New software pricing metrics will force CIOs to change negotiating tactics

CIOs are being forced to rethink almost all elements of licensing negotiations, as well as managing how AI will be used, as major enterprise software vendors look to abandon per-seat pricing and shift to pricing based on consumption and/or agent interactions.

The evidence of such pricing shifts is now all but undeniable. “By 2028, pure seat-based pricing will be obsolete as AI agents rapidly replace manual repetitive tasks with digital labor, forcing 70% of vendors to refactor their value proposition into new models,” said a recent report from IDC. 

Brace for ‘meaningful’ price shifts

Executives from Salesforce and Workday have spoken about such likely pricing changes at recent investor calls. “We are selling back into our base and we’re focused not on just seats, but actually revenue per seat,” Workday CEO Carl Eschenbach told analysts during a recent earnings call. 

That means, said Adam Mansfield, UpperEdge advisory practice leader, “enterprises should brace for meaningful price shifts next year, both through unexpected in-term increases where proper protections aren’t in place and through higher-than-anticipated costs tied to AI products transitioning to consumption-based licensing. Salesforce’s move from per-seat or per-user pricing toward usage-based pricing with Agentforce is a prime example.”

Mansfield said that some enterprises are discovering that they were consuming AI conversations and surpassing the established volume of available conversations far faster than expected. Additionally, “in many cases, they never fully clarified what constitutes a ‘conversation’ or how usage would be counted for their specific use cases, and their order forms are void of the necessary detail. The unfortunate result is that customers inadvertently exceed their volume thresholds and suddenly owe Salesforce more money and, in some cases, it’s a significant amount that could have been avoided with proper up-front negotiation.”

Shifting the risk to customers

Sanchit Vir Gogia, the chief analyst at Greyhound Research, agreed. “Enterprise software pricing is undergoing a systemic reset,” he said. “The per-seat model, long the foundation of SaaS contracts, is collapsing under the force of AI-led workforce contraction, shifting business value away from headcount and toward automation. This is not a marginal evolution. It is a strategic rupture. Vendors are not just retiring outdated pricing logic.”

But, he noted, the critical change is that software companies are trying to shift almost all of the risks to their enterprise customers. 

“Vendors are transferring the cost volatility of AI compute to customers while monetizing customer-side productivity gains as margin. This risk displacement is now embedded in the architecture of consumption pricing,” Gogia said. “What was once a predictable licensing model is being replaced by vague units: credits, interactions, events, deliberately designed to obscure value exchange. In many cases, these units are metered without transparency or pre-defined ceilings. The result is a structural asymmetry: vendors preserve their margin floor while customers absorb the risk of overconsumption. To negotiate effectively, procurement must now develop fluency in AI mechanics, system telemetry, and the behavioral signals that trigger spend. Without this, contracts will outpace control and budgets will unravel fast.”

Tactics for CIOs

This change will force companies to strictly allocate AI usage and manage it, not unlike how T&E budgets are handled. That would mean a sharp change in how AI strategies are deployed, compared with today’s AI efforts. 

Aaron Perkins, CEO at Market-Proven AI, argued that the very high switching costs that enterprises would face if changing major software suppliers would make the threat to go elsewhere less effective. But more importantly, if all vendors abandon per-seat pricing, there won’t be anywhere else to go.

Analysts agree that the first area of negotiation when an vendor pushes a different payment model is to precisely define all terms. 

“A usage model means multiple things to multiple people. Are we talking just when a user is logged in or are we talking about backups that happen after they log off for the evening?” Perkins asked. “Ask the hard questions. ‘Explain to me what constitutes usage.’”

Another tactic is to negotiate strict ceilings on how much usage is being purchased and put the onus on the vendor to reach out to you in writing and ask for permission to go beyond that. If they don’t get that permission, then they have to eat any overage costs.

Perkins stressed that agentic interactions, especially fully autonomous interactions, pose an especially challenging pricing situation. If the vendor controls what the agent does and both parties have agreed to a consumption model, the vendor has an incentive to generate more activity so that it can charge more money.

Jason Andersen, principal analyst for Moor Insights & Strategy, recommends that CIOs start by insisting on a delay before the new pricing model goes into effect. 

“Asking for a one year stay of execution is a tactic people will follow,” Andersen said. This gives the CIO time to track current usage and to prepare pricing models for when the change happens. That means investing more in FinOps and purchasing better observability tools, he observed, noting, “we are starting to see decent software to meter these agents.”

“The vendors are going to respond ‘I am going to give you a billion tokens for X dollars. Tokens or requests, the terms are becoming interchangeable,” Andersen said. That will deliver a lower per-token cost initially because it will be a large upfront purchase. The catch is that if the enterprise goes beyond that initial number, and in the beginning, almost all enterprises will, then there will be a penalty or overuse charge. “The vendor will say ‘If you go over, I am going to keep taking your money but at a higher rate.’ You’ll then have to pass [the cost] along to the line of business,” he pointed out.

To manage consumption by autonomous agents, monitoring systems can be coded to set off an alarm when usage hits a threshold. “You can put policies on the agents themselves that ‘You shall not use more than X tokens,’” he said.

But there are risks. Just as a DDoS attack can force an enterprise to pay far more for hosting, usage metering can expose the enterprise to attacks from a rival company or a cyberthief looking to punish the enterprise. 

“If you have a customer service agent and they launch a series of bots to drive up usage,” that can cost a lot, Andersen said. He therefore recommends adding contract provisions to prevent this: “some sort of fraud protection. If I can prove that I am being attacked, I don’t have to pay for it. But you need to have the right circuit-breakers built into it,” he said.