.png)
.jpg)
Frameworks, core principles and top case studies for SaaS pricing, learnt and refined over 28+ years of SaaS-monetization experience.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Introduction
Salesforce’s grand foray into “agentic AI,” dubbed Agentforce, has been a case study in how pricing is positioning. Launched in late 2024 as an AI-driven digital agent platform, Agentforce’s pricing strategy went through whiplash-inducing changes over two years. What began as a simple-sounding $2 per conversation fee quickly met with customer confusion and backlash, forcing Salesforce to overhaul its approach[1][2]. By early 2026, Salesforce had shifted to a far more complex scheme combining usage-based credits and per-user licensing, essentially reversing course to better align with enterprise buyer expectations. This post analyzes the timeline of these pricing pivots from 2024 to 2026, the context and reactions surrounding each change, and what they reveal about Salesforce’s internal strategy and the broader SaaS market climate. We’ll also compare Salesforce’s model to competitors like Intercom, and distill key lessons for SaaS leaders – chief among them: pricing must align with customer value perception and macro realities, or even a cutting-edge product can stumble.
At launch, Salesforce pitched Agentforce as revolutionary – “AI that will change everything at work” – but its initial pricing model sent a very different message. The missteps and subsequent corrections underscore that how you charge for a product positions it in customers’ minds. In Salesforce’s case, an ill-fitting pricing metric signaled poor platform alignment, denting trust and adoption. Compounding the issue was a tough economic backdrop of software price deflation and AI cost compression, which made buyers especially sensitive to unclear or runaway costs. Let’s dive into how it all unfolded.
2024: Agentforce Debuts with Per-Conversation Pricing – and Immediate Confusion
When Salesforce first unveiled Agentforce (rebranding its earlier “Einstein Copilot” project) in Fall 2024, the company chose a usage-based pricing metric: $2 USD per conversation[1]. On paper, this “pay per conversation” approach sounded straightforward – reminiscent of how one might pay per chat session or case resolution. Salesforce likely intended it to communicate “you only pay when the AI engages a user.” However, the reality proved far messier. Customers and analysts immediately began questioning what exactly counted as a “conversation” and how costs would be tracked[2]. Was a multi-turn exchange one conversation or several? What if an AI agent handled part of an issue and a human finished it – would that still cost $2? Such ambiguities left early prospects uneasy.
“Previously, Salesforce charged $2 per conversation for access to Agentforce... a model that left customers with sticker shock and questioning what exactly counted as a ‘conversation.’”[2]
Even before Agentforce reached general availability (GA on Oct 25, 2024), murmurs of doubt spread in the Salesforce ecosystem. Many enterprise software buyers are accustomed to pricing per user or per month – metrics that align with their budgeting practices. Paying per AI conversation was alien to how they plan costs, and it felt like a potentially unbounded expense. As one pricing expert notes, when pricing is well-aligned, “the customer is not questioning the metric. If they are paying per user, they know that per user is what works for them and that is how they do their budgeting”[3]. In this case, customers were questioning the metric – a clear warning sign of misalignment.
Concerns over cost predictability emerged immediately. For example, a support team leader calculated that just 5 agents handling ~70 AI-assisted conversations each per day could rack up roughly $900 in daily fees – on the order of $20,000+ per month[4]. For mid-sized organizations, that implied “sticker shock” and an untenable budget hit. The seemingly low $2 fee, when multiplied at scale, suddenly wasn’t so low. Salesforce had inadvertently handed CFOs and procurement teams a “blank check” scenario: an uncapped usage-based cost with no clear ROI guarantee[5].
Defining “conversation” was another sticking point. Was it any session an AI agent started? Did a trivial inquiry that went nowhere still count (and cost $2)? Salesforce’s documentation at the time was thin, leading one industry observer to describe the $2/conversation model as “messy”[6]. Customers worried about paying for “short, long, meaningful, or useless” interactions alike[7]. In other words, the pricing didn’t distinguish between valuable outcomes and fruitless chats – you paid $2 whether the AI solved a problem or not. Competitors were quick to point out this flaw: Intercom, for instance, priced its new AI support bot (“Fin”) at $0.99 per resolved conversation, explicitly charging only for successful answers that actually deflected a ticket[8]. By contrast, Salesforce’s flat $2 fee “failed to reflect the value of resolution”, as one community blog noted[7]. An unresolved Agentforce session could still incur a charge, effectively punishing customers for the AI’s failures.
It didn’t help that Salesforce’s own massive installed base had expectations shaped by years of relatively predictable subscription licensing. Many customers initially assumed Agentforce might be an included feature (or a reasonable add-on) to the core platform. Instead, they found a new usage-metered SKU. Some felt blindsided. The Salesforce community’s reaction ranged from genuine excitement about the AI tech to feelings of betrayal among loyalists who saw the pricing as a money grab[9]. On social media and forums, customers vented that Salesforce seemed to be monetizing AI aggressively, rather than rewarding existing customers with value-adds. “It always felt like a black box – hard to predict, hard to explain to stakeholders,” said one product officer of the original model[10].
Salesforce’s pivot in progress: A conceptual illustration depicts the shift from charging $2 per conversation (speech bubbles) to a granular Flex Credit system (coins flowing into a “Flex Wallet”), signifying the pricing model change announced in 2025[11][12].
The numbers tell the story as well. By May 2025, Salesforce disclosed that only ~8,000 of its 150,000+ customers had started leveraging Agentforce[13]. Despite blanket marketing of Agentforce as the next big thing, adoption was stuck in the single digits percentage-wise. Price was not the only barrier (AI use cases themselves were nascent), but it was a major one. In a April 2025 virtual event, experts noted “cost is one of the major impediments to the wider adoption of Agentforce”[14]. The “$2 per conversation” model drew extensive criticism from practitioners, with heavy adoption resistance particularly among mid-market organizations[15]. Many such customers opted to delay AI agent rollouts or use cheaper “overlay” AI solutions from other vendors that could plug into Salesforce data[16]. In short, Salesforce’s pricing strategy was inadvertently driving some users to seek alternatives, exactly what Salesforce feared. (The company has been candid that it views autonomous agents as an existential battleground – if Salesforce isn’t the platform for them, it risks being relegated to just a data provider for others’ AI[17].)
By the end of 2024, the message was loud and clear: Salesforce’s initial pricing didn’t align with customer expectations or perceived value. What was intended as a simple usage fee instead positioned Agentforce as an unpredictable, potentially expensive add-on. For an AI platform still proving its worth, that positioning was fatal to uptake. As one SaaS consultant summarized, “They [Salesforce] don’t know how to price these agents yet, and nobody wants to hand out a blank check – they want caps and predictable ROI.”[5] Facing this feedback, Salesforce had little choice but to revisit its strategy.
Mid-2025: Enter Flex Credits – Salesforce Overhauls Pricing to Salvage Adoption
On May 15, 2025, Salesforce announced a major do-over of Agentforce pricing[18]. In a press release tellingly titled “Salesforce Introduces New Flexible Agentforce Pricing to Accelerate the Digital Labor Revolution,” the company rolled out a trio of changes aimed at “lowering the barrier to entry” and addressing the very complaints users had raised[19][20]:
- Flex Credits (Usage-Based Actions): Replacing the blunt per-conversation charge, Salesforce introduced Flex Credits, a more granular consumption model charging $0.10 per “action” rather than $2 per conversation[12]. In practice, 1 Agentforce action (e.g. looking up a record, sending an email, summarizing a case) consumes 20 credits, and credits are sold in packs of 100,000 for $500[12]. This works out to $0.10 per discrete AI task. The key difference: customers now pay for what the agent actually does, not for an abstract “conversation” unit. As Salesforce put it, “Flex Credits ensure you only pay for the exact actions Agentforce performs – whether that’s updating customer records, automating workflows, or resolving cases.”[21] In theory, trivial or failed interactions would incur minimal cost (since few actions executed), whereas complex successful sessions might use multiple actions and cost more than $2. Low-value conversations that lead nowhere no longer come with a full $2 cost – you pay perhaps $0.10 or $0.20 if the AI only attempted one or two actions before stopping[22]. Conversely, a very involved interaction that actually solves an issue could cost a bit more than $2, but presumably you got value from those multiple actions. This was Salesforce’s attempt to align price with outcomes and value delivered rather than flat interactions.
- Flex Agreement (Hybrid Credits/Seats Contract): Acknowledging that enterprise buyers crave cost predictability, Salesforce introduced a Flex Agreement model allowing customers to convert unused user licenses into credits (and vice versa) as needs change[23]. This effectively created a hybrid enterprise license: companies could commit a certain budget to Salesforce that could be flexibly allocated between traditional seat licenses and Agentforce consumption. If, say, an organization had paid for 100 support agent licenses but later automated some functions with AI, they could swap some of that investment into Flex Credits to cover Agentforce actions instead of “wasting” it on idle human licenses. And if their AI usage didn’t pan out, they could convert credits back to user seats. This transferability was meant to “provide peace of mind” that investing in AI wouldn’t lock customers into one format[24][25]. It’s essentially an Agentforce-specific enterprise license agreement (some dubbed it the “Agentic ELA”). The subtext: Salesforce recognized customers were hesitant to commit budget to AI without escape hatches. Flex Agreement gave CFOs an option to “reallocate spend without getting trapped”, as one analyst observed[26].
- Per-User Licensing Options (Unlimited Internal Use): Salesforce also previewed new Agentforce user licenses and add-ons that would provide unmetered agent usage for a fixed per-user-per-month price[27]. These were targeted for internal/employee-facing AI agents. For example, in Summer 2025 Salesforce launched Agentforce add-on licenses for Sales Cloud and Service Cloud at $125 per user/month (and industry cloud add-ons at $150) that allow a company’s employees to use Agentforce capabilities with no cap on actions[28][29]. They also rolled out premium “Agentforce 1” editions of core clouds (Sales, Service, Field Service) at around $550 per user/month which bundle the Agentforce add-on and a large block of credits (e.g. 1 million Flex Credits per org/year)[30]. In other words, Salesforce began packaging its AI into high-end subscription tiers. And for organizations that didn’t want to buy a whole cloud license, a standalone Agentforce User License was offered at $5 per user/month (requires purchasing credits for the actual usage) just to grant employees access to the AI features[31][32]. This move was a nod to classic SaaS pricing – an unlimited usage plan – to alleviate the “metering anxiety.” An exec in charge of the rollout noted that “CIOs don’t want their staff worrying about how much it will cost to use their AI assistant, and they don’t want surprises at month-end. This per-user unlimited option will kick-start some projects.”[33] Indeed, the appeal of a flat rate for internal use was clear: it put a ceiling on costs and simplified ROI calculations, even if the price ($125+ per seat) was not trivial. Salesforce essentially conceded that one size (usage pricing) would not fit all – enterprises demanded choice, so now “the ‘old’ AI pricing models still exist, but the new unlimited per-employee model makes customers more comfortable that they won’t end up with a 10x budget blowout.”[34][35]
Salesforce’s May 2025 announcement was therefore a comprehensive pivot to a “choice and flexibility” positioning. They explicitly stated that organizations were “seeking a pricing model aligned to how AI agents deliver business outcomes”, and that these changes were “designed to unlock AI adoption at scale” by making costs more controllable[36][37]. Notably, Salesforce even cited external validation: 90% of CIOs say managing AI costs is limiting value[20]. In other words, the company publicly acknowledged that its initial model wasn’t meeting the market’s need for cost management. The new flex pricing was marketed as Salesforce “listening to the ecosystem” and acting on feedback, a narrative echoed by community voices: “It feels as though Salesforce have really listened… the old model always felt like a black box; Flex Credits make it easier.”[10][38] Many in the Salesforce community welcomed the shift. The increased granularity and the safety net of license conversion were praised as “hugely beneficial” and “a huge improvement” for those piloting AI[39]. One Salesforce MVP noted the new model “is more predictable than the previous conversation-based charging, and it’s tied to adding value – you get charged if Agentforce executes an action, which means it’s helping the customer.”[40][41] This sentiment – charging for actions done rather than just talk – was exactly what Salesforce hoped to foster.
However, the story doesn’t end with happy customers and rapid adoption. The Flex Credits overhaul solved some problems but created new complexities. Enthusiasm was “widespread, but there [were] still many existing concerns” after the announcement[42]. Chief among them:
- Forecasting and Overhead: For all its tie to outcomes, the new model is still consumption-based at its core, which means usage must be monitored and forecasted. Instead of counting conversations, customers now had to count potentially millions of micro-actions. This could introduce significant administrative overhead and uncertainty[43][44]. An enterprise tech CEO quipped, “This pricing schema reminds me of my 1991 cell phone data plan, where every month’s bill was an adventure into the great unknown.”[45] The analogy is apt: early cellphone users were charged per minute or per text, often resulting in surprise bills due to complex unit accounting – exactly the scenario buyers wanted to avoid with AI. While Salesforce did roll out a Digital Wallet dashboard for real-time usage tracking[46][47], skeptics noted that without strict contractual safeguards like volume discounts, usage caps, and clear definitions of ‘actions,’ costs could still spiral unexpectedly[48]. In LinkedIn discussions, many predicted that 2026 would bring a backlash against opaque credit-based pricing as customers struggle with budgeting for “AI credits” they don’t fully understand[34][49]. In fact, one operations lead commented: “I never know how much it will really cost me, how much to budget, or how credits are being calculated – I do not need more ambiguity in my life.” (Ouch.)
- Definition of “Action” and Fairness: Salesforce defined an “action” as a discrete task performed by the AI (e.g. updating a record, generating a reply)[50][51]. But some users observed fine print: if an action consumed over 10k tokens of AI processing, it might count as two actions (essentially an overage for very large operations)[44]. This raised concerns about what truly constitutes one action and whether those units correlate to value. For instance, an Agentforce “action” could be something minor like logging an email – not necessarily a business outcome. “Taking an action on the platform – updating a record, etc. – is not necessarily a value to the business, and not all actions provide equal value,” noted one industry expert, who felt the model, while improved, was still not a perfect measure of value delivered[52]. Another commentator was more blunt, calling the new pricing “nonsense on legs” and “a bureaucratic lock-in”, arguing that the ROI doesn’t compute versus alternative AI solutions outside the Salesforce ecosystem[53]. In their view, Salesforce’s model (even with credits) still locked customers into paying for many small tasks, whereas third-party AI services or open-source models might achieve the same outcomes more cheaply. This criticism hints at a platform positioning issue: if customers believe they can replicate Agentforce’s capabilities with external AI at lower cost, Salesforce risks losing on its promise of being the unified platform for AI.
- Complexity vs. Simplicity: Ironically, what began as a single metric ($2/convo) had now morphed into a menu of pricing options: conversation-based, action-based credits (with three payment models: prepaid, pay-as-you-go, pre-commit[54][55]), per-user add-ons, and hybrid agreements. On one hand, this flexibility gave customers choices to fit their preference – something Salesforce emphasized as a positive[35]. On the other hand, it introduced significant complexity into sales conversations. A buyer now has to evaluate: do we gamble on pure usage to possibly save money? Do we lock in a higher fixed cost for peace of mind? Do we buy credits upfront or pay as we go? Such decisions are non-trivial, especially when the technology’s utilization is still uncertain. Some analysts cynically noted that this plethora of options was Salesforce “throwing everything at the wall” to get customers on board – a sign that internally, there was no clear consensus on the single best model, so they went with all of the above. Indeed, the head of pricing at OpenView Venture (Kyle Poyar) observed, “It’s not about one pricing model being better; it’s about providing choice. Salesforce’s old models still exist. But the AELA [unlimited license] makes customers more comfortable there’s a ceiling on pricing… In fact, Salesforce now has different flavors of AI credits: pre-purchase, pay-as-you-go, pre-commit.”[34][56] Choice and agility sound great, but choice can also breed confusion if not well guided. This is a classic case of design-by-committee in pricing: rather than kill their darlings, Salesforce kept them all alive.
Despite these caveats, the net effect of the 2025 changes was positive for many customers. The barrier to experiment with Agentforce lowered – one could start with $500 of credits and see results, rather than fearing an open-ended bill. As a result, by late 2025, Salesforce was able to point to growing adoption and a pipeline of larger AI deals. Internally, the company touted that customers like Adecco, OpenTable, and others were “scaling smarter and driving faster outcomes with Agentforce,” thanks in part to more flexible pricing[57][58]. Salesforce’s Q3 2025 financials showed AI was contributing to revenue growth, and they projected a “sharp jump in monetization on new AI deals” in 2026[59][60]. But the improved uptake likely came at a cost: Salesforce had to permit steep discounts and generous terms to get enterprises committed. (Indeed, industry watchers noted that the company, which “just doesn’t discount ever,” was showing atypical flexibility to make Agentforce cheaper and more accessible[61].) It’s apparent that Salesforce’s priority shifted to driving adoption – even if it meant lower margins or more complex deal structures – to avoid being left behind in the AI platform race[62].
Late 2025–Early 2026: Per-User Pricing and the Push for Big Deals
By the end of 2025, Salesforce’s pricing evolution reached its logical culmination: a full embrace of seat-based licensing alongside usage credits. In September 2025 (around Dreamforce), Salesforce made generally available the Agentforce add-ons and “1 Editions” that had been in preview. Essentially, Agentforce could now be sold just like any other cloud product: at a (hefty) per user subscription, often bundled into existing product packages. For example, a large enterprise could upgrade to Salesforce’s “Unlimited Plus” CRM edition (hypothetical name) that includes unlimited Agentforce usage for all licensed users, rather than buying AI in bits and pieces. This mirrors how Microsoft sells its AI offerings – e.g. Copilot features bundled into premium Office 365 plans.
This shift represented a strategic positioning move: rather than treat Agentforce as a separate usage-based platform, Salesforce started to treat it as an integrated capability of its core platform, worthy of premium pricing tiers. In some sales pitches, the company even began leading with an “AI + data + CRM” platform message, where the AI is part of the suite rather than an add-on line item. The Agentforce 1 Enterprise Edition at ~$550/user (which includes the AI and large data credits) is a prime example – it’s effectively a new top-of-the-line Salesforce license[30]. To a customer, that feels like buying a Salesforce cloud seat, not buying an AI bot usage bundle. This is classic packaging as pricing: by repackaging Agentforce into user-based editions, Salesforce changed the value perception and simplified pricing discussions. (As one pricing advisor often says, “in software, packaging is pricing” – the way the offer is structured determines how you capture value[63]. Salesforce’s re-bundling of Agentforce validates this: they had to adjust the packaging to get the pricing right.)
From Salesforce’s perspective, offering unlimited enterprise licenses was also a competitive response and landgrab tactic. Industry analysts predicted that 2025’s flurry of AI trials would turn into 2026’s multi-million-dollar AI commitments as companies chose platforms to standardize on[64][65]. By providing an “all-you-can-eat” pricing option, Salesforce positioned itself to snag those big deals. An unlimited model let them say: “Don’t worry about usage, just pay this premium per user and deploy AI freely across your org.” This not only alleviates buyer fears, but also deepens Salesforce’s lock-in – if a customer is paying for unlimited Agentforce, they have every reason to use it broadly (and less reason to try outside AI solutions). Kyle Poyar pointed out the parallel to Microsoft’s strategy: it’s reminiscent of bundling Teams for “free” with Office – making it a no-brainer to adopt since it’s already paid for[66][67]. Similarly, Salesforce’s AELA (Agentic Enterprise License Agreement) aimed to make Agentforce ubiquitous within a customer’s environment, thus crowding out third-party AI. Pricing, again, was being used as a weapon for platform positioning.
Customer reception to these late-2025 developments was cautiously optimistic. Large enterprises liked the idea of an “AI ceiling” on costs. After living through years of cloud spend unpredictability, locking in a rate (even a high one) for unlimited AI usage had its appeal. It shifted the conversation from “How much will it cost me if usage spikes?” to “If I invest $X, what value can I drive with unlimited AI?” – a far more positive framing. Salesforce’s sales teams reportedly found it easier to sell the unlimited add-on than to negotiate detailed credit forecasts. One Salesforce sales rep shared that many CIOs chose the $125/user add-on for key departments simply to avoid the headache of metering – they could then encourage their teams to experiment freely, which in turn would (hopefully) lead to faster ROI demonstration. This aligns with a broader truth in enterprise software: predictable budgeting often trumps absolute cost. A slightly higher known cost is preferable to a maybe-cheaper-but-uncertain cost. Salesforce essentially conceded this reality by introducing predictability, after initially offering only variability.
That said, smaller customers and cost-conscious buyers remained wary. For them, $125 per user (on top of existing Salesforce licenses) is steep, especially if their AI usage might have only amounted to, say, $50/month in Flex credits. These customers faced a tricky choice: stick to granular pay-per-use and diligently manage it (with the risk of overage surprises), or overpay for a flat license to avoid surprises. Salesforce’s plethora of pricing models now allowed each customer segment to self-select, but it also meant the onus was on the customer to choose wisely. Negotiation advisors like UpperEdge warned clients to “push Salesforce for clear contract definitions” and “volume discounting” no matter the model[48][68]. They advised piloting with Flex Credits but building in conversion rights to unlimited if usage crosses a threshold – essentially hedging bets. In effect, by early 2026, savvy customers were structuring deals that combined elements of all models: negotiated discounted credit rates (to cap worst-case costs), plus the ability to flip to per-user if needed, plus clauses to swap back if headcount changed. It’s no surprise one observer labeled the whole scheme “bureaucratic”[53] – it required careful contract design to truly make it customer-friendly.
Meanwhile, competitors continued to capitalize on any Salesforce missteps. Intercom’s Fin, for example, gained praise for its outcome-based pricing ($0.99 per resolution), which is “only charging when Fin achieves the outcome you care about – a resolved conversation”[8]. Intercom also smartly allowed usage caps and alerts to be set by customers so they wouldn’t overspend unexpectedly[69]. Freshworks and Zendesk rolled out their own AI assistants with relatively lower price points or included usage quotas to entice adoption[70]. And Microsoft and Google, as mentioned, leveraged bundling and their cloud scale to keep prices comparatively low (e.g. Google’s Vertex AI agents at ~$0.012 per query[71]). The broader context was rapid AI cost deflation – the cost of underlying AI model inference was plummeting (OpenAI, for instance, cut GPT-4’s price by 2/3 in 2023, and model training costs were dropping 70%+ per year)[72]. This put pressure on pricing. Customers became aware that running an AI conversation likely costs only fractions of a penny in compute – so paying $2 or even $0.10 per action started to feel rich. Salesforce had to convince customers that they weren’t just paying for raw AI, but for enterprise-grade orchestration, data integration, security, etc. Nonetheless, the macro trend of AI cost compression and general software deflation made buyers more aggressive in negotiations. One dataset showed that software as a category had seen prices drop about 40% in real terms from 2015 to 2023[73], making clients expect more for less each year. By 2025, many companies were in belt-tightening mode (after 2020-2021’s free spending), with boards scrutinizing ROI on AI projects rather than green-lighting them on hype. Forrester Research even predicted that a quarter of planned 2026 AI spending might be delayed to 2027 due to economic caution, citing “inflated promises” and the need for disciplined economics[74]. All of this created a challenging backdrop for Salesforce’s pricing experiment. It’s likely a big reason why Salesforce pivoted so quickly in 2025 – the market simply did not have appetite for another unpredictable cost center.
Pricing Is Positioning: Why Salesforce’s Approach Fell Short (and What We Can Learn)
Examining Salesforce’s Agentforce pricing saga, it’s evident that missteps in pricing strategy directly undermined the product’s positioning. Marc Benioff pitched Agentforce as the centerpiece of Salesforce’s future – an AI platform deeply embedded in work processes. But the initial pricing positioned it almost like a novel utility service one pays for by the sip, separate from the trusted Salesforce platform value. This was poor alignment. Customers didn’t view “AI conversations” as something independent – they saw AI as an enhancement to the CRM they were already buying. Thus, charging per conversation felt like double-charging or nickel-and-diming for what many assumed would eventually be table stakes functionality. In contrast, bundling Agentforce into editions (the eventual strategy) repositioned it as a feature of the overall solution (albeit an expensive one).
What internal dynamics led Salesforce down the wrong path initially? We can conjecture a few factors:
- Over-indexing on Tech Value vs. Customer Perception: It’s possible Salesforce believed the hype that an AI agent is so valuable (replacing human work, etc.) that customers would gladly pay per use. $2 per conversation might have been benchmarked against the cost of a human agent chat, which can be significantly higher. However, Salesforce misread how customers perceive AI value. In early deployments, AI agents are augmenting humans, not wholesale replacing call center staff. Customers saw Agentforce as a helpful feature, not a billable service on its own. By pricing it separately and usage-based, Salesforce signaled “this is a standalone service you pay extra for,” undermining the narrative that AI was simply the next evolution of their platform. In essence, the pricing sent a conflicting message to the positioning. As Ajit Gohil (a pricing expert) might say, you have to “balance your needs with the buyer’s needs” when choosing a pricing metric[75]. Salesforce’s need was to monetize AI’s technical cost and value; the buyers’ need was predictability and clarity. Initially, that balance was off.
- Internal Goals and Siloes: One can imagine internal stakeholders had different agendas. The product team, aware of the high compute costs of generative AI (and the fact that AI usage could reduce the need for some human user licenses over time), likely pushed for a usage-based model to protect revenue. They know that “with more AI, the users are reducing… and cost of goods sold is higher, so you can’t just do per-user pricing or you’d start losing money”[76]. On the other side, the sales organization knows customers are used to seat licenses and that anything too new or complicated could slow deals. The result of these forces was the Frankensteinian hybrid we saw by 2025 – essentially all models at once, which smacks of design-by-committee. The fact that Salesforce had to introduce conversion rights (Flex Agreement) implies they weren’t confident even internally which model customers would ultimately prefer, so they allowed switching. Bureaucracy and the pressure to satisfy both “growth via AI revenue” and “keep customers happy” likely led to the initial compromise of $2/conversation (simple for sales to quote, theoretically capturing value for product). When that blew up, the subsequent solution was more complex but also more accommodating – indicating Salesforce’s internal priority shifted to “whatever it takes to get customers using this.” It’s telling that by late 2025, the company was willing to raise core product prices and blend AI into enterprise agreements rather than stick to its initial guns[59][77]. This U-turn suggests the top brass realized that forcing a new pricing paradigm could backfire if customers instead slowed or limited adoption.
- Speed vs. Strategy: Salesforce rolled out Agentforce (and its pricing) relatively quickly in response to the 2023 generative AI wave. It’s possible the pricing wasn’t fully tested with customers in pilot programs – or if feedback was gathered, it was overridden by a desire to announce something “simple” at Dreamforce. The execution issues (like poor communication of what counts as a conversation) hint at a rushed job. When introducing a radical new metric, clear definitions and education are critical. Salesforce’s initial documentation left too many gaps (hence the rampant confusion). Later, they had to clarify every detail – down to examples of what one action costs, how many credits typical tasks consume, etc., even publishing an Agentforce ROI calculator[78]. The lesson is that if you’re innovating in pricing, you must also innovate in explaining and supporting that model. Otherwise, customers assume the worst (e.g. “Salesforce is trying to trick us into overage charges”).
- Admitting Mistakes and Adapting: To Salesforce’s credit, they did adjust course relatively rapidly (within 6–8 months of GA). This responsiveness likely saved Agentforce from a worse fate (irrelevance). Many companies fear that changing pricing midstream shows weakness. Salesforce instead framed it as listening to customers and adding flexibility, which likely preserved goodwill. One Salesforce architect noted “it’s encouraging to see Salesforce adjust their pricing strategy to what is happening in the marketplace… The ‘bridge too far’ just got a little shorter for some folks.”[61] In other words, lowering the barrier, even if it meant admitting the first attempt overshot, was the right move. The reality is that in SaaS, optimal pricing often requires iteration (market conditions change, usage patterns emerge, etc.) – but those iterations must be executed carefully to avoid whiplash. Salesforce managed to navigate this by not removing the old model outright (so as not to strand any early customers on conversations) but layering new options on top. While that added complexity, it avoided directly breaking promises. Going forward, Salesforce will need to streamline this menu (perhaps deprecate the per-conversation SKU eventually) once customers settle into the new normal.
What are the actionable lessons from Salesforce’s experience for other SaaS leaders? A few stand out:
- Align Pricing Metrics with Customer Value and Budgeting Norms: If your chosen unit of pricing isn’t something customers intuitively grasp or use to measure success, expect resistance. Salesforce’s “conversation” was too abstract and didn’t map cleanly to outcomes (unlike Intercom’s “resolution” metric which directly ties to a successful support outcome). Moreover, enterprise buyers budget annually for known quantities (users, endpoints, etc.). If you introduce a new variable expense, provide guardrails and be prepared to explain how it correlates to value. A good practice is to test pricing with design partners – would they sign a deal under this model and can they forecast the cost easily? If not, redesign it. As one expert succinctly put it, “when pricing works well…the customer is not questioning the metric”[3]. Strive for that state.
- Complexity is a Last Resort – Aim for Simplicity First: While Salesforce eventually offered “choice,” the simplest model (pay per conversation) turned out to be the wrong metric. The more effective simplicity was perhaps achieved by Intercom: charge per resolution – one fee, clearly defined outcome. Or even Salesforce’s own later approach: $125 per user for unlimited internal use – one number to budget. These are easy to communicate. If you can anchor your pricing on a single, easy-to-calc metric that aligns with value, do so. Complexity (tiered plans, hybrid models, credit systems) should be layered in only if absolutely necessary (e.g. to accommodate widely varying usage or to manage cost of goods). Salesforce jumped to a complex usage scheme perhaps too soon, then had to backpedal to add simpler flat options. A more cautious rollout (e.g. offering both a flat and usage model from the start as a trial) might have been better.
- Provide Predictability (Especially in Uncertain Economic Times): One consistent theme is enterprise customers hate surprises. This is amplified when budgets are tight. During 2024–25’s economic climate, CFOs were scrutinizing every tech expense. A pricing model that could yield a “surprise” invoice was dead on arrival for many. Salesforce’s introduction of caps, conversions, and unlimited plans was essentially about de-risking the spend. SaaS companies should consider offering “safety valves” – whether it’s usage caps, fixed-price tiers, or the ability to true-up/true-down. These features can make buyers significantly more comfortable adopting new functionality. In Salesforce’s case, just the psychological benefit of knowing there’s a ceiling on AI cost (even if high) was key to unlocking larger deployments[35]. Early on, Salesforce had no ceiling and it hurt them. Now, they’ve put a roof and customers can enter the house without fear.
- Don’t Underestimate Change Management (Both Customer and Internal): Rolling out a new pricing model is almost as tricky as rolling out a new product. Internally, Salesforce’s sales reps had to learn to sell Flex Credits (essentially a usage-based “cloud” SKU) which was outside their traditional playbook. There were reports of reps initially struggling to explain the value of $2 conversations or how to size deals – leading to inconsistent messaging. Only after Salesforce equipped them with better calculators and the Flex Agreement narrative could reps turn pricing into a positive story (“look how flexible we are!”). Externally, customers needed education and tools (e.g. the Digital Wallet, usage reports, ROI calculators[78]). The takeaway: if you introduce a novel pricing scheme, invest in education, tools, and support for it. Treat pricing as part of the product experience. If it’s too byzantine to understand or manage, that is a product flaw. In retrospect, Salesforce’s Digital Wallet and estimator tools should have been available at launch, not months later.
- Keep an Eye on Macroeconomics and Competitors: Pricing does not exist in a vacuum. Salesforce launched Agentforce pricing in a frothy moment of AI hype, perhaps thinking customers would pay anything for generative AI magic. But the macroeconomic reality (post-boom cost-cutting and ROI focus) meant customers were far more price-sensitive and skeptical. At the same time, competitors (big and small) were pricing aggressively to capture share. When OpenAI is effectively selling an AI conversation for fractions of a penny (via API), and your competitor offers outcomes at $0.99, pricing high and vague is unwise. Smart SaaS leaders continuously benchmark their pricing against both direct competitors and the broader “cost of DIY”. If a customer can theoretically plug in an API or open-source model and achieve similar results, your pricing must convincingly account for the extra value you provide (integration, security, lower maintenance, etc.). In Salesforce’s case, the value prop is that Agentforce is natively integrated into Customer 360 (data, automation, security controls). The pricing strategy needed to reinforce that (by encouraging integration via flat fees perhaps) rather than challenge it (by imposing separate transactional fees). Ultimately, Salesforce adjusted pricing to reinforce its platform position – bundling AI into the platform offering – but it took a circuitous route to get there.
- Pricing Iteration is OK (but Frame It Right): Few companies get pricing perfect out of the gate, especially with new tech. Salesforce’s error was not that they had to change pricing – it was launching with a misaligned model in the first place. However, they mitigated damage by rapidly iterating and framing changes as customer-centric improvements. SaaS leaders should similarly view pricing as an adaptive strategy. Gather feedback, look at adoption and churn signals, and be willing to tweak. The key is to explain changes as enhancements or new options (as opposed to flip-flopping policy). Salesforce, for instance, never explicitly said “we messed up”; they said “we’re introducing new packaging and pricing built to simplify and accelerate your AI journey”[79]. The tone was forward-looking, not apologetic, which helped maintain confidence. Still, an important caveat: frequent or radical pricing changes can erode trust, so find a balance. In Salesforce’s case, two major changes in about a year was probably the maximum they could do without seriously rattling customers. Now they’ll likely hold this course for a while.
In conclusion, Salesforce’s Agentforce pricing odyssey underscores that pricing is a powerful signal of value and positioning. A misaligned pricing model can confuse customers about what your product really is and for whom. Salesforce’s initial approach made Agentforce look like an experimental addon that punished you if you used it too much – hardly the message a company wants for its flagship AI offering. Through customer feedback (and some humbling market realities), they learned that alignment and flexibility matter more than trying to maximize short-term revenue with a novel metric. Agentforce’s repositioning as both a usage-based and a user-based product reflects Salesforce’s recognition that platform adoption was the priority; revenue will follow if customers actually embrace the AI. In the broader context of SaaS, as software faces deflationary pressures and buyers demand more value for less, companies must be extremely deliberate in choosing pricing that aligns with how customers perceive value. The Salesforce case also highlights the growing trend of hybrid pricing models – mixing subscriptions with consumption – which can be powerful if done right, but perilous if done clumsily. As we head further into an AI-driven software era, finding that sweet spot of fair, transparent, and value-based pricing will be a critical differentiator.
Salesforce’s journey may have been bumpy, but it provides a real-world course on pricing strategy in practice. In the end, the hope is that Agentforce’s pricing is now “working within the engine that’s been created”, with customers no longer questioning the metric and feeling the product “fits their needs”[3]. If not, Salesforce will undoubtedly hear about it – and the cycle of iteration will continue. For the rest of us, the takeaway is simple: Position your pricing as carefully as your product – your platform’s success depends on it.[80][53]
Sources:
· Salesforce Press Release – “Salesforce Unveils Agentforce – What AI Was Meant to Be”, Sep 2024 (pricing at launch)[1]
· Salesforce News – “Salesforce Introduces New Flexible Agentforce Pricing…”, May 15, 2025 (Flex Credits, Flex Agreement details)[21][81]
· SalesforceBen – “Understanding Common Agentforce Pain Points…”, May 2025 (critiques of $2 conv model, introduction of Flex)[7][22]
· SalesforceBen – “How the Ecosystem Reacted to New Agentforce Pricing”, May 2025 (community quotes on old vs new model)[6][40]
· CX Today – “Salesforce Makes Changes to Agentforce Pricing (Again!)”, Aug 21, 2025 (overview of pay-as-you-go, pre-commit options and commentary on Salesforce’s motives)[82][83]
· SalesforceDevOps.net – “Salesforce Shifts to Agentforce Flex Credits: Addressing Adoption Barriers…”, May 2025 (industry analysis of pricing shift, competitive context)[11][15]
· UpperEdge – “Salesforce’s New Agentforce Pricing: What Customers Should Know”, May 22, 2025 (negotiation insight, need for safeguards)[2][48]
· Intercom Help Center – “Fin AI Agent Resolutions” (Intercom’s pricing model at $0.99 per resolved conversation)[8]
· Complete AI Training – “Salesforce hikes AI agent prices, blends seat and usage pricing…”, Dec 5, 2025 (discussion of Agentic ELA, customer desire for predictability)[84][60]
· LinkedIn (Kyle Poyar) – “Salesforce’s AI Pricing Shift: Choice, Flexibility, and Unlimited Options”, Jan 2026 (analysis of Salesforce’s multi-model strategy and 2026 outlook)[34][35]
· SalesforceBen – “The Positives and Concerns Around Agentforce’s Pricing Model” (expert quotes: “black box” model, “1991 cell phone plan” analogy, fairness concerns)[85][80]
· User POV Webinar Transcript (Monetize.ly) – insights on software deflation and pricing alignment (e.g. packaging is pricing; aligning pricing metric to how buyers budget)[3][76].
[1] Salesforce Unveils Agentforce–What AI Was Meant to Be - Salesforce
https://www.salesforce.com/news/press-releases/2024/09/12/agentforce-announcement/
[2] [24] [48] [68] Salesforce’s New Agentforce Pricing: What Customers Should Know - UpperEdge
https://upperedge.com/salesforce/salesforces-new-agentforce-pricing-what-customers-should-know/
[3] [63] [72] [73] [75] [76] Key Youtube Webinars with our POV.docx
file://file_0000000036d871fda43191452e650c9a
[4] [11] [12] [15] [16] [71] Salesforce Shifts to Agentforce Flex Credits: Addressing Adoption Barriers, But Challenges Remain - SalesforceDevops.net
https://salesforcedevops.net/index.php/2025/05/15/salesforce-shifts-to-agentforce-flex-credits/
[5] [6] [10] [14] [33] [38] [39] [40] [41] [42] [43] [44] [45] [52] [53] [61] [80] [85] How the Ecosystem Reacted to Salesforce’s New Agentforce Pricing | Salesforce Ben
https://www.salesforceben.com/how-the-ecosystem-reacted-to-salesforces-new-agentforce-pricing/
[7] [9] [17] [22] [46] [47] [62] Understanding Common Agentforce Pain Points and How Salesforce Addresses Them | Salesforce Ben
[8] [69] Fin AI Agent resolutions | Intercom Help
https://www.intercom.com/help/en/articles/8205718-fin-ai-agent-resolutions
[13] [19] [54] [55] [82] [83] Salesforce Makes Changes to Its Agentforce Pricing Model (Again!) - CX Today
https://www.cxtoday.com/crm/salesforce-makes-changes-to-its-agentforce-pricing-model-again/
[18] [20] [21] [23] [27] [36] [37] [57] [58] [78] [81] Salesforce Introduces New Flexible Agentforce Pricing - Salesforce
https://www.salesforce.com/news/press-releases/2025/05/15/agentforce-flexible-pricing-news/
[25] [26] [59] [60] [74] [77] [84] Salesforce hikes AI agent prices, blends seat and usage pricing, promises 3-10x value amid Forrester skepticism
https://completeaitraining.com/news/salesforce-hikes-ai-agent-prices-blends-seat-and-usage/
[28] [29] [30] [31] [32] [50] [51] Salesforce Agentforce Pricing | Salesforce
https://www.salesforce.com/agentforce/pricing/
[34] [35] [49] [56] [64] [65] [66] [67] Salesforce's AI Pricing Shift: Choice, Flexibility, and Unlimited Options | Kyle Poyar posted on the topic | LinkedIn
[70] Intercom Fin 2 Pricing & Top 10 Better Alternatives - GPTBots.ai
https://www.gptbots.ai/blog/intercom-fin-pricing
[79] Salesforce Announces Pricing Update
https://www.salesforce.com/news/stories/pricing-update-2025/