The Finance AI Problem Isn't Access. It's Implementation.

In March, Mastercard launched a product called Virtual CFO. It runs on 175 billion network transactions, detects cash flow risk, flags payment anomalies, and optimizes supplier timing through your existing bank software. No new login. No implementation project. You already have it if your account is connected.

This kind of product will keep coming. The data infrastructure exists. The models are capable. Access to AI-assisted finance is no longer the constraint for a $5M company.

What I keep seeing in practice: companies that have the tools and can't make them work.

The implementation gap

A Gartner survey of 100 CFOs published this month found only 36% feel confident they can actually deliver on their AI commitments. These are companies with full finance departments. The constraint isn't the technology. It's having someone who can configure it correctly, maintain it as the business changes, and interpret what it flags.

AP automation trained on last year's vendor list miscategorizes new suppliers. Cash flow models don't know about the seasonal contract you signed in January. Anomaly detection thresholds that weren't calibrated to your actual transaction volume flag everything at first, get ignored, then flag nothing because someone widened the parameters to stop the noise.

None of this is a software failure. It's a maintenance and judgment problem. For a $5M company without a dedicated finance resource, it's enough to make the tool useless within 60 days of purchase.

What the tool can't tell you

Every AI finance product does the same thing: it tells you what the data looks like. Cash position in week 9 will drop below $200K. This vendor payment looks anomalous. Receivables are aging faster than 90 days ago.

What comes next requires context the tool doesn't have. Do you pull forward a customer payment? Draw on your line? Hold the hire you planned for Q2? That answer depends on your pipeline, your lender covenants, your read on the customer relationship, and your judgment about what the next quarter actually looks like.

No transaction history captures that. Pattern recognition is backward-looking by definition.

The companies getting real value from AI in finance use it to surface the right questions at the right time. Answering those questions still requires a person who knows the business.

Before you sign up for another tool

Gartner's numbers skew large-company, but the direction is the same regardless of your size. When CFOs with full teams underestimate the implementation burden, the gap is wider for a founder doing finance work on weekends.

Before you buy anything, ask: who configures this for our specific chart of accounts? Who reviews the output each week? When it flags a problem, who decides what to do about it?

If those questions don't have names attached to them, the tool will sit unused within a quarter. Mastercard can build the model. They can't build the operator behind it.