My free users are the biggest line on the infrastructure bill, how do people size a free tier?

Start with the number you have never worked out: what does a free account cost you per month, on average and at the top end?

Take the bill, attribute it, and produce two figures. The average tells you what the tier costs in total. The top end tells you about the accounts eating your margin, and there is usually a very long tail: a handful of accounts consuming more than everybody else combined, which matches what you are seeing.

Then decide the thing you skipped: what is the free tier's budget? Pick a number, monthly, that you are willing to spend on acquisition. Free users are a marketing cost and treating them as one makes every subsequent decision straightforward.

With those two numbers, the limits follow arithmetically. Budget divided by cost per user tells you how many free accounts you can carry, and that tells you where the caps need to be.

Also worth measuring: what fraction of paying customers came through the free tier? If it is most of them, the tier is working and it is worth paying for. If almost all your customers arrived and paid without using it, you are funding something that is not converting, and that changes the answer entirely.

Most people have never checked, and it is the number that decides whether to fix the tier or shrink it.

On where to put the limit: cap the specific thing that costs you money, not everything.

The common mistake is a blanket restriction - a low usage cap on the whole product: which makes the free tier useless for evaluating and stops it converting anybody. Better to be generous everywhere except the one or two operations that carry real cost.

So look at your bill and find what actually drives it. It is usually one of: storage that never gets deleted, an expensive third party call, or compute on one heavy feature. Then:

Limit that specific thing. A storage cap, a monthly quota on the expensive operation, a smaller retention window.

Retention is usually the biggest and gentlest lever. Free accounts keep data for thirty or ninety days rather than forever. This costs almost nothing in perceived value and it stops the bill growing with every account you have ever had, which is your real problem - dormant accounts still storing things.

Delete genuinely dormant accounts, with warning emails first. A meaningful share of your two thousand have not logged in for a year.

Make the paid tier the answer to the limit, so hitting it is a conversion moment rather than a wall.

Two thousand free and forty paying is a 2% conversion, which is worth saying out loud before touching the tier at all.

Thirty days notice is the bit people skip and it is the only part that is not negotiable.

On the accounts already past reasonable, since changing terms on existing users is the awkward part.

What works:

Announce it properly and give real notice. Thirty days minimum, by email, stating exactly what is changing and what it means for them specifically - including their current usage against the new limit. Generic announcements get ignored; a message saying you are using this much and the new limit is that gets read.

Grandfather the small ones, contact the large ones individually. The handful using more than your biggest customer are worth a personal email. Some of them will pay, some are using it for something you would want to know about, and some will leave, which is fine.

Offer a path. The limit plus a paid tier that solves it is a business decision for them. A limit with no option is just a takeaway.

Do not do it silently. Quietly enforcing a new cap on people who built something on the old terms is how you get a bad reputation for very little money.

And write the policy down properly this time, in the terms and on the pricing page, including that limits may change with notice. The reason this is uncomfortable now is that the original tier had no stated bounds, and you are correcting that rather than taking something away.