VP of finance, and the case study part of the process is so sacred to me. I want to test how you use the tools available to you. I don’t give people a lot of information for the case study. We’re a public company, so there’s a lot of data out there. But I want to see how you piece those puzzle pieces together in a way that makes sense, and whether the model output makes sense. You need to be able to create that output, but also truly understand how it works.
Then I focus the case study conversation on whether you’ve really understood the structure you’ve created. Have you shaped it in a way that tests which parts of the model are most sensitive to change? Do you have a strong intuition about how the business is actually working, using the tool you’ve created? And, as we were just talking about, can you communicate that effectively?.
Anyone can create a model by picking their favorite LLM. You can do that really easily. But are you using it as an effective tool that actually enables you to think? You don’t want to outsource your thinking today. You want to use AI as a tool to help and empower you, but it’s still really critical to be the thinker.
A lot of what you’re describing applies to a good hire with or without AI. If anything, AI means you need more accountability and thoroughness to know the thinking is there behind the work product you receive.
Let’s dive in to understand what you’ve automated, and we can talk a little about what we’ve been seeing with public companies. Maybe walk us through a couple of workflows at Figma that you’ve automated. How did you decide what to focus on, and what did you do to automate them?
4. Finance Teams Build With AI
They’re so broad, so maybe I can pick a few that have been fun to see. We’ve moved really quickly over the last year, from trying to coach people to embrace new tools and new ways of working to now, when I feel like I’m constantly being hit in the face with really cool ideas.
We do a segment at the company all-hands called FigNation, where we show people how others have built with AI. Our team consistently shows up with examples. One is Kajal on our credit and collections team. He’s rebuilt his entire workflow, from invoice tracking and cash collections to making sure we can reconcile everything to NetSuite. He’s done it as a homegrown system tied to the systems we use.
That’s freed him up to keep driving down his days receivable by focusing on the things that really matter, and then to look at other tedious tasks he can automate. He’s rebuilt the way he wants to run our credit checks, and he can stitch together datasets in interesting ways so he doesn’t have to go into them manually. Now he can tell anyone, “Go leverage the tools I’ve created to really drive change there.”
On our corporate development team, it’s been fun to see them vibe-code a product as one of the checks we use in diligence now. It’s really interesting to see: How deep is your moat? How interesting is the technology? I think everyone should be able to—
You’ll try to recreate the product? Wow, okay.
Yeah. I think our team has gotten really good at some of this, so you can get through some of the questions: Is this a real thing or not? How difficult is it? There are obviously nuances to a whole bunch of these ideas, but it gives you a solid intuition.
And before, would you have had to go to the product team to do that part of the diligence?
Yeah. We still partner with the product team, but you can take a lot more of a “Hey, this is what I built.”
Sure.
“This is what I’ve seen,” and it builds a lot of credibility there.
I can pause on that. Those are some of the corporate development people, right? Not product people?
Yeah.
Their background is in finance, right?
Yeah, yeah.
It’s fascinating.
Totally. We’re a very document-oriented culture, so they write investment memos and M&A memos. The end readers, or the people digesting them, are folks on our leadership team or other product leaders. More often than not, there are themes and similar questions, and we all tend to have a way of looking at problems as leaders.
What’s been interesting is that they’ve created unique personalities for each of us as leaders, as a set of skills they can run their investment memos by to improve the process.
Do you have an investment committee that exists as an interaction between all of them through the LLM?
Yeah, exactly.
Okay.
That’s exactly what they’ll end up demoing.
Okay.
Or rather, they’ll sit there going back and forth and saying, “Okay, this is what Praveer would say, and this is what Dylan would say, and this is what Sean, our CRO, would say.” You get the areas of the document you want to strengthen or do more diligence on, and the common feedback you need to make sure you address. That reduces the time we all spend sitting together.
There are a lot of areas where we feel this is a tool that helps you augment, replace a workflow, and seek to understand and learn. I’m constantly surprised by how quickly we’ve been able to riff on things. It’s put me in a really optimistic spot. I think we’re going to be able to attract and build the world’s best finance organization, because people want the ability to learn and feel invested in, surrounded by people who want to become the best at their craft. The more I can do to enable that change, the better we’ll be at retaining, growing, and grooming that talent over time.
Are all those tools built directly with the LLM, or are you also looking at off-the-shelf AI tools you can partner with? Would you rather build or buy?
The ones I was describing are a mix. Some people were working in Codex, and I think that’s what the corporate development team is doing. They’re sitting in Codex and trying to work their way through something. We’ve also seen people working within workflow automation tools and building skills around agents using ChatGPT. That’s how some of the credit and collections pieces were built.
And we’ve also started to see folks embrace it and build on our tooling as well. We have a tool called Figma Make, which allows folks to prompt in natural language and create an application or a website. You can then host it, with a back end behind it. Some of those instances are starting to crop up, too.
I’m actually excited about that, because then we can ideally flip into this world where we can give product feedback as well, which I know everyone would love.
I think we’re in the cycle where a lot of slop is being called out. There’s been a lot of conversation about how we went from unlimited tokens to people just calling it slop in meetings. I’ve heard one CFO talk about how using AI led to an unlimited amount of docs being created every day, and you just can’t find anything, because everybody can write a doc now—or 10 docs in a few minutes, when it used to take a week or two.
What are some of the areas where this is showing up? And then naturally, that’s also created conversation around hiring. Do you hire younger people who are more versed in these tools, people with more experience, or more senior people who know what world-class looks like? That’s what worries me a little bit: that we’ll forget what world-class looked like.
Just like in the pre-AI era, you know what a really good model looks like, especially with some experience and having created models. Where do you think this is going? Have you had to call slop on some of the stuff that’s being created with AI?
Sometimes.
Yeah.
Yeah. I think we’ve also done a good job internally building policies around branding when you’ve written something with AI. One of the things that’s really important, as I was sharing earlier, is that you should not be outsourcing your thinking. Anything you’re putting out there that you’re asking someone to consume is going to take them time, effort, and energy.
You want to make sure you hold yourself to a similar bar when you review, read, and summarize something, and make it simple. You can start in a different place; I’m totally cool with that. But I do think it’s on all of us to be really thoughtful about how we use each other’s time.
I’ve been in instances where things look a little half-baked, and you really poke around and ask, “Do you actually understand what’s going on over here?” But I’ve also been in instances where I’m like, “Oh my gosh, your speed of iteration here is pretty damn incredible by virtue of how you set it up.”
A couple of weeks ago, we were going through 2027 planning, and I was sitting in a T&E review. We were looking at our T&E portfolio and asking, “Where do we have the opportunity to make changes to policy? Or where are we not upholding policy where we should?”
Our corporate finance team did a really good job stitching together information that lived in different data sources. We use Navan, Zip, and Ramp, and there’s a lot of fuzzy matching you have to do to get that information working. Was someone actually traveling for an offsite? What if they lived in San Francisco and were just visiting? How do you want to tag and treat that data?
Ninety-five percent of the challenge is just getting your data in the right spot, and these models are good at making these assumptions for you. I didn’t need something to be 100 percent perfect. I just wanted something directionally in the right vein, something that gave me enough information to understand which areas I really needed to crank the dial on.
It was really interesting. They both output a set of suggested changes we could make. I remember giving feedback that morning on some of the construction of how I actually wanted things to look. Thirty minutes later, it was all visualized for me when we got in the meeting. I was like, “Holy shit, this is a really fantastic use case.”
They did it in a way that I don’t think they would’ve been able to do by hand. As much slop as can happen, I also think some magic can really get unlocked if you use the right tool for the right purpose and understand what’s good enough and what needs to be really picture-perfect.
A couple of comments on that. A key unlock is the data, right? With AI, you can consume all these sources of data and quickly do the fuzzy match, get a directionally correct or precise reconciliation across these sources. Whereas before, you had to run the SQL query, wait, and then see it. Essentially, you had to be your own BI analyst on top of it.
The second piece is there’s a muscle being developed around automation. Even if there is slop—for example, I was talking to one of our growth-stage CFOs, who said that in a month or two, 40 or 50 agents had been created by their team. Obviously, they’re not all being used, but it was getting everyone into this motion of thinking, “If I’m doing something manual or tedious and I don’t have a great tool for it, let me build.”
I would guess 90 percent of those tools and agents aren’t being used anymore, but it got people to think, “Instead of doing this manually, let me think automation-first around it,” which I think is almost as important as getting the agent built.
Yeah, I 100 percent agree with you. I also think about how you share it and how you get other people to understand what other people are building. You end up in scenarios where our internal audit team spends a ton of time following up with individuals to make sure they’re adding information to the tickets so it can be tracked for our SOC audit.
Then you realize the internal audit team isn’t the only team doing that. Our GRC team is doing the same thing as they go through all of our ISO and SOC 2 audits. There’s a bunch of back and forth on the AP side when they’re writing approvals. You want to build this motion of triaging, pulling information in, and reconciling it. Why not build a single platform that does that?
This is where I think a systems team is really interesting: coming in to harmonize some of the things you see happening and really productize them, so you don’t have all these individuals building similar things. I think that’s the evolution of what a CIO function should or could be: taking a lot of these bottom-up things people have crafted and having them rebuilt and repurposed in a way that can live on beyond the creator and the initial builder.
I think that cross-org sharing is happening in some ways at our all-company all-hands, but we’re also doing a lot of it internally across our teams.
I was also talking to another one of our growth-stage companies. The head of customer success was saying every one of his reps had a different dashboard that they had vibe-coded, and they had different numbers.
Yeah.
To your point, there has to be some reconciling function, which is a great segue to hear from you guys about what people are doing in terms of the teams that should be building this. Who’s taking the agency? Is it that someone initially builds an agent from your team, and then you take it to an engineering team centrally? We talked a lot about this idea of the finance engineer, which we can unpack further, but we’d love to hear a little bit about that.
Yeah. I think having specialists come up with the ideas and iterate is great; they know what good looks like. They know what they want and the process they’re trying to make work. They’re fantastic at building these initial prototypes.
But do I want that same individual to make sure they’re checking for all upstream dependencies and that, if they were ever to leave, this thing is hosted the right way? And does anyone else across the organization have the same need? Probably not.
This is where I think a world-class CIO organization can come in and really show some magic. I’m really excited. We just had a new CIO join yesterday, a guy named Kasper Sørensen, who joined from Coinbase. A lot of what we talked about in the interview process was how to create centralized PM-type functions within the CIO organization that can partner with the rest of the business teams, find the things that are really working, and then build foundations that others can build on top of. That’s not too dissimilar from how a systems organization would have functioned otherwise, but what’s changing right now is the pace at which you can iterate and how you can empower others to build on top of that.
Is it centralized right now within the company? Do you see that changing eventually?
Yeah. At Figma, we’ve grown faster than we’ve built the tools to support that growth. It’s on me to empower the organization to build, and I don’t want to slow people down by creating infrastructure on top of it, so I think we’re chasing a little bit of that right now. With some teams, we’ve done a better job of centralizing things; other teams have built much faster, and now we’re figuring out how we can help. I think we’re in the messy middle right now.
Mm-hmm.
I think that resonates with a lot of what we’re hearing from our companies, whether it’s a go-to-market engineer, a talent engineer, or a finance engineer. It’s a decentralized set of builders across teams, and they could have different profiles. Sometimes it’s someone with an engineering background on the finance team who’s good at thinking about what workflows need to be automated.
Yeah.
Other times, you’ve borrowed a product engineer from another team to go build. I think everyone’s in this messy middle right now in terms of what that person’s profile looks like.
A lot of it reminds me of data science teams and how they partner with other parts of the organization. The deeper you get into understanding the context of the team you’re working with, the better you operate as an individual. We’ve got data science individuals partnering with our product team. They build as much, if not more, product intuition than our product managers and engineers, and they’re viewed as part of that core team that can help ideate and push change. They’re just leveraging a different tool to enable that.
Similarly, we’ve got people on the data science side supporting our finance organization. It’s really interesting: some of the people who’ve been devoted there actually come from accounting backgrounds. They understand that this stuff has to be really hard and always correct. They understand the nuances we’re describing around checks and balances: What is a refund? What is uncollectible? All that kind of nomenclature. That actually enables them to be better builders, and I could see that translating to the finance engineer you’re describing. You have some shared understanding of what’s necessary. You can build that context over time, but it’s important to build that muscle so you can also push other people’s frame of thinking.
Who owns the definitions? The reason I’m asking is that, especially at an early-stage company, it could be a team of 20 people with just the founders and no finance or business people. They’re building definitions of key metrics that serve reporting, investors, and business decisions, and eventually they bring in finance. How often would you revisit those definitions? In the past, it felt like we revisited them every couple of years. Now, because you can access the code or the Stripe logic much more easily with AI, has that changed? How often do you revisit them?
Yeah, my team owns the outputs there. Maybe we’re a little later than some of those earlier businesses. We have more controls in how we want to talk about it. It’s always been really important to me to have consistency in how we talk about our business. It’s one of the things that’s always empowered me to show up for our board, our leadership team, and other people broadly: they know that what I’m putting out there is the same thing I would’ve put out there a month ago, and the historicals never change. I’m not trying to hide anything.
As we grow and give ourselves the ability to iterate, we constantly have conversations about different sets of disclosure as we go through earnings. We want to be consistent, and we want the decisions we make to be things we can look back on and feel good about, and that have longevity. I think it’s great if people are finding different ways to think about their business, if that’s empowering them to build their business better. But I do think there’s value in consistency as well.
Who on your team has automated the most? What role?
I could say someone, but I’m assuming someone else is probably coming up with something different really quickly. Honestly, it’s individuals who own certain parts of our forecast model, all the way through to individuals on our BizOps team who are automating their day-to-day, like running competitive benchmarking and pricing studies. I think it’s the one-upping of each other that enables you to create an organization that wants to better one another. I’d feel uncomfortable saying there’s one individual, because I truly think this is a team trying to do this together.
Do you think it tends to be people who are more expert, who have more expertise or experience in an area, or is it really across the board? And how much is it coming from you—
Yeah.
—and the role that you’ve automated yourself?
Yeah. Early on, it was helpful for me to show folks, “Hey, if I’m trying to do things in a different way, that should be a signal to you that my expectation is that you are as well.” But in large part, I think what’s important is having a solid understanding of the data sets you’re trying to pull from. The people who’ve been able to move more quickly understand how these things are stitched together. When you understand the things and how they’re architected, you can think more like an architect: “How do I bridge these things in a logical way? Where do I know the model is making a deterministic decision versus inferring? Which tasks is it okay for that to happen on, and which tasks is it not okay for?”
I think others pick up and learn from that. The rhythm we’ve created—the show-and-tell rhythm, the internal hackathons, and the ability to bring in other experts—also helps. I have a lot of conversations with other finance leaders, and I try to bring some of that intel internally so it can get disseminated. My leaders are doing that too, which is really cool to see. Everyone right now is trying to learn from one another, which is really cool to see.
Nice.
Yeah. One of the common themes coming up in my conversations with CFOs, directors of finance, and even individual contributors is that they’re finding it really fun to be working again. Before, you were reconciling something for 2 or 3 weeks, or closing the books for 2 or 3 weeks—or continuously. Now it feels like people are really turning their eyes toward the business and are able to surface insights for the business. There’s a lot of excitement around what’s possible with AI and vertical solutions.
One of the things I’m curious to go back to is the buy scenario. I was talking to one of our hypergrowth companies the other day, and they said, “We just have to buy everything off the shelf.”
“We’re moving so fast.” Do you ever feel that way? Like maybe building is the longer path than just buying one of the solutions out there?
I think for the core systems of record and the things that we just can’t mess up, yes, we’re buying solutions. The areas where our team is having a lot of fun are workflow automation. If I think about my day, what are the areas where I’m almost not thinking? How can I augment my day-to-day so I can reorchestrate how I want to spend my time?
I fully agree with the sentiment you’re sharing: the work has always been interesting, but it’s gotten more fun because you can explore things more quickly. I’ve had folks on my team come up to me and say, “Holy shit, I’m actually having a ton of fun.” I think the policies we’ve created and the freedom we’ve given folks to explore are why some people are saying, “I want to stay here. I really, really like what I’m doing.”
I don’t know if folks are treating access to technology as a recruiting vehicle as much as they could or should. I’ve started doing that in recruiting conversations since I began hearing it from the team, especially when you’re trying to recruit people who aren’t necessarily in technology. They’re like, “Oh my gosh, this is my front entrance into an entirely new world, and if I’m empowered to learn, this is really cool.”
Yeah.
I’d lean into that. I know other people might steal this now, but for me, that was one of the really fun things to do.
No, I agree. I had this idea too: I think we need to start talking about this like we talk about perks and benefits. You get X, Y, and Z if you come to work here, things your former employer doesn’t make available, and now you also get access to all these models or other tools.
Yeah. I think it’s not just that. It’s also about helping people understand what others around them are building. So much of what you’re going to learn comes through osmosis from what others have built. If you get the feeling that you’re joining an organization that isn’t just saying it gives you access to these things, but is actually building—and you can see and feel that through the interview process—those are the places I’d gravitate to.
In 2020, my partner Angela and I wrote a piece about how CFOs needed modern tools. We had this pyramid diagram in there that essentially said the finance team was spending all its time on the data reconciliation layer.
At the top, where the least time was being spent, was strategic finance and strategy questions. I’m curious whether you’ve actually seen your time and your team’s time freed up—not just to do things they enjoy more, but to be more strategic, weigh in on more decisions, or change how you approach your role.
I feel fortunate that, as I’ve grown within the company, that has taken up more and more of my role. As I think about what’s exciting for others, I also recognize the benefit of creating that same space for other people in the organization.
There are things we’ll have to do as daily or weekly rhythms. But if I can take our finance organization—which is split into pillars, one of which is go-to-market—and get the team focused on our self-serve business out of reconciling last week, last month, or whatever it is, they can focus much more on the experiments we’re running, the trends we’re seeing, and where we should be spending additional dollars. That’s way more engaging for them, and I think they’re able to do more of it now.
Some of the little things matter so much. How can I automate the creation of materials that I know are going to look and feel exactly the same, and then focus my time on annotation, discovery, and understanding? We’ve seen that get a lot faster, and the quality improve. We even had one of the leaders of that team—I was actually the leader of that team—get up and talk about how he’d created plug-ins in Figma to ideate on this stuff really quickly.
I’d hope my team would tell me they’re spending a growing share of their time on that bucket of things. But you’re always going to have to chase the complexity in your business and manage it alongside how much time you’re able to reserve for the other pieces.
Totally. One of the things we’re seeing in the CFO tools we’ve invested in is that the product is moving away from tracking workflow toward providing a result, with the audit trail as part of the product.
Yeah.
Finance obviously needs to make sure everything’s correct.
Totally.
You need to be able to track how the product arrived at a particular reconciliation, for example. But then it’s done for you.
Yeah.
So it’s a balance between having the audit trail and having the result. But we’re definitely seeing our teams spend more time running experiments and doing things that used to take a month-long research study. You had to bring in someone from data science, chase down a bunch of people over email, and get everyone in a room. Now you can quickly iterate on this because you can—you have agents running in the background.
And it could even be things on the accounting side, like flux analysis. You’re spending time as an accountant reviewing the GL accounts period over period. If you can use a tool—and there are a whole bunch of tools out there now—that gives you an initial stab at what has changed, that gives you a set of things to chase. Maybe it’s right right away, and I’d love for that to be more and more true. Then they can focus their time elsewhere.
I fully expect that initial draft, that initial set of expectations, that initial idea of what to pull the thread on, can be surfaced via tools. It’s kind of where I was going when I was talking about that T&E piece. It was giving our team sets of threads they could pull, and giving us enough to go and chase them, too.
Right. You have the data and now the time to address those second-order questions and see what’s behind what’s going on.
Yeah. I was thinking about how, when we grew up, the word “reconciliation” was a very negative word. It was associated with a lot of bad memories and a lot of weeks and months spent trying to chase something. Now it’s just something you do, or maybe it’s even a positive word, because you can solve it much more easily and quickly with modern tools and AI. You can find—not necessarily exactly where the errors are—but get there so much quicker.
You still need to go talk to people, obviously, as FP&A will always continue to be the people discipline.
Yeah. It’s one of those things I think I underplayed. I didn’t realize how important communication, relationship-building, and your ability to influence were to this role. I probably walked in feeling like I could be the smartest person in the room and figure it all out. But then I realized it’s a team sport regardless, you can only affect so much change, and you really do have to get other people on the same page with you.
If you can do that effectively, I think you can be a really effective leader. Those are the true skills. You were asking earlier what I screen for. If I can test for and get a sense of someone’s ability to build really solid followership and trust with others, that’s invaluable.
5. What Changes for the CFO
That’s a good segue into what I wanted to chat about next: What do we think changes in the next 12 to 18 months for the team, the role, and the function overall? I don’t know if you guys have any points of view on what might change. Maybe let’s start by talking about how your team is organized.
I would love to be in a spot where we’re only adding individuals if the complexity of the business is growing faster than the needs to support it. I do think we’re in a spot now where a lot of the regular business rhythm pieces can start to move in the reverse direction, versus needing to add people to keep up with them.
For us, we’re going to continue to fight it because we’re adding new products, acquiring teams and businesses, and continuing to evolve what we’re doing. In that world, the complexity is still outpacing where we’re able to truly drive complete efficiency. But we’re definitely getting more efficient, period over period.
Do you think efficiency will plateau, like you’ve got the tools to do what you need to do? Or do you think more of these processes will just get completely automated end to end?
I would hope that more—
Yeah.
—and more can get completely automated. I think maybe I’d start with the pieces that are truly deterministic: How can I feed them enough information so that I can feel the output is completely trustworthy? I think the bar for getting to audit-ready output for a system that’s truly business-critical is extremely high. It’s also going to be an interesting conversation with the auditors about what gets them comfortable.
I don’t take it out of the picture. I do think we’ll get there over time. As I was sharing, if our business complexity were staying the same, I think we probably wouldn’t need the same number of people a year from now. I think you only get there because you’re doing a better job of finding places and rhythms where you can build more consistency in the way you’re operating.
Is Excel under threat?
No, it’s like my love language, dude. In the same way I was describing how I still think a case study is so important, you use it as a tool to build intuition. A lot of the output people are still creating is a model. That model could be hosted somewhere else; it could be something else. But I think you want this tool to give you the ability to refine your understanding and play with things with your hands.
It’s kind of the magic we see with Figma. We can give people the ability to make an initial stab, a first draft, through prompting and natural language. But sometimes it’s so much easier to riff and make changes by hand. Take a color wheel as an example: You’ve got a screen you’ve designed, and you want to change the background. You could sit there with an agent and prompt it to change the background from red to green. Or you could use the color wheel to get all the different shades and a real feel for what’s going on.
I think about Excel similarly. It’s the way I build trust and understanding. There may be something better in the future, but for me, it would be really hard to rip Excel away from me.
I also grew up in Excel, and I still use it to organize my thoughts. If I’m trying to run an analysis, it’s just even the spatial organization of it.
Yeah.
I think the models aren’t quite good enough to create a forecast or a model yet. But I think that organization is still valuable. I don’t know if you feel similarly.
Yeah, totally. Part of what I’ve tried to experiment with now is how I can use one of these models to better understand the output a team is putting together. If you give me a 20-tab model, it’s going to be really challenging for me to thumb through everything. But if I can pass that same context through a model to help me understand the needle movers or the areas that are more sensitive, it’s given me a better starting spot to orchestrate and understand things. I’ve found it really helpful and effective.
The world we’re living in is challenging, too. I was talking to a university professor the other day, and I asked him how they teach AI at the university. They don’t right now. One of the challenges they mentioned was that they had a course, and within a semester they changed the syllabus several times because new models had dropped.
Mm-hmm.
They just have to completely change how they do it, so they sort of give up. They’re going back to fundamentals again, teaching people analytical thinking and Excel and all of that, because I imagine you’re in a similar world. If you were relying too heavily on AI, you’d have to redo how you do your earnings.
A lot has changed, but a lot hasn’t changed.
Yeah.
I think maybe that’s the spirit of where you’re going. The core capabilities of what makes a great finance person are still there. How you leverage the tools at your disposal to make better decisions is always going to be a moving target.
The unlock SQL was for me is now a net-new unlock with AI as a tool and capability driver for me. I think a great finance person has longevity if they continue to adapt.
We wrote this piece titled “You Need a New CFO.” The main point was that there’s a mindset shift. Whether the individual has changed, they have to think automation-first: either build or buy the tools, depending on what’s available and what makes sense. I’m curious if either of you think the profile of a CFO in the next year or two is going to change in any way, or if it’s the same dimensions that are important, but the mindset continues to evolve.
If I were to give advice to anyone, it would be: Hire someone who feels a similar amount of business ownership and wants to bring the same operational rigor as you do as the CEO. They want to win, however possible. They have a set of tools they’re going to use and an understanding of a P&L to orchestrate that change.
You truly want a business leader versus someone who’s a traffic cop. There are a lot of things you need to have as commonplace—a solid understanding of all the accounting fundamentals and the ability to orchestrate a planning process. But I think the things that make people truly unique in this role are: Do you understand the business? Can you make change happen? Can you simplify how the business model works so that others are all running in the same direction?
I think it’s true whether we’re talking about a public company like Figma or a seed-stage or Series A startup that’s starting to scale its go-to-market and getting traction. That’s what founders and CEOs really need from finance: an extra business opinion and insights that are easier to get. But you still go back to the basics, like customers.
Yeah.
Do you understand the customers? Do you understand their behaviors? Do you understand the metrics that come from those inputs and outputs? I think that will always be the same in what we expect from—
Yeah. Maybe the thing I’d add is that creativity is important. Not just creating a fun, funky set of metrics for the sake of being creative, but thinking about your business in a slightly different light and using that to drive change and be opinionated. I actually don’t know how…
I try to do this. I don’t know if I do it as much as I could, but when I come up with these ideas and different ways of thinking about things, I think some great ideas can come from your finance organization. If you set them up well, they can be a layer within your organization that’s just as impactful and really drives a lot of the internal strategy, too.
At the investor and board level, we’re seeing a much deeper set of analysis come through, and rapid iteration. Before, it used to be, “Oh, we have all these questions at the board meeting or before,” and then the team needed to run and cycle through them. That process has short-circuited and, hopefully, is less painful on the finance team’s side.
Last question for you, Ivan. As our portfolio companies, or just companies in general, come to you and say, “Where do I start in terms of automation?” what advice do you give them?
I think it’s the same place where we always started: the integrity of the data. You have to figure out if you have good access to data and whether you’re tracking things. Many times, especially at early-stage companies, they don’t track anything, so you have to set that up.
That often goes against the priorities the first finance hire might have, like, “Oh, I’ve got to close my books on time. I’ve got to send the board update.” Actually, you can’t even do that until you’re on a solid foundation. Sometimes you have to set it up; you might have to rebuild it or change things. We’ve had a number of people come in who can now quickly realize, within several weeks, that maybe revenue is calculated in a way that’s not how it should be calculated for that business model.
More creative ARR at the early stage than I’ve ever seen before.
For sure.
Yes.
For sure. It’s changing fast, too. But it starts with data, and then, based on that, you create the reporting rigor. You create the metrics, and then you create the story. One of the things, too, Praveer, especially since you’ve been on that journey as a CFO, is that storytelling becomes really, really important once you have your eyes set on the IPO.
Mm-hmm.
And you have to all be telling the same story. I feel like that has moved earlier and earlier, because now you expect to tell the story in fundraising for Series A and Series B—not just the story of what the product vision could be, but the story with the numbers and what the business could really be like. That’s gotten easier, but it’s also gotten harder because I think a lot of people overthink it with AI.
Ivan and I have been hosting these CFO and controller dinners for a couple of years now, and we saw a real market change maybe nine months ago. It went from people saying, “Oh yeah, I’m generating this tax memo or this onboarding doc with AI,” to saying, “Okay, yeah, I can actually—I’m doing the core data reconciliation and automating many of the workflows.” It’s really been a step-function change.
If anything, the leaders are saying, “I need to do this; otherwise, I’ll be perceived within my organization as behind everyone else.” This goes back to your point about winning the hackathons. Finance was never the one that was going to be winning the automation hackathons a couple of years ago, and it’s pretty cool to see that’s totally changed.
Yeah. I love this part of it. We have an internal Maker Week. It’s our internally branded hackathon, and in the early days I’d sit there and not really know how to participate. Now, if I had a spare minute, I think I’d have a million different ideas, and I’m sure my team feels very similarly.
Well, Praveer, thank you so much for joining us. This was really fun to talk about and to understand directly from someone who’s gone through this journey of automation over your career. And thanks, Ivan. This was a really fun conversation.
Yeah, thanks for having me.