By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
a16zHow AI Is Turning Figma’s Finance Team Into Builders
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
CHAPTERS
- 0:00 – 1:29
Why AI-oriented finance can build the “world’s best finance organization”
Praveer explains why he’s optimistic about building an exceptional finance team: people want to learn, be invested in, and build alongside strong peers. The hosts frame the episode around a shift in the CFO role toward building systems and automation.
- •Praveer ties talent attraction and retention to learning, investment, and craft development
- •Seema frames a16z’s view that the CFO role increasingly includes building and automation
- •The conversation sets up finance as moving from gatekeeping to enabling and change agency
- 1:29 – 2:48
Automation shifts finance from “traffic cop” to operational business enabler
Praveer describes how he views finance as operational leadership, enabled by automation and strong data foundations. He recounts joining Figma when automation wasn’t in place and using technical work to scale his impact across the company.
- •Finance’s role is positioned as operational leadership rather than policing
- •Early Figma finance work required building and orchestrating data to drive company conversations
- •Modern tools make earlier “force multiplier” approaches more accessible to more people
- 2:48 – 4:33
Pre-AI building blocks: SQL-heavy workflows and “golden tables” for consistent metrics
Praveer gets concrete about the pre-AI stack: long SQL queries, mapping product usage to financial systems, and partnering with data scientists to productionize recurring jobs. The goal was consistency in definitions so teams could stop debating what’s true and focus on running the business.
- •Long SQL queries were used to join product/utilization data with Stripe financial data
- •Outputs became “golden tables” feeding go-to-market systems (e.g., Salesforce/Marketo)
- •Recurring ETL jobs helped standardize definitions and reduce debates over facts
- 4:33 – 7:32
Faster cadence and anomaly detection: moving from weeks to hourly/daily context
Ivan contrasts older monthly/quarterly rhythms with today’s daily reality, and Praveer describes how he uses frequent dashboards to rebuild context and spot anomalies. He emphasizes empowering others to feel accountable to metrics and act quickly rather than waiting for weekly reviews.
- •Ivan discusses how spikes once took weeks to understand; now they can be understood much faster
- •Praveer describes routinely checking hourly/daily dashboards and asking anomaly-driven questions
- •He wants broader team accountability and faster triage as business rhythms accelerate
- 7:32 – 8:42
What makes a great finance hire: curiosity, data fluency, and not outsourcing thinking
Asked what makes someone a great hire, Praveer prioritizes intellectual curiosity, strong questions, and clear communication grounded in data. He adds that AI can remove tedious work, but the core requirement remains: the person must still think and understand the business.
- •Intellectual curiosity and problem-solving to understand the business deeply
- •Fluency with data, strong questioning, and structured thinking
- •Communication of insights matters, especially as AI accelerates production of outputs
- 8:42 – 10:49
Interviewing for judgment with a case study: using tools without losing understanding
Praveer outlines a case-study-driven interview process, including expectations that candidates will use available tools. He evaluates whether they truly understand their model, can identify sensitivities, and can communicate clearly—warning against outsourcing thinking even if models are easy to generate.
- •Case studies are used to test tool use, modeling rigor, and understanding
- •Candidates are pushed on model structure, sensitivities, and business intuition
- •AI can speed model creation, but the bar is still deep comprehension and communication
- 10:49 – 15:04
Figma finance automations in practice: collections systems and “vibe-coded” corp dev diligence
Praveer shares examples of bottom-up automation across the finance org, including credit/collections workflow rebuilding and corporate development using coding to assess diligence. He also describes using AI to simulate leadership-team feedback on investment memos to reduce cycle time.
- •Credit/collections workflow rebuilt end-to-end (invoice tracking, cash collections, reconciliation to NetSuite)
- •Corp dev uses a “can you vibe code a product” diligence check and partners with product with more credibility
- •Corp dev creates leader “personalities” to pre-review investment memos and anticipate committee feedback
- 15:04 – 16:08
Build vs. buy and where tools live: Codex, ChatGPT agents, and Figma’s own tooling
The hosts probe whether teams should build or buy, and Praveer explains Figma’s mix of approaches. He describes teams building in Codex, using ChatGPT-based agents, and experimenting with Figma Make for natural-language app creation that can also generate product feedback loops.
- •A mixed approach: some custom builds, some workflow tools, and experimentation on internal tooling
- •Corp dev work is described as happening in Codex; other workflows use ChatGPT agents
- •Figma Make is used to prompt-build apps/websites with hosting and a backend, enabling internal product feedback
- 16:08 – 19:43
Calling out AI “slop”: respect readers’ time and focus on data readiness over perfection
Ivan raises concerns about AI-generated document overload and losing “world-class” standards, and Praveer agrees slop happens. Praveer emphasizes internal policies for disclosure/branding, the responsibility to respect others’ time, and that most difficulty is still getting data into the right place; he shares a travel-and-expense review where fuzzy matching across systems produced directionally useful insights quickly.
- •AI can create too many half-baked docs; teams must maintain a high review/clarity bar
- •Internal norms/policies help clarify when AI is used and preserve accountability
- •A T&E review example shows fuzzy matching across multiple tools and systems, with ‘directionally correct’ outputs enabling faster iteration
- 19:43 – 23:57
From one-off agents to shared platforms: centralizing what works and making it durable
Seema and Praveer discuss how rapid agent creation builds an automation-first muscle but can create fragmentation and inconsistent dashboards. Praveer argues for cross-org sharing and a systems/CIO function that productizes effective bottom-up prototypes into shared, maintained platforms that outlast individual builders.
- •Agent sprawl can be useful for learning, but sharing and standardization become critical
- •Examples of repeated cross-team needs (audit follow-ups, GRC/SOC processes, AP accruals) suggest platform opportunities
- •Praveer describes the evolution toward a CIO-led central function to rebuild and repurpose prototypes sustainably
- 23:57 – 26:25
Centralization in the “messy middle” and the emerging finance-engineer model
Praveer says Figma is still catching up—empowering teams to build first, then adding infrastructure later—so centralization varies by team. The group compares this to data science partnering models and discusses how domain context (including accounting knowledge) can make builders more effective in finance-specific automation.
- •Figma is still balancing speed with infrastructure; some areas are centralized, others are not
- •A ‘finance engineer’ profile may mirror data science partnering: embedded context plus building skills
- •Accounting and finance domain knowledge helps builders handle checks, balances, and correct terminology
- 26:25 – 30:24
Owning metric definitions: finance controls outputs and prioritizes consistency over time
Ivan asks who owns definitions as companies scale and definitions evolve more quickly due to accessible logic and data. Praveer states his team owns the outputs and underscores the importance of consistent definitions and stable historicals so stakeholders trust reporting and disclosures over time.
- •Finance owns metric outputs and prioritizes consistency in how the business is described
- •Consistency supports trust with boards, leadership, and broader stakeholders
- •Iterating can be useful, but Praveer emphasizes longevity and comparability of decisions and disclosure
- 30:24 – 38:20
AI tools as a recruiting perk and a catalyst for more strategic finance work
The conversation turns to how automation affects morale, recruiting, and time allocation. Praveer and Ivan describe finance work becoming more fun as tedious tasks shrink, and Praveer notes he now explicitly uses access to tools and a visible building culture as part of recruiting; he also describes shifting teams away from reconciliation toward experiments, trends, and decisions.
- •Access to tooling and an observable building culture can help attract candidates
- •Automation can free teams to focus on experimentation, trends, and resource allocation discussions
- •Praveer stresses still needing core leadership skills like communication, trust-building, and influence
- 38:20 – 40:35
What changes in the next 12–18 months: headcount, deterministic automation, and audit readiness
Seema asks what changes soon, and Praveer says that if business complexity held steady, fewer people might be needed because more routine rhythms can be automated. He expects progress toward end-to-end automation for deterministic processes, but highlights the high bar for audit-ready outputs and the importance of auditors’ comfort with new approaches.
- •Efficiency gains may reduce the need to add headcount if complexity doesn’t rise
- •Deterministic processes are prime candidates for deeper automation with trustworthy outputs
- •Audit readiness remains a high bar, and auditor comfort will influence the pace
- 40:35 – 46:57
Excel isn’t going away: models as a way to build intuition, with AI as a lens on complexity
Ivan asks whether Excel is threatened, and Praveer calls it his “love language,” arguing models help build trust and intuition through hands-on iteration. He and Seema discuss Excel’s spatial thinking value, and Praveer adds he’s using AI to interpret large multi-tab models by surfacing sensitivity drivers and needle-movers to orient his review.
- •Excel remains central for hands-on iteration and building trust in understanding
- •AI can help digest complex models by highlighting sensitivities and key drivers
- •The group emphasizes fundamentals (analytical thinking) despite rapidly changing AI tooling
- 46:57 – 50:11
Where to start with finance automation: data integrity first, then metrics rigor and storytelling
In closing, Ivan explains that automation starts with data integrity and access—many early-stage companies aren’t tracking properly, so foundations must be built even if it conflicts with immediate priorities. From there, teams establish reporting rigor, metric definitions, and a coherent numbers-backed story, which Seema notes has evolved from simple docs to core workflow automation across the market.
- •Start with data integrity and tracking; foundations may need rebuilding before automation pays off
- •Then establish reporting rigor, metric definitions, and a consistent narrative/story with numbers
- •AI makes iteration easier but can also lead to overthinking; leaders are increasingly expected to automate core workflows
