• CFO Secrets
  • Posts
  • šŸ¤ One-on-one Interview Series: Babak Siavoshy, CFO of Anduril

šŸ¤ One-on-one Interview Series: Babak Siavoshy, CFO of Anduril

How is one of the fastest growing companies on Earth using AI?

Header

A word
Ramp

Over $1B is processed on Ramp cards each month, revealing clear market trends.

AI is a fast-growing line item in companies' budgets, and Ramp data shows that AI spending is moving out of the ā€œexperimentalā€ phase and into an ā€œoperationalā€ one.

Download the report to discover what AI tools companies have been bullish on, what else they're spending on, and where they're cutting back.

One-on-one withā€¦

Iā€™m on a mission to better understand what AI means for CFOs.

That started last week with an interview with Eric Glyman, the founder of Ramp.

But we need to hear from those at the coal face too. Time to talk to some fellow CFOs.

Babak Siavoshy is CFO at Anduril, the defense technology company. Babak recently closed a funding round valuing Anduril at a reported $14bn, making it one of the fastest-growing businesses on the planet.

Babak has an interesting take on AI. Heā€™s selective about how and where he uses it. Heā€™s also had an interesting path to CFO with a valuable lesson for any budding finance pro.

Letā€™s get into it.

How AI will change accounting and financeā€¦

Secret CFO (SCFO): Do you think the AI hype is justified?

Babak Siavoshy: The state of AI in 2024 is similar to the state of the internet in the 90s. It is a technology that will undoubtedly change the world, so it merits the hype itā€™s getting. But many of the current applications are probably not going to survive the decade. Whether or how much of an AI bubble we are in is debatable, but thereā€™s definitely a lot of froth out there.

Longer term, Iā€™m very bullish. Some people have become desensitized to the magnitude of the progress weā€™ve seen in a short time. Just two years ago, virtually nobody believed we would have chatbots that easily pass the Turing test. And weā€™ve seen record-breaking revenue growth at multiple companies, suggesting thereā€™s more than just valuation bloat. How sticky that revenue is, how much of it is infrastructure, etc., remains to be seen, but I think weā€™re past the point of saying itā€™s fake.

The question now is, ā€œWhat are the actual long-term applications going to be in the business world?ā€ I think that will be driven by what the tools are good at. One observation is that the current language-based AI tools are not precise. They hallucinate and make up facts. So they will be less useful where you need high degrees of precision, at least in the near term.

But there are many use cases today where the tool doesnā€™t have to be very precise, so long as itā€™s making a human way more efficient. I think of these as 80% use cases ā€“ basically where being 80% right but 10x faster is still a really good outcome, assuming a human can take over and fix the last 20%. In these 80% of applications, the AIā€™s work doesnā€™t have to be perfect, but it can still add value if it saves a ton of time and increases productivity. Many of the early applications of AI ā€“ in customer service, coding, and design ā€“ have in common that they donā€™t need to be perfect to be useful.

SCFO: How is AI changing finance and accounting?

Babak: I think there are promising use cases emerging, but itā€™s useful to talk through the challenges first.

The main challenge with finance use cases is that current AI language models donā€™t promise precision ā€“ in fact, they promise that some percentage of their interactions will be hallucinations. And in CFO-land, precision matters. If you had an analyst who said, ā€œHereā€™s everything with a big asterisk, I may have hallucinated 5% of the data,ā€ youā€™d probably discount that personā€™s work significantly.

The other challenge is that a lot of the finance role ends up being about people, and about things like trust, accountability, and credibility. These human dynamics determine the quality of your inputs ā€“ whether people in the business give you timely and context-rich information ā€“ and the quality of your outputs, including whether those same people trust your analysis and act on it.

You canā€™t delegate credibility and accountability to AI, and if the models are not precise, then they risk eroding the trust the finance function has built within the organization. You need to have a person ā€“ and not a black box ā€“ accountable for the fidelity of the data at the end of the day. AI can be helpful in some ways, but itā€™s not going to democratize accountability.

On the positive side, Iā€™m bullish about use cases where AI tools are helping make a finance or accounting professional more efficient, for example by sifting through a large corpus of materials and drawing some initial insights, or putting together a draft of an analysis with citations to source information. Iā€™m particularly interested in cases where the AI model is able to show how it arrived at its conclusions so that an accountable human can check its work.

SCFO: So, you think the unlock is where AI can produce an answer alongside an audit trail all the way back to the source - then build confidence over time through manual checking, and eventually automating the process?

Babak: Exactly. You donā€™t even need to automate it completely. Having an analyst that works 100 times faster than a person but is 80% accurate can be very useful, but only if you have a way to check its work and get to 100% in a later step. And fundamentally in these use cases the common theme is youā€™re not replacing your best employees, youā€™re making your best employees more useful and productive and efficient.

In these cases, AI can help a smaller team manage a higher volume of tasks, pushing the point where a human needs to intervene further downstream. This allows for a more strategic and engaged finance function.

Thatā€™s similar to some of the AI and automation use cases that Anduril works on as a company. We often focus on areas where using a computer to get to 80% or 90% accuracy very efficiently is hugely beneficial, even if computers cannot yet automate 100% of the task ā€“ because the computer is making a person more efficient and allowing them to exercise their judgment and close the gap.

SCFO: How exactly is Anduril using AI in finance today?

Babak: We do not - yet - have many direct applications of AI in the finance team - for the reasons I stated. So, today, most of Anduril's use of AI in finance relies on vendors powered by AI (Ramp is one). But I do see more widespread adoption across the finance function coming quickly as new tools are launched that include the ability to ā€˜show their workā€™.

SCFO: What do you think CFOs should be doing to begin engaging with AI? And avoid being left behind.

Babak: Again, I think businesses should treat the growth in AI as similar to the rise of the Internet in the late 90s. It will be transformative for many industries ā€“ but there are probably going to be false starts along the way. Itā€™s just as risky to be a false start as it is to ignore its transformative effect on your industry. You donā€™t want to be pets.com. You want to be amazon.com.

What about AI and fraud risk?

SCFO: Given the nature of your business, Iā€™m fascinated to know how you think about the role of AI in fraud and cybersecurityā€”both in terms of the risks it creates and the tools it provides to defend against those risks.

Babak: Iā€™m not an expert in cybersecurity, but one way to describe what changed between 2021 and 2022 with the big ChatGPT release is that you now have a machine that can pretty convincingly sound like a human. It seems reasonable to assume this technology will make it dramatically easier to engage in social engineering attacks.

Social engineering is effective because no matter how tight your security, thereā€™s always a human involved who has the authority to bypass it. Now, you have a robot that sounds like a human and can charm or trick that person into giving up information that gives the fraudster an advantage. And you have the ability to deploy this robot effectively for free, in any language you want, against many victims. This was unthinkable a few years ago.

I think weā€™ll see this impact on consumers before enterprises. Nigerian Prince scams work even though they seem silly because if you reach out to enough people, you will find a few who still fall for them. With AI, fraudsters can go deeper into conversations without human involvement, managing a thousand conversations instead of just a few. This allows them to cast a wider net and get more leverage. These, again, are 80% use cases where precision matters less than speed and volume.

On the positive side, thereā€™s nothing unique about a fraudster. If AI is useful to them, itā€™s also useful to legitimate businesses and useful to security teams fighting back against fraudsters and cyber actors. So, the flip side is that AI can help in areas like sales, marketing, or even national security by making existing workflows more efficient and freeing up human resources for more strategic tasks.

What is Anduril?

SCFO: Can you help someone unfamiliar with Anduril understand the business?

Babak: Anduril is a defense technology company with a mission to transform US and allied military capabilities with advanced technology. Warfare - today and in the future - is increasingly defined by large numbers of intelligent, autonomous, and software-defined systems, requiring an approach to capability development and adoption more closely aligned with the commercial market than the traditional defense industrial base. Rapid development and adoption of these critical technologies are fundamental to national security.

SCFO: Anduril is clearly looking at defense in an innovative way versus the incumbents. But how specifically is Anduril unique?

Babak: Iā€™ll highlight three things that set us apart.

First, software is at the core of everything we do. Andurilā€™s products are built on top of Lattice, a software platform that uses AI among other technologies to handle everything from command and control to autonomy on all of our products and many third-party systems. While our products span all domains ā€“ from autonomous fighter jets to autonomous submarines, to space-based systems ā€“ they all share the Lattice software platform in common.

Now, I know that in the commercial space, the idea that a software-focused company would drive innovation is not novel. Software ate the world more than a decade ago, and thereā€™s a reason Apple and not Motorola invented the iPhone ā€“ it had to be a company with software in its DNA. But you have to remember that the defense and aerospace world is still a couple of decades behind the commercial world. Our largest defense contractors are still traditional shipbuilders and aircraft makers, and theyā€™ve struggled to adapt to new technology.

Second, Anduril is built from the ground up to build products at scale, and affordably ā€“ what the Department of Defense calls affordable mass. Whether itā€™s the latest fighter, bomber, or submarine, the current defense industrial base has struggled to build anything at scale in recent decades, and Anduril is built from the ground up to solve that problem.

And third, a new business approach is critical to moving quickly: in most cases, Anduril uses private capital to take on the costs of research and development before selling to the U.S. government. We try to understand a problem or operational need and design our products to solve that mission. This means we take on the risk, not the customer. The result is that weā€™re able to iterate and innovate far faster than legacy defense companies and traditional government contract timelines and save the government and taxpayers money.

Letā€™s learn more about Babakā€¦

Babak photo

Source: Anduril

SCFO: I noticed from your background that your path into finance was through the legal profession. Is that right?

Babak: Yes, I started at Anduril as employee number 20. I was the first lawyer at Anduril, and either founded or helped grow a number of different functions at Anduril, including legal, proposals, contracts, M&A, pricing, compliance, and financing ā€“ eventually running several of these teams as the Chief Legal Officer. As a CLO, I became more heavily involved in the financial side of business, working closely with the strategic finance team as we matured our M&A, pricing, contracts, investor relations, and fundraising functions. About two years ago, I took on the full finance role.

SCFO: How did all of that prepare you for your role as CFO?

Babak: It was less about my legal background ā€“ though understanding how our contracts work is incredibly helpful! ā€“ and more that Iā€™ve gained a deep understanding of our business that helps me be an effective CFO.

Iā€™ve been here from the beginning, doing everythingā€”from giving product demos to writing proposals and working on contracts, and meeting customers. That depth and variety of experience make me better able to explain the root cause of financial results and provide practical solutions. Itā€™s not just about showing numbers, itā€™s about understanding the shared language and the literal things happening in the business. For a high-growth startup, having someone deeply embedded in the business as the finance lead is incredibly important. This is the advice I often give to other startups: put a special weight on how deeply the finance lead understands the business.

SCFO: Absolutely. Numbers are the easy part of finance. Itā€™s everything else that plugs into it that matters, and that requires a deep understanding of the business.

Babak: Itā€™s also the bear case for AI in financeā€”it doesnā€™t inherently have that deep understanding. A lot of finance is about personal relationships and moving the business, which breaks down into a finite number of interactions. AI canā€™t change that. It might equip you with better data, but the human element is still critical.

Summarizingā€¦

Babak makes some crucial points. The most strategic areas of finance have two things in common; they are judgmental - not absolute. And they require precision.

Take pricing/margin analysis for new proposals for example. They are art, as much as science. But still need absolute precision in thinking and output.

Todayā€™s AI tools do not sufficiently ā€˜show their workā€™ to be relied on for those strategic tasks. The stakes are simply too high.

But they can be used to support the people doing that work. Pulling the factual data that sits underneath those judgments.

And it wonā€™t be long before products are available that can show that audit trail from source information/reasoning, all the way through to insight. And thatā€™s exciting.

The winners will be the ones that lean in and learn the tools. Itā€™s always been true, but now more than ever. Iā€™ve got my work cut out.

A huge thank you to Babak and the Anduril team for making time for the interview. And thanks to Ramp for making this series possible.

Weā€™ll be back next week for the second of four ā€˜One-on-oneā€™ interviews diving deep into the future of AI in finance. 

And, as always, you can expect your regularly scheduled email on Saturday, with a new series starting on investor relations.

Learn

Reply

or to participate.