What the next generation of AI has in store for finance
3 ways finance leaders could see value, and which one I'm betting on
As we release our first LLM-powered feature at Anrok, I put together some thoughts on the future of AI in finance. While no one can forecast how AI will manifest, I’ll share the broad strokes of where the technology might be applied in finance over the next 2 to 3 years.
(For my thinking on where we are today: The fog of AI and what investors are missing)
An inherent assumption in my predictions is that we will see a step function change in LLM capability in the next release of models from OpenAI and Anthropic.
The possibilities
What are the ways in which the next generation LLM could augment a finance leader's workflow? If I were to distill the plethora of answers I’ve heard, there are three general directions this could take:
AI Product: Software companies utilize LLMs in product offerings
AI Agent: Users provide their LLM-powered AI agent with access to their software subscriptions
Custom software: Finance leaders write custom software with LLM-powered AI coding assistants
I notice people are overwhelmingly anchored to the first direction. Every CFO dinner conversation is a frantic exchange of what financial tooling one has tried, of little to no avail.
The third direction hinges on how large the step function change between models will be. Even in a superintelligent world, the principles of economics still apply. Economies of scale will still mean that most finance teams would benefit from using off-the-shelf software. Until we get to a post-scarcity world, Hal Varian’s Information Rules still provides a relevant blueprint on digital supply and demand dynamics.
My money is on the second as the primary way finance leaders see value in the near term. Now, we could still benefit from software companies incorporating LLMs into their products or see legacy software companies meaningfully disrupted by a 10x better AI-first upstart. What I’m trying to underscore here is that many of us aren’t considering the other possibilities.
The near term prediction
Imagine an intelligent agent that can log in to your NetSuite and Stripe accounts and reconcile transactions with your bank account. Like an analyst on your team, it can click into the different accounts, identify anomalies and discrepancies, and take screenshots to provide supporting evidence in your general ledger. It uses the same tools that you do today.
Similar to us, the LLM can document its work. Interestingly, when an LLM is asked to think through a problem step-by-step and show its work, the results are more accurate. AI researchers call this prompting phenomenon as “Chain of Thought” (Wei et. al, 2022).
Take this one step further. You spin up another copy of the AI agent to then check the other model’s work. It flags any areas for the manager (you) to review. This “acceleration of existing trends” future might seem less attractive than the “AI in all of our products” for many. But this is precisely why we underestimate this possibility.
To borrow a phrase from the former CTO of Microsoft:
“Many of the most powerful effects of the information highway, especially in the early days [like where I believe we are at with AI in 2024], will come by accelerating trends that are already well under way.”
—InterOffice Memo: Road Kill on the Information Highway (1993)
What intelligent agents tell us about possible losers
When I first signed with Airtable in 2018, my then-boss Howie Liu shared a reading list before I started. One of the books on the list was Engines that Move Markets by Alasdair Nairn. A quote that stayed with me is:
“When a new technology arises, it is a lot easier to spot the losers than to find the winners.”
—Engines that Move Markets
With claims of GPT-5 on the horizon, the importance of having some mental model of industries to avoid cannot be understated. Every Fortune 500 CEO (and CFO) should, at the very least, have an opinion on where NOT to deploy their capital in a superintelligent world.
If one is bullish on possibility number two—AI agents augmenting the finance workflow—the losers are obvious. Any business that relies on cheap knowledge workers is at risk of disruption. The cost of quality bookkeeping, data cleansing, and IT consulting goes way down. If Tata Consulting, Infosys, or outsourced bookkeeping services don’t start leveraging AI assistants to augment their services, they may soon find their cost basis too high to compete.
In some ways, if you aren’t already using software to automate such bookkeeping workflows, you are already at a disadvantage. Forget AI. For example, many fractional CFOs and accounting firms use Anrok’s software to automate nexus study analysis and augment their services. This is a win-win for both the firm and their client. Firms can provide faster and more accurate analysis to clients at a lower cost, and take on more clients or upsell strategic services in their free time.
There are qualities that make a business immune. Namely, businesses where prestige and authority matters. I am bullish on The Big Four and reputable accounting, audit, and advisory firms. Their prestige and authority are valuable intangible assets. Today, clients pay premiums for the legitimacy from a Deloitte, EY, KPMG or PwC seal of approval on their audits. This "reputational capital" constitutes a moat. Though new technologies may reduce their costs and provide new ways to compete for market share, the differentiation from the average IT consulting business lies in the signaling value of their brands.
The bottom line
No one really knows what the capabilities of future models may hold, but we ought to know what to avoid. The bottom line is that we need to couch predictions in existing trends and principles of economics. The tools that worked best over the last few years could still be relevant in the few years ahead. ⊞