Every established SaaS company now has a bunch of new“AI-native” challengers claiming to reinvent your category. But here’s your advantage: data.
You already know your customers—how they buy, what they use, what drives renewal, and what triggers churn. You can see real usage patterns and understand where value lies. That gives you a huge edge: the ability to launch, test, and reprice.
For twenty years, SaaS was expensive to build but cheap to scale. Agentic AI flips that: it’s cheap to build and expensive to run. Every interaction burns compute. If you’re not careful, your AI service can scale your cloud bill faster than your revenue.
That’s the paradox: you need adoption to prove value, but adoption can crush you if you don’t manage costs. Be too cautious, and no one adopts; be too generous, and GPU costs spiral. The solution starts with how you define and charge for usage.
It starts with your Unit of Measurement (UOM).
What is a UOM, you ask?
Your UOM is the metric that captures how and why customers consume and pay for your service. Get it right, and your revenue grows in sync with customer value. Get it wrong, and you’re either burning cash or chasing fake growth.
As I recently told CNBC, too many companies (I’m looking at you, Adobe, Microsoft, and Google) are hiking prices without a clear value rationale. Your pricing should never be driven by the fact that you have a bunch of new data centers to pay for. Instead, it should be tied to how customers succeed with your product.
Here are three smart ways to find your UOM (and three not-so-smart ones).
Start with what the customer gets.
Too many teams ask, “What can we meter?” The better question: “What value does the customer actually get?”
For a support AI, the answer isn’t API calls—it’s tickets resolved or hours saved. For a marketing AI, it’s not prompts—it’s published content that performs. This pricing feels fair and builds trust because it maps directly to progress, not process.
When your metric aligns with customer success, expansion and renewal follow naturally. When customers win, you win.
Pick a metric that scales, maps to cost, and is dead-simple to track.
A good metric grows with customer value, is visible in analytics, and roughly correlates to cost. Take an AI transcription service that charges per hour of audio processed. It’s simple, measurable, and tracks with compute usage.
But balance is key: your UOM has to protect your margins while encouraging exploration. If customers can clearly see value and you can stay profitable, you’ve found the right UOM.
Embrace repricing!
You’ll never get pricing perfect on the first try. Your initial model is a hypothesis. Companies like Intercom, Zendesk, and Salesforce constantly adjust pricing and service terms based on what they learn.
Once customers start using the product, you’ll see where value actually lives: who’s over-consuming, who’s under-using, and where the margins appear. Use your data to identify which usage metrics correlate with value, and price accordingly.
But velocity matters. If it takes two quarters to approve a change, you’re too slow. Build financial systems that let you test, tweak, and reprice fast.
Don’t price based on system activity.
“API calls,” “inferences,” “GPU hours”— these are your costs, not your customers’ value.
Charging for system activity makes customers feel like they’re paying for inefficiency. It breeds friction and distrust. Instead of experimenting, they’ll spend their time auditing bills.
If your customers are afraid to use your product because they know the meter is running, you’ve already lost.
Don’t copy your competitors.
Just because OpenAI charges per token or Salesforce charges per seat doesn’t mean you should.
Your product creates value in its own way. Maybe it saves time, improves accuracy, or drives revenue. Copying someone else’s metric guarantees you’ll miss your own.
Competitor pricing is a reference point, not a strategy. Your pricing should tell the story of your differentiation, not someone else’s.
Don’t pick a UOM you can’t explain in one sentence.
If your pricing needs an FAQ or a slide deck to explain, you’re in trouble.
“AI insights generated” might sound clever, but no one knows what it means, or how to count one. Ambiguity leads to billing disputes and churn.
Your UOM should be obvious, transparent, and boring. It should be something a customer can repeat to their CFO without translation.
The Balancing Act
As we discussed with Michael Mansard of the Subscribed Institute, this process is a constant balance between cost, adoption, and value. You need adoption to realize value, but adoption can kill you if you don’t control costs. Tighten too much, and no one adopts.
As the great AI pricing analyst Kyle Poyar puts it: “There is no such thing as a perfect pricing model. The best you can do is choose one that tells the story of what you do, who you’re for, and why you’re better—and have a plan to manage the inevitable downsides.”
Defining your UOM isn’t a one-time project. It’s an infinite loop of testing, learning, and refining. In other words, it’s very zen.
