Salesforce Has Rewritten Its AI Pricing Four Times in Two Years. The Product Underneath Hasn't Changed.

An earlier piece in this series looked at how Bonterra and Blackbaud handle donor data. This one looks at a different kind of vendor behavior: charging more for "AI" without much sign the AI changed anything.

Every SaaS renewal notice this year seems to mention AI somewhere in the price increase. What's harder to find is a vendor that can point to what the AI actually does differently for the price. Salesforce is the clearest case, and it's worth walking through in detail, because the company has now tried four different ways to charge for the same feature — each one confirmed off its own pricing page, not secondhand.

Agentforce is Salesforce's existing CRM with a conversational agent layered on top

Agentforce — including Agentforce Nonprofit, the rebrand of Nonprofit Success Pack — didn't rebuild Salesforce's core platform. It added an agent on top of it. Reviewers on G2 and Capterra describe the nonprofit version as "basically the same platform with just a few added features," and note it still requires heavy configuration to handle basic nonprofit workflows like program management and reporting. The AI is the new part. The infrastructure underneath it, and the friction reviewers already complained about before Agentforce existed, is not.

That gap between "we added an AI layer" and "we changed what the product actually does" shows up most clearly in how Salesforce has priced it.

Four pricing models in under two years

Salesforce has re-metered Agentforce four times since its late-2024 launch:

• Late 2024: $2 per conversation, flat — charged the same whether the agent resolved the issue or the customer gave up and asked for a human.
• May 2025: Flex Credits — $500 per 100,000 credits, with a standard action costing 20 credits (about $0.10) and a voice action costing 30 credits (about $0.15) — introduced after customer resistance to the flat per-conversation fee.
• Per-user licensing, stacked on top: Salesforce's live pricing page lists an Agentforce User License at $5/user/month (requires Flex Credits on top), Agentforce Add-ons at $125/user/month for "unmetered" internal use, and Agentforce Industries Add-ons at $150/user/month — three more tiers of the same feature, each solving a problem the last one created.
• July 2026: Agentforce Help Agent, with pay-per-resolution pricing. Salesforce's own announcement states the mechanism plainly: "you are charged when Help Agent resolves an issue autonomously, start to finish… If the customer asks for a human or walks away unhappy, there is no charge." Trade coverage (SalesforceBen, Enterprise DNA, CIO) reports the rate at $2 per resolution — a figure Salesforce's own release doesn't actually state.

For Agentforce Nonprofit specifically — the version aimed at the sector this newsletter covers — none of this pricing history is even visible up front. Salesforce's own announcement page for the product directs customers straight to sales, with no public number to check the pitch against.

Four rewrites of the same feature, each dropped after the last one drew pushback, is not what pricing looks like when a product is delivering clear, stable value. It's what pricing looks like when the vendor is still trying to find a number customers won't reject.

Meanwhile, Salesforce has been telling a different story to investors than to customers. In September 2025, Marc Benioff said the company had cut its support organization from roughly 9,000 people to about 5,000 — some 4,000 roles — crediting Agentforce for absorbing the difference. Salesforce has since gone through multiple further layoff rounds (including a WARN filing for 86 California roles in June 2026) even as Agentforce hit $1.2 billion in annual recurring revenue, up 205% year over year. The AI is framed as strong enough to replace thousands of jobs and simultaneously in need of a fourth pricing model to convince the customers paying for it.

It's not just Salesforce

The same mechanism shows up elsewhere, in less dramatic but equally telling form. Zendesk's own live pricing page lists Suite Professional at $115/agent/month and a Copilot add-on at $50/agent/month, only unlockable on that tier or higher; add the Workforce Engagement Bundle (another $50) and a team pays $215/agent/month — 87% over the $115 base, for the same ticketing product plus an AI layer. And Blackbaud, which announced a Microsoft 365 Copilot integration for Raiser's Edge NXT in August 2026, discloses no price for it anywhere on its own newsroom release or product page — consistent with a vendor that already runs on quote-only pricing and a flat $600/year "PCI compliance" surcharge charged to every customer regardless of volume. Blackbaud's own research, published the same year, found only about a third of social-impact professionals think their organization uses AI effectively — a strange market to be selling an undisclosed-price AI upgrade into.

Neither Zendesk nor Blackbaud has rewritten its pricing model as many times as Salesforce has. That's the point of leading with Salesforce: it's the sharpest, most-repeated version of the same behavior, not an isolated one.

The pattern

Forrester's 2026 guidance to enterprise buyers is blunt: software vendors are "raising prices and piling on usage charges" through new metering units — credits, tokens, per-resolution fees, per-seat AI tiers — layered on top of the existing subscription. The Register put it more directly in July 2026: AI vendors have found someone to pay their infrastructure bills, and it's the customer, not the vendor's own margins. Forrester separately warns about "AI washing of layoffs" — cutting headcount and crediting AI for the savings as a way to justify price increases without having to prove the tool did the work. Salesforce's own numbers — the layoffs, the ARR growth, the four pricing rewrites — are close to a textbook case of exactly what Forrester is describing.

What to ask before you pay for it

• What does the AI feature actually change about the outcome you get — not the workflow, the outcome — compared to before it existed?
• How many times has this vendor changed how they charge for this feature? Each change is a data point on whether the last price actually matched the value.
• If the vendor cut staff and credited AI for it, ask whether your price went down to match. If support headcount fell and your bill didn't, the savings went somewhere — not to you.

A companion piece looks at what's underneath the price tag: Agentforce is an AI layer running on top of the same CRM data organizations already have, and that data is often fragmented, duplicated, or stale across systems that were never reconciled. An agent can't tell an authoritative record from a bad one — it just answers, confidently, either way. Four pricing rewrites without a product rebuild is one symptom of that; unreliable AI output is the other. A third piece looks specifically at how AI is reshaping per-seat pricing itself, across Salesforce and other vendors.

---

A Note on Our Position: Salesforce and Zendesk aren't competitors to engine9 — neither builds nonprofit-specific fundraising or engagement tools. Blackbaud's core fundraising product does compete more directly in that space, and we want to say that plainly rather than downplay it: we have an interest in how organizations think about vendors like Blackbaud, even if this piece doesn't touch its core CRM, just the Copilot-for-Raiser's-Edge integration. What we don't have is a stake in how any of these three vendors price their AI features specifically; every claim above points at their own public pricing pages and public statements, not at anything about us. If we have misread or misrepresented anything here, we will correct it.