Salesforce has stopped selling AI as an add-on and started baking it into every license. On the surface, it looks like a gift: Agentforce, Slack, Tableau, and Premier Support now come standard, along with a pool of “Flex Credits” to fuel your AI ambitions. But look closer, and the math tells a different story. The new Core, Advanced, and Max tiers aren’t designed for production-scale AI. They are designed for pilots. A single data-unification job can burn through your entire annual credit allowance in one run, and overage charges have no cap.
In this exclusive interview with CIONOW, Dr. Joseph Sweeney, Advisor, IBRS, an Australian ICT and digital advisory firm, decodes the real cost of Salesforce’s new model, explain why 78% of IT leaders are already facing unexpected AI bills, and offers a three-step formula for CIOs to model total cost of ownership without getting burned.
Salesforce, like most of the major core systems vendors, has stopped selling AI as an optional extra and started selling integral to the product. Why? Because AI is not, despite the hype, selling itself. Agentforce is still used by only a fraction of Salesforce’s 150,000-plus customer base, its stock was down 22 per cent for the year before its August earnings.
In short, Salesforece is using AI is a licensing packaging lever to pull the customer base upward. This is also an effort to boost Agentforce adoption as the default AI platform within it ecosystem, rather than a separate buying decision. Folding security and analytics into every tier partly addresses data readiness and governance objections most often blamed for slow AI adoption.
Based on what you have seen, how were the Flex Credit allowances per tier (e.g., 500,000 for Core, up to 2.75 million for Max) likely calculated? Are they sized for pilot/POC usage or for real production AI at scale?
The Salesforces allowances are sized for a pilot, not production. Core’s 500,000 credits equal about 25,000 standard agent actions a year. In a 200-seat organization, that is roughly one agent conversation per user per month. Nobody builds an agentic enterprise on one conversation a month. These are tasting portions, and they are deliberately so: the allowance gets every customer started, and the meter gets Salesforce paid the moment usage becomes real.
The credit values at list price (US\$2,500 for Core, US\$5,000 for Advanced, US\$13,750 for Max) are immaterial against the seat price uplift, which confirms they function as an adoption incentive rather than a production capacity commitment.
Is the claim credible that a single large Data 360 or data-unification job can burn through a tier’s entire annual Flex Credit allowance in one run? Have you seen customer examples that support or contradict this?
Yes, this is credible, and the arithmetic is on Salesforce’s own rate card. Data 360 unification is billed at 75,000 credits per million rows processed at the base tier. Core’s entire annual allowance is 500,000 credits. So a single unification run over roughly seven million rows consumes the credit allocation in one job.
A 100-million-row customer estate, unremarkable for a large enterprise, would burn through Core’s allowance about fifteen times over and Max’s nearly three times. No customer has publicly confirmed such a bill yet, but the rate card makes the outcome a matter of multiplication, not speculation.
One nuance worth noting: monthly tier thresholds make a single concentrated backfill cheaper per credit than spreading the same work across months, so how the job is scheduled matters as much as its size.
From a buyer’s perspective, how should a CIO model total cost of ownership now: base license + expected overage credits + bundled features they may never use? What’s a realistic range for AI overage costs in year one?
Model it in three layers.
Third, sweat the assets you already own. Treat bundled products that duplicate them as zero value, not as savings. On overage: there is no cap and no circuit breaker, so consumption above the pool bills at your contracted rate. Organizations running agents in production should budget overage at 50 to 150 per cent of the included credit value in year one. Anything with heavy Data 360 unification or real-time workloads can exceed that in a single month.
A a side note, 78 per cent of IT leaders report unexpected charges from consumption-based AI pricing, and consumption models are overshooting budgets by around 40 per cent on average versus roughly 5 per cent for seat-based licenses.
How do you expect this pricing model to influence CIO behavior: more conservative AI rollouts, shift to alternative platforms, or deeper negotiation around credits and tiers at renewal?
All three, in that order of likelihood. The near-term effect is quieter, more carefully metered rollouts. Salesforce clients will gate agent deployments behind consumption dashboards rather than ambition. The strategic effect lands at renewal, where credits, overage rates and unmetered tiers become the negotiation surface, and Salesforce has already signalled it will deal. The Agentic Enterprise License Agreement, a fixed-price all-you-can-eat contract, exists precisely to take consumption risk off the table for customers who push. Platform switching is the least likely response as the switching cost of CRM dwarfs a 13 per cent price rise, and every rival is moving the same direction.
My key recommendation: The CIOs who win will be the ones who arrive at renewal with their own consumption data rather than Salesforce’s.
In the broader context of 2026 SaaS pricing, is this more like Microsoft Copilot bundling, Atlassian’s usage-based AI meters, or something unique to Salesforce? What pattern should CIOs watch for next?
It is not just Microsoft doing this sort of bundling of AI (and there is more to come from them in November!). It is a trend right across all major platforms. Atlassian, Canva, Oracle, Google with Workspaces… this is the future.
