Contrary to a16z, AI won't fix your finance

Infinia ยท

Recently, a16z published "You Need a New CFO." Two numbers in it are going to change a lot of founder conversations.

Anthropic's finance team gets a near-complete monthly review done in about 30 minutes. And leading companies are aiming for finance at 2% of headcount, down from 5%.

Both numbers are true. But, a lot of founders at $1M to $5M ARR are about to take exactly the wrong lesson from them.

The lesson they'll take is: "AI can do finance now. I don't need a finance person."

The lesson they should take is: "AI can do finance now, if someone has already done the hard part."

The 30 minutes is the end of the story, not the start

A 30-minute monthly review doesn't happen because someone plugged a model into a spreadsheet.

It happens because a team spent a long time on the boring work that comes first:

AI is reading clean data. It isn't fixing messy data. It's reading it confidently.

And that's the problem.

Fast and wrong is worse than slow and wrong

Here's what happens at a typical $2M ARR company that "adds AI to finance."

The founder connects the accounting system, the billing tool and the bank to an AI assistant. They ask it for net revenue retention, churn by cohort and runway.

It answers in seconds. The answers look polished. They come with charts.

But churn is defined three different ways across the billing tool, the CRM and the board deck. Two enterprise customers are on annual contracts the billing system treats as monthly. One expense category has been miscoded since last spring.

The AI doesn't know any of that. It produces a beautiful, confident, wrong answer.

When a bookkeeper gives you a wrong number slowly, someone usually questions it, because it took a week and it looks a little off.

When AI gives you a wrong number instantly and formats it nicely, nobody questions it. It goes straight into the board deck.

That's wrong faster.

The part nobody can automate

Before you bring AI into finance, three things have to be true, and no tool will do them for you.

Your numbers trace to their source. Every figure in the deck can be followed back to an invoice, a contract or a bank line. If it can't, the AI is guessing, and so are you.

Your definitions are written down. What counts as churn. When revenue is recognised. What an "active customer" is. If three people would answer differently, the AI will pick one at random and never tell you which.

Someone owns the process. Not the tool. A person who decides how the month closes, checks what the AI produces, and is accountable when it's wrong.

That third one is where founders will push back hardest.

"But the whole point is fewer finance people"

Yes. At scale, it is.

The a16z argument is about companies that already have a finance function and can now run it with fewer people. Going from 5% of headcount to 2% assumes you started with 5%.

At $2M ARR, most companies started at zero. There's a founder, a bookkeeper, and a spreadsheet the founder rebuilds before every board meeting.

You can't shrink a finance team you never built. And you can't automate a process that doesn't exist yet.

So for most early-stage companies, AI doesn't remove the need for senior finance judgment. It makes the need more urgent, because the tools are now powerful enough to scale whatever mess you feed them.

The people who'll disagree

Some will say I'm defending an old job against a new technology. I'm not. AI in finance is real, it's coming fast, and founders who ignore it will fall behind.

Some will say a smart founder with good tools can do this alone. Some can. If your data is clean, your definitions are written down and you have the time, go ahead. Most founders at this stage have none of the three.

And some will say this is what someone who introduces fractional CFOs would say. Fair. Judge the argument, not the messenger.

The right order

If you have a raise or a board meeting coming, here's the sequence I'd follow.

  1. Fix the data. Make every number in the deck traceable to its source.
  2. Write the definitions. One page. Churn, ARR, active customer, revenue recognition.
  3. Give someone the process. One person who owns the close and the numbers, even part-time.
  4. Then bring in AI. Now it makes good numbers fast instead of bad numbers fast.

Do it in that order and you might get your own 30-minute monthly review one day.

Do it backwards and you'll get a very fast, very confident path to an awkward board meeting.

AI is only as good as the data it runs on. Somebody still has to own the data.

Infinia introduces founders to fractional CFOs who own a number, not a calendar. One introduction, free.