A recent effort to decorate for my wife's birthday party made me reflect on the choices enterprise IT leaders must make when procuring agentic AI services and programs. You might think I'm full of hot air, but let me inflate this concept for a moment.
At a store's party counter, I was told they couldn't inflate balloons for me. Their industrial helium tank had run empty. But they informed me I had another option – to purchase another smaller tank of helium and fill the balloons myself. Instead, I chose to leave. I went online and ordered balloons from a service that delivered them inflated and directly to my door on the morning of the party. It was more expensive, but I didn't have to do any work and I got the outcome I wanted.
My balloon procurement had me considering three different business models to achieve the same result. Each had its pros and cons to weigh:
- Flat-rate pricing. Pay a set price for a tank of helium. Least expensive option and predictable budget, but I have to work to achieve the outcome I want. Might run out of helium before all the balloons are filled.
- Consumption-based pricing. Pay a metered rate for the helium until all the balloons are inflated. I'll be guaranteed to fill all the balloons I want, but the budget is less predictable. Might pay for lost air.
- Outcome-based pricing. Pay for the balloons to be filled for me and delivered. Least amount of work, outcome guaranteed, and predictable budget. Most expensive option.
Sound familiar? Any enterprise IT leader navigating agentic AI procurement will recognize these approaches from vendors. Except it's not helium the vendors are providing access to, it's inference compute. And the outcomes that leaders want to achieve are a lot more complex than a set of floating balloons.
The business model that a vendor offers with agentic AI is informed by the business model they started with. Software as a service (SaaS) vendors are adding an additional layer of consumption metering. Business process outsourcers are focused on outcome-based pricing. For enterprise technology buyers, not only can they select a business model that suits them, but they can negotiate at some crucial leverage point.
Infotech's Global IT Leadership Survey shows that four business models rise to the top when IT leaders are asked about their preferred payment model for AI/software solutions. The most popular, at 36.4%, remains subscription per user. The next most popular is open source software with a support services contract at 23.1%, followed closely by consumption-based or utilization-based pricing at 22.1%. A one-time purchase rounds out the top four at 13.9%.
Are SaaS vendors full of hot air on outcome-based pricing?
Salesforce is an example of a SaaS vendor that's expanding to an AI-driven business model since launching Agentforce. Agentforce pricing has changed multiple times in recent months as the vendor that popularized cloud CRM navigates its way to AI-driven CRM.
Customer-facing agent consumption can be priced per conversation, an older approach that launched alongside Agentforce in 2024, or more recently a price-per-resolution option. Alternatively, both customer-facing and internal agent consumption can be priced using flex credits, with a standard number of credits required to complete different actions.
The "per conversation" model only applies to customer-facing interactions, so flex credits are needed for any back-office AI interactions. The "price-per-resolution" model is only available for Help Agent, a pre-packaged agent designed to help resolve customer inquiries. Salesforce describes this as "outcome-based pricing," and there's a shift in risk here because Salesforce won't charge for the agent interaction if the customer requests escalation to a human agent. Still, dig into the details, and you'll see Salesforce defines the resolutions and imposes upper limits such as 10 minutes of voice interaction before ticking the meter up.
Customers can focus on getting discounts for flex credits they commit to buying in advance. The more credits and the longer the contract, the steeper the discount is likely to be. Other points for optimization include where customers should pay the per-seat add-on for unmetered use, and if unused credits can roll over to the next period.
When the “customer is always right” turns into “the customer always negotiates”
Another company that’s delving into AI-powered pricing around its services, but in a very different way, is Horizon Media. The advertising agency launched agentic buying capabilities as part of its Blu platform in July. The platform packages together agentic AI, data sources, and analytics tools for clients to build their own audience groups and deliver ads to them.
At Horizon Media, the pricing model is flexible and based on the needs of the customer. As customers do their own analysis on the platform directly, they run into variable costs and find reason to quibble with the outcomes resulting from their consumption. As the executive put it, this leads to a constant negotiation, as many as "dozens of times a week with each client."
The clients insist they shouldn't have to pay when they perceive no value was delivered. The agency, which paid for the inference costs and data calls to respond to the prompt, has to respond that the users need more training to achieve productive results. Most of the time, the matter is settled with the client paying most of the consumption costs, but not all of them.
Outcome-based pricing was the obvious choice for my party balloons because the outcome was easy to verify. A balloon is inflated, or it isn't; it either floats or it doesn't. It's not so clear-cut with agentic AI. The space for dispute in the contract is exactly where a buyer's leverage sits. Whether it's the definition of an outcome or what exactly is being metered, focus your negotiation efforts and push back there.
At my party, no one examined the invoice or asked how the balloons were prepared. Enterprise buyers who put agentic AI at the core of their operations are facing more difficult choices and service relationships. The CFO will be reviewing the bill, and the vendor's definition of 'done' will matter as much as its price. When the stakes are that inflated, know your leverage points before you sign on the dotted line.
Brian Jackson is a principal research director at Infotech, where he focuses on technology trends and has developed Infotech's annual Tech Trends report. He writes columns on Forbes Technology Councils and Spiceworks Insights. He is regularly featured as a technology expert by media including CIO.com, CIO Dive, Techopedia, Chorus Radio, CBC, and elsewhere.