CEO and Co-Founder of Digits, a financial management platform providing the next generation of tools that give operators a full understanding of their business in real-time. Using state-of-the-art machine learning and advanced statistical analysis , Digits converts millions of data points into a living model of your business.
You taught yourself to program when you were 12, what attracted you to computer science at such an early age, and what were you programming?
Like most late 90s kids, I spent hours playing computer games and my parents felt that wasn't the best use of time. When I was 12, they gave me a book for Christmas – Mac Programming for Dummies – and to be honest I read it and didn't really get it. A few months later I picked it back up and tried the “Hello World” example, and on a whim I changed it to make the text print in bright orange. The moment that text turned orange, a lightbulb went off in my head and I realized I could make this computer do anything I ever wanted. I was instantly hooked. I spent basically every night from then on coding, and over the course of middle school and high school I built and released a series of Mac desktop shareware apps, including a Histogram graphing application and a plugin editor for the game Escape Velocity.
You’ve started 2 companies prior to Digits, what were these companies?
In 2007, I co-founded Increo, a real-time document collaboration startup that let you comment, draw on, and mark up documents in your web browser. We were acquired by Box in 2009. Two years later, in 2011, I co-founded Crashlytics, a mobile crash reporting tool which was acquired by Twitter in 2013 and then again by Google in 2017. Today, Crashlytics is the de facto crash reporter for iOS and Android and runs on over 6 billion MAU, substantially every active smartphone on earth.
Could you share the genesis story behind Digits, and how it originated from your experience operating Crashlytics?
At Crashlytics, I was really struck by the difference in quality and ease of use between the dashboards we had on the product side (Google Analytics, real-time performance monitoring, A/B testing, etc.) and what we had on the business/finance side (QuickBooks, Excel models, PDF reports). It was crazy to me that any question seemed so tedious and manual to answer, and so slow – we were waiting 2-3 weeks after every month to get our financial reports. I started Digits with a single goal: make small-business finance real-time, interactive, and intuitive.
What is the real-time financial data premise behind Digits?
My Crashlytics experience was reinforced by what I saw at Twitter. As Head of Consumer Product, part of my responsibilities were working with the finance team to oversee the org's finances. I remember asking them a budgeting question where the answer was effectively, “We haven't run those books. Give us a few weeks.” I was like, a few weeks?! We have 100+ people in finance and this is the core product-engineering team at the company! It showed me that what I experienced at Crashlytics wasn't weird, it was the accepted status quo at tons of companies of all sizes.
In today's fast-paced world, you need to make business decisions in real-time, in the moment. Your finances should run at the same speed your business runs.
Why have legacy software companies struggled with offering real-time financial data?
Legacy is the key word here. The current financial systems are relics from the digitalization era of the 70s-80s. The vast majority of banks are still running COBOL mainframes. The major accounting software packages are 20-30 year-old codebases. At Digits, and with the launch of Digits AI, we’re addressing this at the foundational level, reimagining the basis of financial accounting through the lens of the latest machine learning technology and modern software architecture.
Can you discuss the types of machine learning algorithms that are used?
We're so excited about Digits AI because it seamlessly combines the strengths of both major fields in machine learning: generative large language models and predictive similarity models. We've fine-tuned generative language models to aid customers with tedious tasks like understanding financial questions and explaining accounting terminology, and we have trained proprietary predictive models on over $300 Billion in small-business transaction volume, so they understand the core concepts of double-entry accounting. Combined with our custom-designed financial modeling engine, Digits AI represents a technological breakthrough in the application of state-of-the-art machine learning models to business finance.
Generative AI models are often poor at math, how does Digits solve this problem?
You're exactly right – generative AI models are brilliant at creative pursuits but notoriously weak at fact-based exercises and math in particular. You might have seen some examples with ChatGPT that are pretty hilarious. Solving this was a huge effort and an absolutely massive breakthrough for our engineering team: Digits AI combines the power of large language models with our new, proprietary financial modeling engine and custom-designed query language that is both easier to translate and more expressive than traditional SQL-based approaches. This allows Digits AI to understand the user’s intent and request computations over their data without knowing the data’s underlying schema or encryption keys. Brought together, we can deliver 100% accurate responses to every request.
Digits has pioneered a three-tier architecture to protect customer data bringing bank-grade security to large language models, could you discuss what this technology is, and how it keeps data secure?
At the base lies Digits’ new, proprietary financial modeling engine. Customer data is encrypted at rest using AES-256, with object encryption keys protected via advanced techniques like per-secret envelope encryption, and is siloed business-by-business, never leaving the secure confines of Digits' US-based infrastructure. At the second tier, Digits AI leverages custom-trained, proprietary deep-learning models to understand the unique attributes of small-business finance and double-entry accounting. At the third tier, Digits AI interacts with public LLM APIs using only anonymized and obfuscated customer identifiers. At no point is raw customer data ever exposed to third-party models or systems.
You’re quite bullish on the future of AI, why do you think the AI boom is just beginning?
ChatGPT has experienced the single fastest adoption curve of any technology in the history of humanity for good reason: unlike the internet wave 25 years ago, you don’t need to get a computer and sign a contract with an ISP. Unlike the mobile wave 15 years ago, you don’t need to buy an expensive smartphone. And unlike the crypto wave 5 years ago, AI is easy to use, and it’s actually useful.
I deeply believe AI will be an incredible, accelerating force across the vast majority of industries. The underlying technology is real, it works, and it is already showing fascinating parallels to how humans learn and think. I predict in 10 or 15 years, we will look back and say the advent of AI was a greater transformation of the human experience than mobile. It’s on par with the invention of the computer chip, or flight, or the printing press, in how it’s going to impact the world around us. And It’s happening 100 times faster.
What is your vision for the future of Digits?
Digits AI is just our first step in building the future of real-time, intuitive, automated small-business finance, and we're already hard at working expanding its capabilities. In the (near) future, Digits will speed-boost your bookkeeping, give you simultaneous cash- and accrual-views of your business, auto-generate your monthly financial statements, and help you budget and forecast without the tedium of debugging and maintaining models. The next five years are going to be transformative in business finance and accounting, and our goal is to pioneer the software that makes all of this delightful.
Thank you for the great interview, readers who wish to learn more should visit Digits.
- The Black Box Problem in LLMs: Challenges and Emerging Solutions
- Alex Ratner, CEO & Co-Founder of Snorkel AI – Interview Series
- Circleboom Review: The Best AI-Powered Social Media Tool?
- Stable Video Diffusion: Latent Video Diffusion Models to Large Datasets
- Donny White, CEO & Co-Founder of Satisfi Labs – Interview Series