Why Flat Pricing Is Running Out of Road

Flat, per-seat pricing has been the default in SaaS for so long that it barely feels like a choice anymore. One price per tier, sometimes multiplied by seat count. The assumption underneath it is simple: every customer on the same plan costs roughly the same to serve.

The assumption was never fully true. Cost to serve has always varied by customer. Some accounts hit your API harder, or store more data, or lean on support more than others, even on identical plans at identical prices. Flat pricing worked for as long as it did because the variance was usually small enough, and too costly to measure precisely.

AI changed that math, the same way it changed the chargeback math inside IT departments. The variance in what each customer really costs you grew large enough, and fast enough, that averaging it away now means either overcharging your cheapest customers or losing money on your most expensive ones, and often both at once.

The data backs this up. Roughly 41 percent of SaaS companies now formally monetize AI features to some extent. So a majority still have not, and are quietly absorbing AI cost inside a flat price built for an older world where that cost barely existed. Among the ones who have started charging for AI, the split is almost even, a little over half default to bundling it into subscription tiers, and just under half have already moved to usage, consumption, or outcome-based pricing for that specific feature. Credit-based pricing, a kind of pre-paid usage bucket, has surged as a stopgap in between. Several people who study this space closely describe credits the same way: a bridge, not a destination, a workaround while companies figure out what they actually want pricing to reflect.

That phrase, “a bridge, not a destination,” is worth sitting with, because it’s the pricing version of the exact same problem the rest of our posts have been circling. Credits are an estimate dressed up as a solution. They let a company avoid the hard question - which customers and features actually drive our cost - by selling a bucket of usage upfront and hoping the average works out. It’s headcount-based chargeback wearing a retail costume.

Here’s the harder question. If you don’t know what a customer actually costs you, on the infrastructure and AI layers specifically, every pricing decision you make is a guess. You might guess well but you’re still guessing. Guesses can go wrong in two ways, and it’s not symmetrical. Underprice your heaviest users and you fund their usage with your other customers' margin, quietly, for as long as nobody notices. Overprice your lightest users and you hand a reason to leave to the exact customers who cost you the least to keep.

This is not a new failure mode that AI invented. It’s the oldest failure mode in subscription pricing, cross-subsidy, where your best customers by cost-to-serve are quietly paying for your worst ones. AI just made the subsidy large enough to show up on an income statement.

The uncomfortable part is that pricing teams have known this in the abstract for a long time. Usage-based pricing has been rising in SaaS since well before AI, roughly sixty percent of companies had adopted some form of it a few years ago, well ahead of the current AI wave. What’s different now is not the idea. It’s the excuse for not acting on it. When cost-to-serve variance was small and expensive to measure, flat pricing was a reasonable default. Once the variance is large and the measurement is cheap, flat pricing stops being a reasonable default. That’s the case now, thanks to instrumentation standards that already exist, and for AI and cloud infrastructure specifically.

This doesn't mean every company should jump to pure usage-based pricing. Customers value predictability. Finance teams value forecastability. A pricing model that swings wildly with usage creates real friction on both sides of a deal.

That's why hybrid models are winning. A base subscription covers the predictable part. A usage-based layer covers the part of the product that actually varies in cost. Customers still get the predictability they want. The bill still tracks the real cost underneath it.

This is not a compromise. It’s pricing that finally matches the shape of the underlying cost structure instead of averaging over it.

A hybrid model is only as good as the attribution underneath it. You cannot build a usage-based layer on a cost structure you cannot see per customer. Most companies are missing the data, not the pricing model.

Decide your price architecture after you can see cost per customer, not before. Otherwise you are just building a more complicated version of the same guess.

The companies that get this right in the next few years will not be the ones with the cleverest pricing page. They’ll be the ones who did the unglamorous work first, attributing cost to the customer and the feature, so that whatever pricing model they choose after that is actually built on something real. Everyone else will keep tuning tiers and adding credit packs, treating a measurement problem as if it were a marketing problem, and wondering why the fix never quite holds.