AI in finance and the benefits for enterprises

In my view, the equivalent formula for AI will look like:

AI ROI = (business process value x volume) / AI costs

While the numerator is straightforward, a few notes on the AI costs denominator:

• Data: Cleaning and labeling data are real costs.

• Quality assurance: Checking outputs for accuracy requires real resources. This is especially important in enterprise and finance use cases.

• Setup: These include process mapping, prompt engineering, and security reviews.

• AI vendor costs: Given compute costs, I expect usage-based pricing to be the default for AI.

Winning buyers: larger orgs

Importantly, those “AI costs” are relatively fixed.

Using an existing AI use case of accounts receivable (A/R) management as an example, the all-in costs for a $400 million revenue and $100 million revenue would be relatively similar, at least relative to that 4x revenue disparity.

For that relatively similar AI cost, the $400 million org is able to automate 4x the amount A/R processes.

Moreover, that $400 million org has more absolute dollars to invest in getting their AI strategy right. Let’s assume each firm allocates 0.5% of revenue to AI (a meaningful portion of their IT budget) — $2 million dollars of spend should win out versus $0.5 million.

Lastly, that $400 million org should have a larger pool of data to build models from.

These factors imply that larger-scale AI adopters should be the most advantaged.

However, quality data and (modern) interoperable tech stacks are also critical. Abraham Thomas captures this well:

“Quantity has a quality that’s all its own, but when it comes to training data, the converse is also true. ‘Data quality scales better than data size’: above a certain corpus size, the ROI from improving quality almost always outweighs that from increasing coverage. This suggests that golden data — data of exceptional quality for a given use case — is, well, golden.”

Winning users: Many roles, daily gains = higher leverage

Generative AI will drive further adoption and improvements upon existing AI use cases — like matching receipts to transactions in Brex today — reflected in the Deloitte AI report above.

Despite any clear net new killer apps at this moment, I think the finance function captures how AI will diffuse broadly into personal workflows with “human in the loop” as the default.

Like the subtle gains created by Calendly lowering scheduling friction, some examples:

• Broaden surface area of data analysis — “interesting to look at” sensitivity (recent example here) and correlation work that would be impractical for a human analyst could be first attempted by an AI bot. I predict the first wave here will focus on “chat with your data” use cases (see Brex’s “OpenAI For Finance Teams” product announcement).

• Automate more forms of internal reporting like weekly slide decks.

• Small gains from “bots in the background” will compound.

And those time savings will be redeployed into more strategic work instead of layoffs. Said differently, each worker will become higher leverage.

These AI-driven productivity gains will be especially important in teams that scale in size with transaction or customer volume, like:

• Customer service

• SDRs

• And many other volume-linked operations functions, including many finance “back office” processes

Winning users: examples of AI benefits

To frame the gains AI automation can produce, Blackline (a publicly traded financial close software company) estimates their automated transaction matching reduces manual efforts by 70%.

Similar automation — like receipt matching — is also at the core of Brex’s product suite and a prime example of shifting employees' focus from the “repetitive and routine” to “outliers and strategic.”

Here’s how Brex envisions some of the AI benefits for specific users:

For employees

• Automatically fills in documentation requirements, including receipts and memos

• Automatically routes spend to the correct budget so employees don’t need to manually assign their budgets each time

• Answers finance-related questions

For managers

• Highlights anomalous or high-risk transactions so managers can focus on outlier expenses for review

• Automatic follow-up with employees on missing documentation or policy violations

For travel managers

• Assist in planning group offsites

• Automatically book travel according to your preferences

For finance teams

• Gain insights into company spend among different employees, departments, or even compared to other companies

• Close the books faster with automatic categorization

• Provides real-time insights based on spend trends

• Highlights expenses that need a closer look

• Automatically follow up with employees on missing documentation or policy violations

• Easily build expense policies based on different risk levels/similar customers

• Automate procurement workflows

Listen: Lightcast controller Ron Cook says AI will help most in these areas

See @OnlyCFO for even more on Brex + AI

Winning vendors: specialists + incumbents

Emergence Capital — an enterprise-focused VC firm — has been publishing excellent content on AI that I agree with, especially the value of specialization:

“But in business settings, similar to all cloud software, AI’s value will be most powerful when tightly focused.” — Gordon Ritter, Emergence Capital

That said, I believe that incumbent software vendors are also advantaged due to:

• Low barriers to adding AI: As shown by the explosion of AI product releases — see Ian Ito’s B2B SaaS Generative AI Tracker — and the existence of AI plugins in a short period of time.

• Customer acquisition costs + capital: Achieving scale in SaaS was already slow and capital intensive — Key Banc’s flagship SaaS benchmarks show reaching $25 million of ARR requires over 7 years and $31 million of capital. Due to higher compute costs, AI startups will likely have lower gross margins than traditional SaaS models and therefore need even more capital to hit scale. Whereas incumbents only need to roll out — at negligible cost — their new AI modules to their existing customers.

• Trust + security: Not only do incumbents have a security track record with their customers, but they are also financially aligned. A ServiceNow or Brex is less likely to take risks with a marquee, high ARR customer than a potentially desperate AI upstart might take.

Predictions: business + P&L impact

In the wave of hype and excitement over AI, framing the business and P&L impact of AI is a useful exercise.

Goldman Sachs research forecasts AI lifting economic productivity by 1.5% per year, raising global GDP by 7%, and corporate profits by 30% over a 10-year horizon. Central to this forecast is AI taking the place of roughly 25% of human workloads.

While not fully comparable, my own estimates of AI’s P&L impact are more conservative than Goldman Sachs’ due to:

• Degree of human replacement: While the recent pace of innovation in AI has been staggering, I have more skepticism regarding full human labor replacement.

• P&L flowthrough: Some AI savings will be shared with customers, AI vendors, and spent internally.

• Existing non-AI optimizations: Businesses have not been asleep — back office and supply chain automation programs have been going on for decades, leading to less “low-hanging fruit” and margins for AI to fix.

Implied AI ROI: Goldman’s 30% corporate-wide profit lift implies returns on AI investment far beyond traditional software ROIs.

My poll (n = 154) on AI’s profitability impact yielded similar results with EBITDA margin gains below Goldman Sachs estimates.

However, either forecast suggests AI is worthy of the hype and adjusting corporate priorities.