When More Means Less: The AI Pricing Paradox

Distinguished Professor Hemant Bhargava of the UC Davis Graduate School of Management presented on "AI Pricing: The Value-Revenue Paradox" at the Wharton Human-AI Research's 3rd Annual Business & Generative AI Conference on September 4, 2025, in San Francisco. Bhargava is the director of the Center for Analytics and Technology in Society at UC Davis.

Bhargava's research shows that seat-based pricing dominated the software-as-a-service (SaaS) industry, and has quickly became popular for AI tools (e.g., ChatGPT’s $20/per-user-per-month model). But with AI’s promise to increase productivity and reduce headcount, seat-based pricing can create a paradox where firms that enjoy greater value and savings pay less; a paradox that does not bedevil traditional software.

Wharton's research conference brings together academics and industry researchers to connect, collaborate, and explore bold ideas at the intersection of Generative AI and global business transformation. Through focused discussions on new research and innovative thinking, the conference examined how GenAI is reshaping business models, industries, and economies worldwide.

View Bhargava's full presentation.

Learn more about Professor Bhargava.

Learn more about the conference.

Video Transcript

Distinguished Professor Hemant Bhargav, UC Davis Graduate School of Management

All right, thanks, Karthik. I’ll spend one minute on a different topic before I get into this talk on the value-revenue paradox.

In my editorial positions, I’ve been working with others who set the AI-use policy for authors. But I think we have a much bigger problem coming at the journal level. We just completed a paper on what journals need to do to adopt AI, and our recommendations go against the direction most journals are taking. It’s something I want to spread the word about and discuss further with colleagues.

Now, back to this talk. This paper is about a paradox we’re observing in how firms are pricing AI products. For software, one of the most common enterprise pricing models is per-seat pricing—you see it everywhere. When AI products started being released, many firms adopted a similar model. For example, we pay about $20 a month for ChatGPT. GitHub Copilot, Salesforce and many others are also sold to enterprises on a per-seat basis.

This has worked fairly well for traditional software. One goal of a pricing model is to align payment with value, so that firms receiving more value pay more. That’s been the logic behind software pricing. But AI products present a new challenge.

AI can replace human labor and drive significant productivity gains. Companies talk about big cost savings in areas such as coding, customer service, paralegal work and more. Here’s the paradox: with per-seat pricing, firms that reduce headcount the most—and thus gain the most value—end up paying less because they have fewer employees using the AI.

Unlike with other technologies—say, electronic medical records—you don’t cut the number of doctors after adoption. With AI, however, greater headcount reduction leaves fewer people to pay for the product. That’s the paradox I’ll explain.

Our model imagines a customer base of many firms with different numbers of employees and different capacities to leverage AI. Some firms use AI without replacing humans. Others aggressively downsize to maximize cost savings. When vendors set a per-seat price, firms with fewer employees but higher cost savings may pay less while gaining more.

The model introduces a parameter—omega—that represents the AI’s inherent capability. A higher omega product enables greater headcount reduction. In cases where omega is high, the paradox emerges: firms getting the highest value from AI actually end up paying less.

This doesn’t occur with traditional software, which usually drives only modest headcount reductions. But AI tools promise large-scale labor substitution.

So, what does this mean for vendors? If you improve your AI’s capability, you might actually worsen the paradox. Should firms then rethink pricing models? Perhaps token-based pricing, or better yet, outcome-based pricing, where payment is tied directly to results delivered. We’ve seen similar approaches in digital advertising with cost-per-click and cost-per-action replacing cost-per-impression.

Of course, these models raise their own challenges, such as measurement, auditing and trust. But the industry needs to explore them, because simply adopting per-seat pricing as we’ve done for decades with software may not be sustainable.

Other complexities exist—private price discrimination, segmented pricing, negotiated discounts. But none of these approaches eliminates the paradox entirely. In fact, they may worsen it, with more capable firms extracting deeper discounts.

So, the broader point is this: AI substitution technologies demand rethinking pricing. Otherwise, we risk a persistent misalignment between value created and price paid.