Ivan Makarov

1. Key Themes

From Traffic Cop to Business Enabler

Melwani's core thesis is that the finance function's identity is shifting. Automation isn't mainly about speed; it changes what finance people spend their time on. He frames the role as an operational leader rather than a gatekeeper. Describing his early years at Figma, he said: "I've always felt like the role of a finance individual should be this operational leader" [00:03:01]. He added: "that's the magic of shifting this role from what can be viewed as like a traffic cop sort of finance-y type function to one that's a business enabler and a true change agent within the organization" [00:03:41].

The "Messy Middle" of Bottoms-Up Automation, and the Role of a Central Systems Team

Specialists inside each function build the best prototypes because they know what "good" looks like, but they shouldn't own the productionization. Melwani explained: "do I want that same individual to be the individual that has to go and make sure that they're checking for all upstream dependencies and making sure that like, you know, if they were ever to leave, that like this thing is hosted in the right way?" [00:22:56]. His answer is a central CIO/systems group acting as product management: "how do you create these like, you know, 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 these like foundations that other can others can build on top of?" [00:23:10]. He is candid that Figma is mid-transition: "I think we're in the messy middle right now" [00:24:14]. Seema Amble confirmed the pattern across a16z portfolio companies, citing a customer success head whose reps each had vibe-coded dashboards showing different numbers [00:21:55].

Data Integrity Is the Real Unlock (Not the Model)

Both Melwani and Ivan Makarov emphasized that automation sits on top of clean, consistently defined data. Melwani on his early SQL work: "if I can get consistent in our definitions, in the way that we're talking about things, no longer are we ever debating like what's true and what's fact" [00:04:55]. On the T&E analysis: "95% of the challenge there is just like getting your data in the right spot, and these models are good at making making these assumptions for you" [00:00:38]. Makarov's advice to early-stage companies: "You start with the integrity of data. You have to figure out if you actually have a good access to data, if you're tracking things" [00:47:05].

Don't Outsource the Thinking

The recurring guardrail is that AI is a tool, and judgment can't be delegated. Melwani: "I think you don't want to outsource your thinking today... You want to use AI as a tool to help you and empower you, but it's really, really critical to still be the thinker" [00:09:48]. He ties this to the cost of "slop": "Anything that you're putting out there that you're asking someone to consume, it's going to take them time, effort, and energy" [00:00:38]. Figma has policies including labeling content written with AI [00:17:02].

AI Tooling as a Recruiting and Retention Vehicle

Melwani argues access to modern AI tools is an underused talent lever: "I don't know if folks are actually treating access to technology as like a recruiting vehicle as much as they could or should" [00:32:00]. He is using it to pull candidates from non-tech backgrounds: "This is like my front entrance into like an entirely new world" [00:32:00]. He expects it to produce a flywheel: "I think we're going to be able to attract and build the world's best finance organization because so much what people want is the ability to learn" [00:00:00]. Makarov extended the idea: "we need to start talking about this like we talk about perks and benefits" [00:32:30].

Culture of Show-and-Tell and Friendly One-Upmanship

Diffusion of automation is driven by social mechanics, not mandates. Melwani: "it's the one-upping of each other that actually enables you to create an organization that wants to better one another" [00:28:36]. Figma uses all-hands segments, hackathons ("Maker Week"), and cross-company finance-leader conversations to spread ideas [00:30:18]. Amble noted that even if 90% of agents built get abandoned, the muscle matters: "it got people to think, 'Okay, instead of doing this manually, let me think automation first around it,' which I think is almost as important as getting the agent built" [00:20:24].

Determinism vs. Inference as the Architect's Core Skill

The people who automate fastest understand data architecture and know where a model may infer versus where it must be deterministic. Melwani: "Where do I know where the model is actually making a deterministic decision versus inferring? And what tasks like is it okay for that to happen and what tasks is it not okay?" [00:29:20]. He expects full end-to-end automation to start with deterministic processes, with audit-ready output for business-critical systems being "extremely high" bar and "an interesting sort of conversation with the auditors" [00:39:39].

Finance Moving Up the Strategy Pyramid

Amble referenced a 2020 a16z piece in which finance teams spent most time on data reconciliation and little on strategy. Melwani described the shift: "If I can take the team that's focused on our self-serve business and immediately get them out of reconciling last week or last month... and then very much more focused on what are the experiments we're running... like that is way more engaging to them" [00:33:59]. Makarov observed that "the word reconciliation was a very negative word" and is now becoming something you simply do quickly [00:37:03].

Headcount Leverage Is Coming, but Complexity Offsets It

On the 12-18 month outlook, Melwani was direct: "If our business complexity was staying the same, I think we probably wouldn't need the same number of people a year from now" [00:00:50]. But Figma keeps adding products and acquiring teams, so complexity growth is outpacing efficiency gains for now [00:39:04].

Soft Skills Remain the Differentiator

Despite the tooling, Melwani stressed influence and relationships: "I didn't realize how important communication, relationship building, your ability to go and influence was 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" [00:37:32].

2. Contrarian Perspectives

Excel Isn't Going Anywhere: It's a Thinking Tool, Not Just a Calculator

When asked if Excel is under threat, Melwani answered flatly: "No, it's like my love language." His reasoning is about trust and tactile understanding, not capability: "I think you want this tool to give you this ability to refine your understanding and play with things with your hands" [00:40:33]. He drew an analogy to Figma itself: prompting an agent to change a background color is possible, but "I could take the color wheel and you get all different shades of it and you get the real feel" [00:40:56]. Amble added that models "aren't quite good enough to like create a forecast or a model quite yet" and that the spatial organization still has value [00:41:32]. This runs against the narrative that spreadsheets are about to be displaced by agents.

Most Agents Built Will Be Thrown Away, and That's the Point

Conventional wisdom treats unused tools as waste. Amble offered the opposite view, citing a growth-stage CFO whose team built "40 or 50 agents" in a month or two: "maybe 90% I would guess of those tools and agents are not being used anymore, but it got people to think, 'Okay, instead of doing this manually, let me think automation first'" [00:20:24]. The value is the habit formation, not the artifacts.

Don't Standardize the Dashboards Too Early, but Do Standardize the Definitions

Melwani holds a strong position on consistency, especially for a public company: "they know that what I'm putting out there is the same thing that I would have put out there a month ago and the historicals never change, and I'm not trying to hide anything" [00:27:13]. Even as AI makes it trivially cheap to redefine metrics or build bespoke views, he sees consistency as a source of credibility with the board and leadership: "I do think that there's some value in consistency as well" [00:27:52].

Diligence by "Can You Vibe-Code It?"

Figma's corp dev team, which has finance backgrounds rather than product backgrounds, now tries to recreate a target's product as a diligence step. Melwani: "can you vibe code a product is one of the checks that we use in diligence now, which is actually really fun to see. Like how deep is your moat? How interesting is the technology there?" [00:12:15]. Amble's reaction ("Wow. Okay.") reflects how unusual this is. The implication is that product defensibility can be pressure-tested by non-engineers in hours.

AI Won't Teach the Next Generation Finance, so Fundamentals Come Back

Makarov relayed a professor's experience: the syllabus for an AI course changed several times in a single semester due to new model releases, so they returned to teaching fundamentals: "analytical thinking and Excel and all of that" [00:42:50]. Melwani agreed: "A lot has changed, but a lot hasn't changed" [00:43:06]. This runs against the instinct to rebuild curricula around current tools.

3. Companies Identified

Figma

Collaborative design and product development platform, publicly traded; Melwani's employer. Mentioned as an AI-forward finance organization and as a product being used internally for finance workflows. Melwani described plug-ins built in Figma by a finance team leader: "he had gone and create a set of plug-ins to go and you know, in Figma, to go and ideate on this stuff really quickly" [00:34:45]. Also mentioned: "We have a tool called Figma Make, which allows folks to be able to prompt through natural language and create an application, a website. You can then host it" [00:15:37].

Andreessen Horowitz (a16z)

Venture firm hosting the podcast and an early Figma partner. Melwani said: "our business took off during the early days of the pandemic. I think that was actually the time where we started to partner with the folks at Andreessen" [00:05:55]. Amble also noted a16z's CFO and controller dinners and the "You Need a New CFO" piece [00:48:34].

Stripe

Payments platform. Melwani's early financial data mapping: "mapping it to all the financial information we had in Stripe" [00:04:13]. Makarov also referenced accessing "the Stripe logic much easier with AI" [00:26:40].

NetSuite

ERP system of record. The credit and collections workflow reconciles "everything to NetSuite" [00:11:11].

Navan, Zip, and Ramp

Travel/expense and procurement/spend management tools. Used in Figma's T&E review: "we use Navan, we use Zip, we use Ramp, and there's a lot of fuzzy match that you have to go through" [00:18:10].

Codex (OpenAI)

AI coding agent used by Figma's corp dev team: "they're sitting in Codex and trying to, you know, work their way through something" [00:15:13].

ChatGPT (OpenAI)

Used for workflow automation: "building skills around agency using ChatGPT, and so that's how some of their credit and collections pieces were being built" [00:15:13].

Salesforce and Marketo

Go-to-market systems fed by Melwani's early golden tables: "pass it to Salesforce, or Marketo, or whatever systems that we're using at the time" [00:04:38].

Webflow

Makarov's former employer, where he saw COVID usage spikes: "When I was at Webflow at Figma, we saw, for example, a huge spike in usage in COVID" [00:05:28].

Coinbase

Previous employer of Figma's new CIO, Casper Sorensen [00:23:41].

4. People Identified

Praveer Melwani

Figma CFO; joined in 2017 as the first business operations hire and helped take the company public. Mentioned as the guest and as a model of the operator-CFO who built his own SQL tables and still reviews hourly dashboards: "I'm constantly in there going and taking a look at, you know, hourly dashboards or sort of daily metrics on things to make sure that I'm also sourcing things that I think maybe look like anomalies" [00:05:55].

Kajal (Figma credit and collections)

Rebuilt his entire workflow with AI. Melwani: "Kajal in our credit and collections team, he's gone and rebuilt his entire workflow, everything from invoice tracking, cash collections, making sure that we can reconcile everything to NetSuite, and he's done it in a way where it's like a homegrown system" [00:11:11].

Casper Sorensen

New Figma CIO, joined from Coinbase. Melwani: "We just had a new CIO join yesterday, a guy named Casper Sorensen who joined over from Coinbase" [00:23:10]. His mandate is to create centralized PM-type functions that productize the best bottoms-up automations.

Dylan and Sean (Figma leadership)

Dylan is a Figma leader, and Sean is the CRO. Figma's corp dev team built AI personas of them for investment memo review: "This is what Praveer would say and this is what Dylan would say and this is what like, you know, Sean would say, our CRO" [00:13:53].

Seema Amble

a16z partner and host; co-author of "You Need a New CFO" and a 2020 piece on modern CFO tools with partner Angela. She contributed portfolio observations on finance engineers, agent sprawl, and audit trails.

a16z partner and host; former Webflow finance leader. Contributed views on data integrity, definitions, and storytelling.

Angela (a16z partner)

Co-wrote the 2020 piece with Amble on why CFOs needed modern tools, featuring a pyramid showing finance time mostly spent on reconciliation [00:33:27].

5. Operating Insights

Hire for Case-Study Depth, Not Output

Melwani's VP Finance interview centers on a case study with sparse information. The test is whether the candidate understands the model they built, not whether they can produce one: "have you really, really understood the structure that you've created? Have you molded it in a way where you've tested what are the pieces of that model that are the most sensitive to any change?" [00:09:19]. Anyone can get a model out of an LLM; the filter is sensitivity analysis and intuition for how the business works.

Run Your Own Review with a Model to Navigate 20-Tab Spreadsheets

Melwani uses an LLM to digest team-built models: "if you give me like a, you know, 20-tab model, it's going to be really challenging for me to go and thumb through everything. But if I can pass that same context through a model to help me understand what are the things that really are the needle movers or what are the areas that are more sensitive, it's actually given me like a better starting spot" [00:41:55]. This is a leader-side tactic for faster review.

Build Investment Committee Simulators from Leadership Personas

Figma's corp dev team built personas of each executive to pressure-test memos before presenting. Melwani explained the benefit: "you get the areas of the document that you either want to go in plus up, do more diligence on. What are common sets of feedback that I need to go and make sure I ask the questions over? And then it reduces the cycle time that we're all sitting together on" [00:13:53].

Target "Directionally Right" Where Perfection Isn't Needed

Matching rigor to the use case is a core operating principle. On T&E: "I didn't need something 100% perfect. I just wanted something that was directionally in the right vein... that gave me enough information to understand what are the areas I really needed to crank the dial on" [00:18:32]. Pair this with the discipline to know "what is good enough and what needs to be like really, really picture-perfect" [00:19:08].

Leaders Must Visibly Change Their Own Workflow

Melwani described how early signaling set expectations: "if I'm coming and trying to do things in a different way, like, hey, that should be a signal to you that my expectation is that you are as well" [00:29:05]. He also uses every spare minute across hackathons and show-and-tell to model behavior.

6. Overlooked Insights

Collections and Credit Teams as Hidden Automation Leaders

The only individual contributor named for automation excellence is on credit and collections, not FP&A or corp dev. Kajal "rebuilt his entire workflow" and then "he can now go to anyone to say, 'Hey, go and leverage the tools that I've created'" [00:11:40]. The implication is that the least glamorous, most rules-based back-office functions tied directly to cash (collections, AP accruals, SOX ticket chasing) are often the highest-ROI and most automation-ready, and their builders become internal distribution channels for tooling.

Duplicate Work Across Audit, GRC, and AP Signals a Shared-Platform Opportunity

Melwani offhandedly listed three teams doing the same underlying task, chasing people for information to close tickets: internal audit (SOX), GRC (ISO and SOC 2), and AP (accruals). "Why not go and build a singular platform that does that?" [00:21:00]. This is a quiet product-and-investment thesis: cross-functional "evidence collection and follow-up" workflows are being rebuilt in parallel inside companies, which is the type of seam where an horizontal agent product, or an internal platform, wins.