Stop Pricing AI Like a Utility: The Case for Price Differentiation

I was recently in a discussion about how a software company should monetize its AI capabilities. We were debating whether certain AI-powered features should be included in the base product, while others should warrant incremental charges.

One of the participants made an interesting argument in favor of charging for all AI usage. He offered an analogy:

"Think about water. You use water for drinking, and you use water to flush your toilet. But regardless of what you use it for, you pay for the water you consume."

It's an intuitive argument. If AI creates a variable cost every time it's used, why wouldn't we charge customers according to how much they consume?

But the analogy also highlights something I think is getting lost in the current enthusiasm for usage-based AI pricing: the importance of price differentiation.

Water is generally priced according to consumption because it's a relatively undifferentiated commodity. The water used for drinking might create considerably more value than the water used to flush a toilet, but distinguishing between those uses at the point of supply would be impractical.

Where differentiation is possible, we routinely see very different pricing. Bottled water, for example, commands a substantial premium over tap water because it addresses different customer needs.

This raises an important question for AI pricing: if we can distinguish between different uses of AI, and those uses create dramatically different amounts of value, why would we choose to price them all the same way?

The danger of confusing usage with value

One of the most fundamental principles of pricing strategy is that different customers, capabilities, and use cases create different levels of value, and therefore support different willingness-to-pay. Effective pricing seeks to capture an appropriate share of that value through price differentiation.

Yet the prevailing logic in AI monetization often starts somewhere else: AI has a variable cost of delivery, that cost increases with usage, so customers should be charged according to how much AI they consume.

The problem is that the resources consumed by an AI action may have little relationship to the value it creates.

Consider two AI-powered actions within an enterprise software platform. One summarizes a lengthy email conversation. The other identifies a contractual risk that could expose a business to a significant financial loss.

Both consume tokens and computing resources, but the value they create is dramatically different. It's entirely possible that the email summary consumes more tokens than the risk identification. If we charge based on tokens consumed, we could end up charging more for the less valuable action.

Cost is an important consideration in pricing, but it should be a constraint on our economics, not the primary determinant of what customers pay. Otherwise, we risk under-monetizing our most valuable capabilities while overcharging for relatively mundane ones.

There's a compelling reason why so many companies adopt cost-plus consumption-based pricing for AI: simplicity. It's relatively straightforward to implement, easy to explain, and avoids the difficult work of determining what different AI actions are actually worth. For companies primarily focused on driving adoption, that trade-off may be entirely reasonable.

But as the AI market matures and attention shifts from adoption toward monetization, that simplicity can leave substantial value on the table. 

Not all AI deserves to be monetized equally

Recognizing the difference between usage and value has another important implication: some AI-powered features are better included in the core product than monetized separately. 

As AI becomes embedded in enterprise software, customers will increasingly expect certain capabilities as standard. Summarization, natural-language navigation, and routine assistance may simply become part of the core product experience. Attempting to monetize every interaction could create unnecessary friction and undermine the value proposition of the broader platform.

Other capabilities are fundamentally different. An AI agent that resolves complex customer cases, identifies high-value sales opportunities, or automates entire business processes may create substantial incremental economic value. These capabilities can justify a meaningful price premium.

The challenge is to design a pricing model that reflects those differences. That might mean including some capabilities in the base subscription, charging separately for others, or differentiating prices according to the value of individual actions.

Effective AI monetization requires a deliberate choice about which capabilities to charge for, how to charge for them, and how much value to capture.  

Of course, including AI capabilities in the base package doesn't mean ignoring their cost. Vendors should budget for a reasonable level of expected consumption and ensure that the economics of the base subscription can support it. Beyond that, fair-use allowances can protect against unusually heavy consumption, with customers paying incremental charges when they exceed those limits. These charges can often be relatively modest and are easier to justify as a matter of fairness rather than as an attempt to monetize every interaction. If a credit system is already in place, excess usage can simply draw against the customer's credit balance, although other mechanisms can work equally well. 

Credits can enable price differentiation, if designed correctly

Credits have become a popular mechanism for monetizing AI, but there's a fundamental misconception about what they actually are. Credits are not a price metric. They're a virtual currency. Customers purchase credits, then spend them on specific capabilities or actions. The currency itself tells us nothing about whether the underlying pricing model captures value.

A system where every 1,000 tokens consumes one credit is fundamentally different from one where different actions consume credits according to their relative economic value. The former is essentially metered compute consumption. The latter can become a sophisticated mechanism for price differentiation.

Imagine an enterprise software platform offering multiple AI-powered actions. A routine document summary might consume one credit, a sophisticated contract analysis might consume 20, and an automated workflow that completes a substantial business process might consume 100.

The vendor can establish differentiated prices for different types of work while allowing customers to purchase and manage a single pool of credits. This makes it possible to align pricing with multiple sources of customer value without creating a confusing collection of separate purchasing mechanisms.

The challenge lies in determining the appropriate number of credits for each action. That requires a structured approach to assessing value and cost, rather than assigning arbitrary weights. 

Designing a value-aligned credit system

At Monevate, we've developed a four-step framework for designing credit systems that balance economic viability with value capture.

1. Define valuable actions. Identify the distinct actions your product performs, determine which warrant monetization, and estimate their underlying delivery costs.

2. Establish relative value. Assess the relative customer value of those actions alongside their cost profiles. Develop monetization weights that reflect both willingness-to-pay and economic risk.

3. Translate actions into credits. Convert those relative weights into a practical credit structure, with simple, understandable consumption rates for different actions.

4. Establish the credit price. Triangulate across delivery costs, required margins, competitive alternatives, customer willingness-to-pay, and the cost of non-AI solutions.

The resulting architecture should then be tested across customer profiles and usage patterns. Do the prices make sense for individual actions? Is the model economically sustainable when customers disproportionately use certain capabilities? Is it sufficiently intuitive that customers can understand and predict what they'll spend?

A well-designed system should differentiate pricing according to economic value while remaining straightforward for customers to understand and use.  

Don't let AI turn your product into a commodity

There are circumstances where straightforward consumption-based pricing makes perfect sense. For infrastructure providers selling standardized computing resources, usage may be the most logical basis for charging. For certain AI applications, workload can also be a reasonable proxy for value.

But enterprise software vendors selling differentiated capabilities and business outcomes should be cautious about adopting infrastructure-style pricing models. Their strategic advantage lies in what their software accomplishes, and their pricing should reflect the value of those accomplishments.  

AI offers an extraordinary opportunity to create and capture new forms of value. Our pricing models should recognize those differences rather than flattening them into a single commodity measure.

Recovering the cost of AI is relatively straightforward. Capturing a fair share of the value it creates is the real commercial challenge. And for all the novelty of AI, the solution rests on one of the oldest principles of effective pricing: price differentiation.

‍