Canva's paid plans all include AI.

Here is what you actually get for the money.

Canva's published plan allowances, as of September 2026. The same monthly allowance buys either Premium or Ultra uses, at roughly ten to one.

Look at the Pro line.

200 Premium uses, or 20 Ultra. Same allowance, different denominator.

Canva did not change what it charges for. It still charges for AI uses.

What it changed is the definition of a use.

Plain version first: not every unit of your metric is worth the same to the customer.

Some units are worth a lot. Some are worth nothing at all.

The word for how evenly that value is spread is metric density.

Stripe is the perfect case. It charges a percentage of the money moved. One dollar is identical to the next dollar, and each one is worth the same to the customer.

Density 1.0. You cannot do better than that.

Now the opposite.

Lyfegen is a Swiss insurtech that handles repayment flows between insurers and pharma companies. They priced per patient.

The average patient was worth about $20 to the insurer. So charge $10 a patient and split it down the middle, right?

No.

A few patients on expensive experimental treatments were worth $1 million or more. The vast majority were worth exactly $0, and nobody could tell which was which until the platform had already processed them.

So the insurer looks at $10 per patient and sees itself overpaying on 95% of its patients.

‍

Here is the part that matters.

You argue fairness at the metric level. Your customer argues it at the unit level.

You say the average patient is worth $20.

They say three quarters of these patients are worth nothing.

Deadlock.

And the customer responds in two ways, every single time.

First, they push hard on price. Your $10 becomes $2.

Second, once the deal is signed, they sort. They stop sending you the units they believe are worthless, so you only ever process flow they have already creamed.

Everyone loses. Adoption stalls, you never see full volume, and your customer is now running a pre-sorting process just to feel fairly treated.

‍

You can put a number on this.

Median value of a unit, divided by average value, multiplied by the share of units at or above the median.

Stripe: $1 divided by $1, times 100%, equals 1.0.

The rule of thumb underneath the formula is simpler. Your median unit should be worth roughly what your average unit is worth.

Where you land:

• 0 to 0.25: poor. Heavy discounting, stunted adoption. Fine for testing demand, not for running a business on.
• 0.25 to 0.50: sub-par. Long negotiations. Survivable if your packaging is strong, but a better metric would unlock real money.
• 0.50 to 0.75: strong. Minor issues only. Stick with it unless something obviously better is available.
• 0.75 to 1.0: excellent. Sales and adoption go smoothly, and it will survive packaging that works against you.

If you are sitting at the low end, there are two moves. Only two.

Pick a different metric altogether.

Or keep the metric and tighten its definition until the units look more alike.

Slack is the clean example of the second one. It does not price per user. It prices per ACTIVE user.

Someone who has a login and never opens Slack is worth nothing to the buyer. An active one is worth everything. Same metric, narrower definition, far higher density.

The version I use in the book: a customer has high demand for M&M's, but values the red ones three times as much as the brown ones. Price per #M&M and you capture that demand at low density. Refine to #RedM&M and you are in business.

Which brings us back to Canva.

“One AI use” is a low-density metric. An image upscale and a generated video both count as one, and they are worth nowhere near the same to the person doing them.

So Canva refined the definition. Standard, Premium, Ultra, with an Ultra generation drawing ten times what a Premium one draws.

That is the Slack move, applied to credits.

With one wrinkle worth noticing.

Canva looks to have weighted those tiers by what each generation costs it to run, not by what each one is worth to the customer.

Those two things point the same way often enough to be useful. They are not the same thing.

Density is a statement about how value is distributed for the customer. Weighting by your own compute bill raises it only insofar as your costs happen to track their value.

Which is last month's cost-versus-value problem wearing new clothes.

‍

Two things density is not.

It is not value chain position. Density governs how much of the value you manage to monetize. Where the unit sits in the value chain governs whether there is demand for it at all. High demand for a unit nobody will pay much for is still a bad metric.

It is not packaging. Canva's $100 AI Pass, and the fact that allowances never pool across a team, are packaging and fencing decisions. Both are smart, and strong packaging can carry a metric with mediocre density a long way. Neither raises density by a single point. Different lever, different job.

How to check your own:

1. Pull the distribution, not the average. For your current metric, estimate what one unit is worth to the customer across the whole population. What shape does it make?
2. Compare median to average. Divide the median by the average, then multiply by the share of units at or above the median. Where does that put you on the scale above?
3. Find the units worth nothing. What share of your metric's units deliver no value to the customer at all? That is the share they will argue about, and eventually stop sending you.
4. Check whether they are already sorting. Is volume per customer lower than the contract implied? Which customers are filtering what reaches you?
5. Refine the definition before you replace the metric. Which sub-set of your current unit carries most of the value, and can you define and measure that sub-set? That is the ACTIVE user move, and it is far cheaper than re-metering the whole model.
6. Only then go looking for a different metric. If no definition of this unit gets you past roughly 0.5, what would you have to measure instead?

Choosing the right metric is the decision everyone talks about.

Whether the units inside it are worth the same is what decides how much you get paid.

Most pricing projects do the first and never check the second.

One Idea

Picsart just swapped the unit under its video quota

Seedance 2.5 used to be sold in videos. Pro got about 83 of them a month.

Now it is sold in seconds. Pro gets about 250 seconds, inside the same credit allowance. Ultra went from roughly 416 videos to 1,250 seconds.

The price did not move at all.

So did customers get more, or less?

Depends entirely on how long their videos were.

Which is the tell. “One video” was never a stable unit of value. A three second test clip and a thirty second finished piece both cost the customer one video, and they are worth nothing like the same to the person making them.

That is a low-density metric, and it produces exactly the behaviour you would predict. People ration. They avoid short experiments because a throwaway test burns a whole unit. They feel robbed on every generation that does not come out usable.

A second is uniform. Median and average sit almost on top of each other. Nobody has to decide whether this particular render is worth spending a unit on.

Picsart did not change what it sells. It changed the definition of the unit until the units were worth the same.

Same move as the active user. Same move as Canva's tiers.

Take your own main metric and ask the blunt version of the question: is the median unit worth roughly what the average unit is worth? If it is not, you will spend the next two years discounting and wondering why volume came in under what the contract implied.

Fix the unit before you touch the price.

‍