Regional Pricing for an AI Product: The Margin Math

Regional pricing for an AI product means charging different prices in different countries, usually a discount for markets with lower purchasing power. The standard SaaS playbook says do it: a seat costs you nothing to serve, so a $9 sale in Brazil beats a $0 non-sale. That logic does not survive contact with an AI product, because your cost of goods is not zero, it is denominated in US dollars, and it does not shrink when your price does.

Here is the arithmetic, and then the three structures that actually work.

Why the SaaS playbook breaks

Take a product at $40 a month with a genuinely typical AI cost structure. Say the average active user burns 1.8 million input tokens and 400,000 output tokens a month across a mid-tier model, landing around $9 in API spend. Add $2 for storage, egress, and the rest of the infrastructure. Your COGS is roughly $11, your gross margin is 72%, and everyone is happy.

Now apply a 60% purchasing-power discount for a market where $40 is unrealistic. The price drops to $16. Your COGS does not move. It is still $11, because tokens are priced in dollars regardless of where the user sits.

A 31% gross margin is not a growth business. And that is the average user. The heavy user in the discounted market, the one who found your product because it solves a real problem for them, might burn $18 in tokens against a $16 price. You are now paying for the privilege of serving your most engaged customers.

This is the part traditional SaaS advice cannot help with, because traditional SaaS has a marginal cost near zero and AI does not. If you have not modelled your own numbers yet, reducing AI API costs is the prerequisite to any pricing decision, and it sits inside the wider set of AI monetization strategies that regional pricing is only one branch of.

The three structures that hold up

1. Discount the seat, not the usage

Split your price into an access fee and a usage allowance. Discount the access fee regionally by whatever your market research supports. Keep the usage allowance identical everywhere, priced at a consistent multiple of your token cost.

At $40, that might be $22 access plus 1.5 million tokens included. In the discounted market it becomes $9 access plus the same 1.5 million tokens. Your margin compresses on the part that costs you nothing and holds on the part that does.

This is the structure most AI companies converge on eventually, and arriving at it deliberately saves a repricing later. It maps naturally onto usage-based versus flat-rate pricing, with the regional dial applied only to the flat component.

2. Route discounted tiers to cheaper models

If a market cannot support your price, it can often support your product on a smaller model. Serve the discounted tier with a model that costs a quarter as much per token and be transparent that the tier uses a faster, lighter model.

This is honest, it is common, and it works when your task tolerates the quality drop. It fails badly when it does not, and the failure is invisible to you and obvious to the customer. Test the smaller model on your actual workload before you build a pricing tier on it, not after.

3. Regional annual pricing only

Offer the regional discount exclusively on annual plans, paid upfront. This does two things: it converts the discount into cash you hold, and it filters for users who are committed enough that their usage patterns are predictable. Monthly regional pricing attracts the churn-heavy end of a market where your margin is already thin.

If you do not have an annual plan yet, offering an annual plan for an AI product is worth doing before layering regions on top of it.

The floor calculation

Before you set any regional price, work out the number below which you will not go. It is not complicated:

1. Take your 90th-percentile monthly token cost per active user, not the mean. The mean hides the users who will disproportionately choose a discounted tier.

2. Add your fixed per-user infrastructure cost.

3. Multiply by 1.6. That is your floor, and the 1.6 covers payment processing, support, and the fact that your token costs will rise before they fall if usage grows.

In the worked example above, if the 90th percentile is $17 in tokens plus $2 of infrastructure, the floor is about $30. Which tells you plainly that a $16 regional price for that product is not viable, and the conversation should move to a smaller-model tier or a lower usage allowance rather than a straight discount.

Doing this calculation first is what separates regional pricing as a growth strategy from regional pricing as a slow leak. It also gives you a defensible number when a customer or a reseller pushes back on the discount you did not offer them.

Enforcement, briefly

Regional pricing invites arbitrage. Someone in a high-price market will find the low price. Your options, in ascending order of friction:

• Do nothing and accept the leakage. Reasonable if the discount is under 30%.

• Require a billing address and payment method registered in the region. Catches most casual arbitrage, annoys legitimate travellers and expats.

• Require both of the above plus consistent IP geolocation. Catches nearly everything and generates support tickets from people with VPNs, which is now a large number of people.

Most companies over-invest here. The revenue lost to arbitrage is usually smaller than the revenue lost to the support burden of policing it. Start permissive and tighten only if you see real volume.

When not to do this at all

Regional pricing is a growth lever for products with a large addressable market and a low support burden per customer. If you sell a high-touch product to a small number of businesses, the right move is usually to negotiate individually and skip published regional pricing entirely, which keeps your public pricing simple and your discounting deliberate.

And if your product has not found its price in its home market yet, do not add a second variable. Get one price working first.

FAQ

Should I use purchasing power parity data to set regional prices?

Use it as a starting reference and then check it against your floor. PPP indices are built for consumer baskets, not for products with dollar-denominated COGS, so they routinely suggest prices below what an AI product can sustain.

How many price regions should I have?

Three to five. Every region is a support surface, a billing edge case, and a thing to maintain. Most companies get 90% of the benefit from a home price, a mid tier, and a low tier.

What if my token costs drop later?

Then you have a decision, and it is not automatic. Whether to lower prices when AI costs drop is worth thinking through before it happens, because the market will notice the cost drop before you announce anything.

Can I raise a regional price later?

Yes, and it is harder than raising a home-market price because the discount is often why the customer chose you. Build the increase into the plan from the start with annual terms rather than retrofitting it. The mechanics in raising prices on an AI product apply, with less room for error.