Model freemium to paid conversion
A free user is a cost centre until proven otherwise. Freemium is a bet that a small paid slice pays for the whole free base; compute the conversion rate at which that bet breaks even.
Procedure
Measure the per-free-user monthly cost: hosting, storage, support, and email, summed and divided by active free users. Say $0.08 per free MAU per month.
Measure the paid plan's gross profit per month: price minus the variable serving cost. $15.00 price less $1.50 cost is $13.50.
Write the break-even condition: paid_count * gross_profit >= (free_count + paid_count) * free_cost.
``` python3 -c "from decimal import Decimal as D free=D('100000'); gp=D('13.50'); fc=D('0.08') print((fc*free)/(gp-fc))" # 596 paying users to cover free-user cost -> 0.60% conversion ```Compare to the observed rate. At 3% converting you are well past break-even on servicing; the real cost is usually acquisition spend, not hosting.
Add the support tail. Heavy free users cost far more than the average; bucket by P95 usage and re-run so one abused free tier does not hide in the mean.
Cap the free tier's expensive resource (compute, seats, storage) so per-free-user cost is bounded by design, not by hope.
Keep the conversion rate an exact ratio of integer counts, and prices in minor units; a float percentage hides the point where the tier tips to a net loss.
Recompute when infrastructure cost per free user moves 25% or more, since it, not price, usually drives the sign.
Distinguish free-to-paid within a window from free-to-paid ever; the second is larger and easier to fool yourself with.
Decide the free tier's job explicitly — acquisition, virality, or a loss-leader — and measure against that job.
Pitfalls
- Dividing free-user cost by all users and understating it per free user.
- Assuming the average free user's cost when the P95 user can be a hundred times dearer.
- Counting conversions at any time rather than within a window, overstating the rate.
- Leaving an expensive free resource, such as exports or compute, uncapped.
- Treating support cost as fixed when it scales with free-user count.
- Float percentages that hide the sign flip exactly at the break-even point.
Verification
```
python3 -c "from decimal import Decimal as D; print((D('0.08')*D('100000'))/(D('13.50')-D('0.08')))"
# ~596 paying users, i.e. 0.60% of a 100k free base
```
Pass when the break-even conversion rate and the observed rate are both stated. Report the break-even rate, the observed rate, and the per-free-user cost bucket behind it.