Podcast Summary: Metrics That Measure Up

Episode: The Use and ROI of AI in Finance
Host: Ray Rike
Guest: Sowmya Ranganathan (Former Controller, OpenAI; CEO, Lumera)
Date: October 1, 2025

Episode Overview

In this lively episode, Ray Rike talks with Sowmya Ranganathan about the practical deployment and return on investment (ROI) of AI in finance departments, drawing on Sowmya's experience as the former Controller at OpenAI and now as founder and CEO of Lumera. The conversation provides candid insights, concrete workflows, and actionable advice for finance leaders exploring AI implementation, especially in rapidly scaling SaaS and cloud environments.

Key Discussion Points & Insights

1. Sowmya’s Journey to AI in Finance (01:14 – 03:13)

• Background: Sowmya is a CPA who began her career in Canada, moved to San Francisco nearly incidentally, stayed for 12 years, and built experience at Square, robotics startups, Rippling, and most recently OpenAI.
• Joining OpenAI: She joined OpenAI as Controller, coincidentally just as ChatGPT launched for paid users, marking a period of enormous growth and transformation.
• New Venture: Left OpenAI to found Lumera, a company focused on bringing AI-driven automation to finance teams everywhere.

2. Inside OpenAI’s Finance Team During Hypergrowth (04:06 – 05:52)

• Environment: High-growth, intensity, and no playbook could be reused due to the company’s unique challenges and pace.

  “Maybe you thought you were a star player and someone changed the rules of the game right as you were in the game. That's, that's realistically how it felt.”
  — Sowmya (04:21)

• Small, Empowered Team: Only ten people in finance, building from first principles, and constantly iterating.

3. Core Finance Challenges at Scale (06:24 – 09:18)

• Data Explosion: Main challenges around revenue accounting and infrastructure cost accounting due to massive transaction volumes (both from the user side and from compute/GPU usage).
  • Manual approaches (e.g., Excel) quickly became impossible due to dataset sizes.

  “You have a CSV file from a vendor for compute data. You open it and it says, Excel is unable to open this file because it exceeds a million rows. And you're staring at like, okay, now what do I do?”
  — Sowmya (07:46)

• Forecasting Complexity:
  • Revenue: Somewhat prepared due to existing API business.
  • Cost: Required talking to researchers and engineers, not just working from spreadsheets or historicals.

  “If you thought of looking at historical data and tried to pull it forward, like that kind of forecasting mindset just would not have worked at this company hands down.”
  — Sowmya (08:25)

4. Early Use of AI for Financial Workflows (09:49 – 14:45)

• Hackathons & Practical AI: Started with a hackathon where finance staff used ChatGPT to generate predictive models and automate complex data handling.
• Case Study: Financial Close:
  • Used ChatGPT as a ‘writing assistant’ for policies and memos.
  • Key AI breakthrough: Handling massive CSV data by getting ChatGPT to generate Python scripts for data processing.
  • Moved from a one-off script to a cloud-based, real-time, Databricks-deployed solution, enabling scalable automation and cross-team use.

  “I have this, like, four gigabytes of CSV data. Excel won't let me open it. Tell me how I can do my work with this file... [ChatGPT] said, 'Python is really efficient at processing CSV files’...you just tell me what you need done and I can write the Python code for you.”
  — Sowmya (11:17)

• Output QA & Hallucination Risk:
  • Static code created from AI was tested with historical data and edge cases before being put into production.
  • For LLM use cases (like extracting data from contracts), reconciled AI results against authoritative systems (Salesforce/CPQ) and flagged discrepancies for manual review.

  “The hallucination risk really is limited to the part where it's writing the code for you. And once you have this code, it's a static Python script... you can test it for all your edge cases upfront...”
  — Sowmya (14:45)

5. Lessons Learned and Best Practices for AI in Finance (18:02 – 23:31)

• Lumera’s Approach:

  • Started as AI workshops teaching finance teams to use ChatGPT as an “engineer in their pocket.”
  • Evolved into productized script generation and secure hosting/automation, enabling finance teams without internal engineering resources.

  “In an AI world, economics is wildly different because you describe the problem and your code comes 90, 95% done and the last 5% is usually tweaking...”
  — Sowmya (19:59)

• Data Privacy for AI in Finance:

  • Advocate for ChatGPT Enterprise (or similar) for regulatory compliance.
  • Suggest writing code with dummy data and only running scripts on sensitive data in secure environments.
  • Lumera offers such an environment for smaller companies lacking in-house resources.

6. Evaluating and Starting AI in Finance (23:31 – 27:39)

• Where/How to Start:
  • Don’t chase shiny AI tools; instead, audit your biggest pain points and bottlenecks, and let that drive solution selection.

  “My recommendation is always...do an assessment of where the problems are today. What are your process bottlenecks? ...Once you get a good understanding of where the top painful bottlenecks are...then start thinking about what does the solution look like. Sometimes it’s AI, sometimes it’s just a big dataset and you need something to automate it.”
  — Sowmya (23:31)

• Getting Started – Talent Needs:
  • You likely don’t need to hire AI experts for initial prototyping; start by experimenting internally with ChatGPT or similar chatbots.
  • Bring in experts/partners only when operationalizing prototypes at scale or for compliance/audit needs.

Notable Quotes & Memorable Moments

• On the pace of change at OpenAI:

  “It was the kind of job where there's no playbook, you could rinse and repeat. Nothing you thought worked in a different place would work here because this was just operating at a whole different level.”
  — Sowmya (04:21)

• On using ChatGPT to generate code:

  “It felt so silly that it didn't strike us to go ask for it to begin with, because of course, our engineers...were talking about how good it was at writing code. But then that's what kind of helped me connect the dots.”
  — Sowmya (12:01)

• On the shift AI enables for finance orgs:

  “I truly think this is one of like the big unlocks for accounting and finance teams to be able to have an engineer in their pocket...”
  — Sowmya (19:13)

• On selecting AI solutions wisely:

  “It's not AI technology looking for a problem, it's identifying the challenge and opportunity and then decide which technology might be the most applicable. And it might not be AI...”
  — Ray (25:23)

Timestamps for Important Segments

| Timestamp | Segment Description | |-------------|--------------------------------------------------------| | 01:14–03:13 | Sowmya’s career journey to OpenAI and Lumera | | 04:06–05:52 | Life inside finance at OpenAI during explosive growth | | 06:24–09:18 | Core data and forecasting challenges at scale | | 10:36–14:45 | How AI (ChatGPT) first helped automate the close | | 14:45–17:11 | QA, hallucination, and reconciliation for AI outputs | | 18:28–19:52 | Lumera’s model for AI enablement in finance teams | | 21:01–23:02 | Data privacy and enterprise considerations | | 23:31–25:23 | How to select use cases/problem areas for AI | | 26:02–27:39 | How to get started, talent requirements, next steps |

Actionable Takeaways for Finance Leaders

• Start with your bottlenecks: Audit your process pain points before chasing AI solutions.
• Prototype with chatbots: Use ChatGPT or similar to quickly test what’s possible, even for non-technical teams.
• Be pragmatic with data privacy: Stick to enterprise versions, use dummy data for coding, and run code in secure environments.
• Invest in teach-ins or workshops: Upskill your finance team to feel comfortable leveraging AI tools for real process improvements.
• When & how to bring in experts: Only partner with AI vendors or consultants when operationalizing at scale or when regulatory/audit requirements demand.

Rapid Fire Recommendations (27:39–29:53)

• Who to follow in AI: Stay connected to frontier model companies and watch technology shifts closely.
• Must-have tool: ChatGPT remains the daily driver for most AI-driven productivity in finance.
• How to learn more from Sowmya:
  • Website: lumerahq.com
  • Substack & LinkedIn for hands-on use cases
  • New podcast interviewing finance leaders about AI tech stacks

This episode demystifies the real-world use and ROI of AI in finance through firsthand stories and practical guidance—highly recommended for any finance professional considering their first or next leap into AI.