Your Financial Statements Assume a Human Customer. That's a Problem Now.

If AI agents are choosing your product — not a human buyer with loyalty and memory — your standard financial statements are already lying to you a little. Here’s what breaks, and the four-part framework for fixing it.

The Framework

Every core metric investors use to judge revenue quality assumes a human on the other end. Customer lifetime value and acquisition cost collapse when the “customer” has no loyalty or memory to build. Churn analysis can’t predict agent-driven shifts because agents don’t behave like retained accounts — they behave like whatever the underlying model was last trained to prefer. Stress testing fails the same way: historical recession performance and sensitivity models can’t account for a customer base whose preferences move with model retraining cycles instead of economic indicators. The mechanism is specific — when a foundational model retrains, its default selections shift. A coding agent choosing an authentication provider, or a procurement agent picking a supplier, can flip overnight because updated training data changed what the model defaults to, not because the market changed. As Neel Somani puts it: “The accounting hasn’t caught up to the reality of who’s actually buying.” Authentication providers and payment processors are the categories most exposed right now.

Segment Revenue by Source

The first fix is disclosure-level: break out revenue as human-driven, agent-driven, or hybrid, rather than reporting a single blended growth number. A company reporting 40% revenue growth means something very different if that growth is concentrated in agent-driven accounts with no switching cost keeping them in place.

Disclose Foundational Model Concentration Risk

If a meaningful share of revenue flows through decisions made by one or two foundational models, that’s a concentration risk analogous to customer concentration — and it deserves the same investor-facing disclosure treatment, since a single model retrain can move the number.

Establish an “Agent Penetration Rate” Metric

A new metric — the share of revenue or transactions initiated by an AI agent rather than a human — gives investors a way to see the exposure directly instead of inferring it from footnotes, and gives operators a number to actually track over time.

Where It Doesn’t Apply

None of this matters yet for a business with negligible agent-initiated revenue today. But the exposure grows quietly — the first sign is usually a growth number that looks great and a churn model that stops predicting anything, which is exactly the blind spot AI-native finance functions are starting to build tooling around.

Source: Entrepreneur

Content created with AI assistance and reviewed for accuracy.