← PM Digest
$LLM economicsJun 2026 · 4 min read · Kumar Aniket

Your token bill is a product decision

Every AI feature has a meter running, and most teams treat the reading as a finance problem — something to reconcile at month end. It isn't. It's a product decision you're making blind.

The cost of an AI product isn't the model's price per token. It's tokens-per-request × requests-per-user × users — and every one of those multipliers is something a PM chose, deliberately or by accident. The system prompt nobody trimmed. The retry loop that fires on every timeout. The autocomplete that calls a frontier model when a small one would do.

When you attribute spend down to the feature and team level, the conversations change. "AI costs are up 40%" is an argument. "Summarization retries doubled after the provider timeout change" is a fix. The first one produces a cost-cutting mandate; the second one produces a two-line PR.

The uncomfortable part: you can't delegate this. If the PM can't read the meter, the PM isn't pricing the product, the infrastructure bill is. Unit economics — cost per request, cost to serve a user, what caching is actually saving you — belong in the same dashboard as activation and retention. That's not a finance ask. That's the job now.

Built something like this? I'm always happy to compare notes.

aniket.kgp25@gmail.com →