ExpertsFMThe New CFO: How AI Is Rewriting the Finance Function | Figma CFO Praveer Melwani
The a16z Show · November 5, 2026
5 Selected Lessons:
heuristic · 2 case · 1 priority · 1 reasoning_pattern · 1
Consistency of reported outputs over cleverness or speed
Source
The a16z Show — The New CFO: How AI Is Rewriting the Finance Function | Figma CFO Praveer Melwani (November 5, 2026) · at 26:56 ▸
When this applies
when communicating recurring metrics or decisions to stakeholders (investors, board, leadership) who rely on trust over time
The takeaway
Prioritize stability and non-retroactive consistency in recurring external communications over the temptation to constantly 'improve' how things are measured.
Why it works
If definitions or historical reported figures shift retroactively, stakeholders lose the ability to trust that today's numbers mean what yesterday's meant, undermining the entire basis for decision-making built on those numbers.
One level up
Trust with external stakeholders is built on predictability of interpretation, not on the sophistication of any single report.
Where it stops holding
This stops working when the current metric or definition is simply wrong and the cost of continued inconsistency with reality exceeds the cost of a one-time redefinition.
Detail
hierarchy:
- Historical figures must never change retroactively
- Decisions should be ones leadership can look back on and feel good about
- Innovation in how people measure things is fine if it genuinely helps them run the business, but shouldn't come at the cost of consistency
rationale: Being predictable is what lets him 'show up with our board, with our leadership team' credibly.
In their words
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 I felt has always been empowering that, like, how I can show up with our board, with our leadership team, with other folks broadly is that they know that what I'm putting out there is the same thing that I would have put out there a month ago.
Headcount scales with complexity growth, not baseline workload
Source
The a16z Show — The New CFO: How AI Is Rewriting the Finance Function | Figma CFO Praveer Melwani (November 5, 2026) · at 0:00 ▸ · at 37:32 ▸
When this applies
when deciding whether a function needs more people as its tooling and automation improve
The takeaway
Judge future staffing needs by comparing the rate of complexity growth against the rate of automation-driven efficiency gains, not by extrapolating past headcount trends alone.
Why it works
Automation reduces the labor needed for a fixed level of complexity, but if the business itself adds complexity (new products, acquisitions) faster than automation offsets it, headcount still must grow; the net effect depends on the race between automation gains and complexity growth.
One level up
Resource needs for any function are a function of the derivative of complexity relative to the derivative of capability, not of either alone.
Where it stops holding
This stops working when complexity growth is unpredictable or discontinuous (e.g., sudden new regulation or market shift), making the 'race' framing less reliable as a planning tool.
Detail
approach: Explicitly separate 'automatable regular business rhythm' from 'new complexity being added' (new products, acquisitions) when forecasting headcount.
signals:
- business rhythm processes starting to move in reverse direction (need fewer people)
- but complexity from new products/acquisitions still outpacing full efficiency
rationale: 'If our basic complexity was staying the same, I think we probably wouldn't need the same number of people a year from now... but we're going to continue to fight it because we're adding new products, we're acquiring teams and businesses.'
In their words
I think for us, we're going to continue to fight it because we're adding new products, we're acquiring teams and businesses, we're continuing to evolve what we're doing. And so it's always more, I think, in that world, the complexity there is still outpacing where we're able to truly drive complete efficiency.
Buy core systems of record, build workflow automation
Source
The a16z Show — The New CFO: How AI Is Rewriting the Finance Function | Figma CFO Praveer Melwani (November 5, 2026) · at 28:58 ▸
When this applies
when allocating build-vs-buy decisions for any internal tooling portfolio
The takeaway
Match build-vs-buy decisions to the asymmetry of downside risk versus iteration value for each category of tool.
Why it works
Systems of record carry catastrophic downside if wrong (compliance, audit, correctness), favoring proven off-the-shelf reliability; day-to-day workflow augmentation carries low downside and high upside from fast iteration, favoring custom build.
One level up
Allocate scarce building effort toward the areas where fast iteration creates compounding advantage, and buy where failure is unacceptable.
Where it stops holding
This stops working when off-the-shelf vertical solutions for workflow automation become mature and reliable enough that building in-house no longer offers a speed/customization edge worth the maintenance cost.
Detail
rule: For core systems of record and things that 'can't mess up,' buy proven solutions; for workflow automation and daily-process augmentation, build in-house for speed and fit.
threshold: 'Things that like we just can't mess up' = buy; 'almost like not thinking' repetitive daily tasks = build/automate.
rationale: Building lets the team re-orchestrate how they spend their time and keeps work more engaging, while buying protects against unacceptable correctness risk.
In their words
I think for like the core systems of record solutions and like the things that like we just can't mess up, like, yes, we're going and buying solutions for that. The areas where I think our team is having a lot of fun, it's the, you know, it's kind of where you're going. It's like the workflow automation.
Use AI to understand large artifacts, not replace the trusted hands-on tool
Source
The a16z Show — The New CFO: How AI Is Rewriting the Finance Function | Figma CFO Praveer Melwani (November 5, 2026) · at 40:30 ▸
When this applies
when deciding whether a new AI capability should replace or augment an existing trusted manual tool for building deep understanding
The takeaway
Use AI as a navigation/comprehension layer on top of trusted manual tools rather than assuming it should replace the tool that builds your intuition.
Why it works
Hands-on manipulation (e.g., spatial organization in Excel, a color wheel in design) builds tacit understanding and trust that purely prompted output doesn't provide; AI is better used to help navigate/summarize what's already built (e.g., finding sensitivity in a 20-tab model) than to replace the medium of understanding itself.
One level up
New capability-generation tools and trust-building/understanding tools serve different jobs; don't conflate 'can generate it' with 'should replace the way you understand it.'
Where it stops holding
This stops working when the manual tool's output no longer needs deep human intuition (fully deterministic, low-stakes tasks), in which case full automation without hands-on involvement is fine.
Detail
rule: Keep using the hands-on tool (Excel) to build and refine your own understanding; use AI primarily to help parse/summarize someone else's large existing output and surface the most sensitive variables.
rationale: The models aren't yet good enough to create a trustworthy forecast/model from scratch, but they are very good at helping a reader triage a complex existing one.
In their words
If I can, you know, 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 a better starting spot to go and orchestrate and understand things, which I've actually found really, really helpful and effective.
Bottoms-up automation builds, then centralized via a CIO/systems function
Source
The a16z Show — The New CFO: How AI Is Rewriting the Finance Function | Figma CFO Praveer Melwani (November 5, 2026) · at 19:29 ▸ · at 22:08 ▸
When this applies
when many individual teams or employees start independently building automation tools solving overlapping problems
The takeaway
Let domain specialists prototype freely, but route proven bottoms-up tools through a central platform/systems function that harmonizes, productizes, and sustains them beyond their original creator.
Why it works
Specialists are best at prototyping because they know what 'good' looks like for their own workflow, but letting each prototype live as a one-off creates duplicated effort, orphaned tools, and fragmented truth (different dashboards, different numbers) when the builder leaves or scales elsewhere.
One level up
Decentralized experimentation plus centralized consolidation outperforms either pure top-down design or pure fragmented tool sprawl.
Where it stops holding
This stops working when the central team is too slow or bureaucratic to keep pace, in which case premature centralization just recreates the old bottleneck it was meant to solve (the 'messy middle' problem Figma itself admits to).
Detail
situation: Multiple Figma teams independently built similar AI ticket-tracking/reconciliation tools (internal audit, GRC, AP); a growth-stage company's customer success reps each vibe-coded their own dashboards with differing numbers.
decision: Figma is building a centralized CIO organization with PM functions to find what's working bottoms-up and rebuild it as shared, durable infrastructure rather than individual one-offs.
outcome: Described as currently 'in the messy middle' — some teams centralized, others not yet, with ongoing effort to harmonize.
lesson: A world-class systems/CIO function's new role is to find and productize the best bottoms-up automations rather than dictate tools top-down.
In their words
This is actually where I think, like, you know, a systems team is really, really interesting to come in and then harmonize some of the things that you see are happening, really productize it so that you don't have all these individuals going and building similar things.