Salesforce closed out July 2026 with a split screen. On one side: Agentforce Help Agent and Agentforce Customer Service Portal went generally available, backed by a pay-per-resolution pricing model that charges $2 only when the AI actually closes a case—and Agentforce plus Data 360 annual recurring revenue crossed $1.2 billion in Q1 of fiscal 2027, up roughly 205% year over year. On the other side: KeyBanc Capital Markets and Bernstein, two of Wall Street’s most-followed research shops, published back-to-back notes saying customers are telling them Agentforce “just isn’t there” yet, and that more CIOs plan to cut Salesforce spending over the next year than plan to increase it. Bernstein downgraded the stock. Both things are true at once, and that tension is exactly what CRM buyers evaluating any vendor’s AI agent pitch right now need to understand.
Key Takeaways
- Agentforce Help Agent and Agentforce Customer Service Portal reached general availability in July 2026, deployable in fewer than ten setup steps.
- Pricing is pay-per-resolution: a flat $2 per autonomous resolution, sold in pre-purchased packets of at least 1,000, with no charge if a customer asks for a human or walks away unresolved.
- Salesforce reports Agentforce plus Data 360 ARR at roughly $3.4 billion combined, growing over 200% year over year, with more than half of bookings coming from existing customers expanding deployments.
- KeyBanc and Bernstein both published analyst notes in July 2026 citing weak customer feedback, messy underlying data, and slower-than-expected proof-of-concept-to-deal conversion.
- Bernstein downgraded Salesforce stock, arguing consumption-driven AI monetization will take longer than the market expects.
- Real customer deployments—Grupo Falabella, Engine, AAA Washington—show genuine automation gains, but all three still lean on human-in-the-loop design rather than full autonomy.
- The lesson for buyers: outcome-based pricing lowers the risk of a bad pilot, but it doesn’t eliminate the data-readiness work that determines whether the agent performs at all.
What Actually Shipped: Help Agent and Pay-Per-Resolution Pricing
Agentforce Help Agent is Salesforce’s attempt to remove the two biggest objections to AI customer service agents: setup complexity and pricing risk. It’s designed to go live in minutes with a guided configuration flow, and instead of a seat license or a flat platform fee, Salesforce is charging per outcome. Customers pre-purchase a packet of resolutions (minimum 1,000) at a flat $2 each. A resolution only counts if the agent handles the interaction autonomously from start to finish. If the customer escalates to a human, or leaves without a fix, Salesforce doesn’t charge for it. Messaging sessions run up to two hours; voice sessions are capped at ten minutes.
This is a meaningful shift from the credit-based consumption pricing Salesforce introduced with the original Agentforce launch. It’s also a direct response to the pricing objection Bernstein flagged in its own survey work: the majority of customers it polled said they were unwilling to pay extra for AI capability bolted onto a CRM they already license. Pay-per-resolution is Salesforce’s argument that the pricing itself shouldn’t be the blocker—you only pay when the software does the job a human agent would otherwise have done.
The Adoption Numbers, Read Two Ways
Salesforce’s own figures are genuinely strong on paper. Agentforce ARR alone hit roughly $800 million by the end of fiscal 2026, then jumped to $1.2 billion in the first quarter of fiscal 2027—about 205% year-over-year growth. Combined with Data 360, the two AI-adjacent product lines are tracking near $3.4 billion in ARR, and Salesforce points out that more than half of that quarter’s bookings came from existing customers expanding deployments they’d already started, not from discounted new-logo pilots.
KeyBanc and Bernstein aren’t disputing those numbers directly—they’re questioning what’s underneath them. KeyBanc’s survey work found two consistent themes across the customers and partners it spoke with: internal data isn’t organized well enough to support meaningful AI work, and the product experience itself “just isn’t there” for a lot of the use cases customers actually want. The firm also noted that partners are only now starting to convert proof-of-concept engagements into real pipeline deals, months after many of those pilots began. And in KeyBanc’s CIO survey, more respondents said they expect to deprioritize Salesforce spending over the next twelve months than said they’d increase it.
Bernstein’s downgrade note struck a similar chord: consumption-driven monetization at Agentforce “will take longer than most expect,” and the product is likely to succeed mainly inside Salesforce’s existing CRM base rather than expanding into the broader enterprise workflows Salesforce has been pitching it for, given how many hyperscalers and SaaS vendors are shipping competing agent functionality of their own.
Salesforce’s rebuttal, through a company spokesperson, was that Agentforce is “the fastest-growing product in Salesforce history,” pointing to named customer wins at Engine, Grupo Falabella, and AAA Washington as evidence the deployments are real, not just pilots sitting on a shelf.
What the Real Deployments Actually Show
The named case studies are worth looking at directly, because they illustrate both the upside and the limits analysts are flagging.
Grupo Falabella rolled out Agentforce on WhatsApp in its Colombia operation and moved digital interaction share on that channel from under 50% to over 70% within three weeks of launch, autonomously resolving roughly 60% of WhatsApp inquiries—a real, measurable shift in channel mix, achieved in just over two months from kickoff to launch. Engine reports autonomous agents handling about 50% of customer support cases, including handling cancellations in seconds and giving employees always-on HR, IT, and finance answers inside Slack. AAA Washington frames its deployment explicitly as an “augmentation strategy”—the agent handles insurance recommendations and routine requests, but the design intent is to make human agents more effective, not to remove them.
None of these are full replacements for a support organization. They’re targeted deflection of a specific volume of routine, well-defined interactions, layered on top of human teams that still handle everything else. That’s a legitimate and valuable outcome—it’s just a narrower one than “autonomous AI customer service,” and it lines up with what Bernstein means when it says the product performs best inside familiar CRM workflows rather than as a general-purpose enterprise agent.
Why This Matters Beyond Salesforce
Every major CRM and ERP vendor is currently making some version of the same pitch: HubSpot with its Prospecting Agent and Breeze suite, Zoho with Zia’s autonomous “digital employees,” Microsoft with Dynamics 365 Sales and Service Agents, NetSuite with its EPM planning agents. Salesforce is simply the vendor far enough ahead in disclosure—and large enough in market cap—that analysts are scrutinizing its numbers in public. The gap this reveals applies across the category: headline ARR growth for an AI product tells you adoption is happening somewhere, but it doesn’t tell you whether it’s happening at the depth and reliability a buyer needs before they can retire the human process it’s supposed to replace.
Pay-per-resolution pricing is a genuinely useful innovation here, because it shifts risk away from the buyer. You’re not paying a platform fee for an agent that sits half-configured; you’re paying for outcomes it actually delivers. But it doesn’t remove the underlying prerequisite both analyst firms keep landing on: an agent can only resolve what your data lets it resolve. If your knowledge base is stale, your case history is fragmented across systems, or your product/service catalog isn’t structured cleanly, no pricing model fixes that. You’ll simply pay less for a lot of escalations instead of paying more for a lot of failed automation—which is a better failure mode, but still a failure mode.
Benefits and Challenges
The benefits are real and worth taking seriously: outcome-based pricing that removes most of the financial downside of trying an agent, genuine channel-shift results at reference customers, fast setup relative to a traditional bot-building project, and a credible roadmap of continued investment given Salesforce’s ARR trajectory. The challenges are equally real: the data-readiness gap analysts keep hearing about from customers, a still-unsettled question of how much of that ARR growth is expansion within an already-committed base versus genuinely new adoption, and a competitive field where every other major platform is racing to ship the same capability, which will compress differentiation over the next 12–18 months.
Implementation Best Practices and Common Mistakes
Organizations that get real value out of an agent like this tend to do a few things before they ever turn it on. They audit and consolidate the knowledge sources the agent will draw from—help center articles, macros, past case resolutions—rather than pointing it at whatever exists today and hoping. They start with a narrow, well-bounded case type (order status, cancellations, simple billing questions) instead of trying to cover the full support taxonomy on day one, which is exactly the pattern behind Falabella’s and Engine’s results. They set explicit escalation rules so the agent hands off cleanly rather than looping a frustrated customer, since under pay-per-resolution pricing a clean handoff costs nothing but a bad autonomous attempt that doesn’t fully resolve the issue still burns trust. And they treat the first 60–90 days as a measurement period, tracking resolution rate, escalation rate, and customer satisfaction on resolved cases specifically, rather than declaring success from raw deflection volume alone.
The most common mistake is the one both analyst notes point to directly: buying the platform before doing the data and process work the agent depends on, then blaming the product when it underperforms. The second most common mistake is treating pay-per-resolution as a free trial with no oversight—because you’re not charged for failed resolutions, it’s tempting to launch broadly and see what sticks, but a poorly scoped agent still damages customer experience even when it doesn’t damage your invoice.
| Consideration | Salesforce’s Position | Analyst Concern |
|---|---|---|
| Pricing risk | Pay-per-resolution removes cost for failed/escalated interactions | Doesn’t address whether customers see enough value to expand usage |
| Growth numbers | ~205% YoY Agentforce ARR growth, $1.2B run rate | Growth concentrated in expansions of existing accounts, not new logos |
| Product readiness | Points to named wins (Falabella, Engine, AAA) | Broader customer base reports data readiness and product gaps |
| Scope of impact | Positions Agentforce as an enterprise-wide agent layer | Bernstein expects success mainly within core CRM workflows |
CRM Experts Online’s Perspective
We implement CRM and ERP systems for a living, and the pattern in this story is one we see constantly, regardless of which vendor’s logo is on the AI feature. The technology usually isn’t the bottleneck—the data underneath it is. Before we’ll recommend a client turn on any autonomous agent, whether it’s Agentforce, Zia, or a Copilot Service Agent, we do a data and workflow audit first: is the knowledge source clean, is the case taxonomy consistent, are escalation paths actually defined, and is there a narrow enough first use case to prove value before expanding scope. Pay-per-resolution pricing is a good development for buyers because it lowers the cost of finding out the answer is “not yet”—but it’s still on you to do the readiness work that determines whether the answer is yes. If you’re evaluating Agentforce Help Agent, or weighing it against Zoho, HubSpot, or Microsoft’s equivalent, we’d rather help you scope a 90-day pilot against a single, well-defined case type than watch you buy a resolutions packet against a knowledge base that isn’t ready for it.
FAQ
What exactly does Agentforce Help Agent charge for? A flat $2 per autonomous resolution—an interaction the agent handles start to finish without human involvement and without the customer walking away unresolved. Escalations and abandoned sessions aren’t billed.
Do we need Service Cloud to use Help Agent? Salesforce positions it alongside the new Agentforce Customer Service Portal as part of its service agent lineup; existing Service Cloud customers are the primary near-term audience, though check current packaging with your account team since Salesforce pricing structures change frequently.
Is the skepticism from KeyBanc and Bernstein about the product or about Salesforce’s stock? Both. Their research is aimed at investors, but the underlying evidence—customer interviews about data readiness and product maturity—is directly relevant to any buyer evaluating the product, not just shareholders.
Does pay-per-resolution pricing mean there’s no risk in trying it? It removes most of the financial risk of a failed pilot, but not the time cost of preparing your data and defining escalation rules, or the customer experience risk of a poorly scoped agent going live too broadly.
How does this compare to what Zoho, HubSpot, or Microsoft are doing? All four are moving toward outcome- or resolution-based pricing models and autonomous service agents in 2026. The core evaluation criteria are the same across vendors: how clean is your underlying data, how narrow is your first use case, and how clearly are escalation paths defined.
What size of support volume justifies trying this? The clearest wins so far are organizations with high-volume, repetitive request types (order status, cancellations, common billing questions) on channels like WhatsApp or web chat—not necessarily raw company size.
Should we wait for the product to mature before piloting it? Not necessarily—the pay-per-resolution model is specifically designed to make an early, narrowly-scoped pilot low-risk. The mistake is piloting broadly before your data is ready, not piloting at all.
Conclusion
Salesforce’s ARR growth and Wall Street’s skepticism aren’t actually contradicting each other—they’re describing the same early-adoption phase from two different vantage points. Real customers are getting real automation value from Agentforce Help Agent, and real customers are also telling analysts their data isn’t ready and the product hasn’t matched the pitch yet. Both will keep being true until the vendor and the buyer side both close the readiness gap. If you’re weighing Agentforce—or any competing agent platform—against your own support or sales operation, the highest-leverage next step isn’t reading another vendor deck, it’s an honest audit of whether your data and workflows could support autonomous resolution today. Schedule a consultation with CRM Experts Online and we’ll help you scope a pilot that tells you the real answer before you commit budget to one.
Further Reading
- Salesforce: Agentforce Help Agent Announcement
- The Register: Salesforce’s Agentforce Isn’t Winning Over Clients, KeyBanc Analysts Claim
- Investing.com: Bernstein Downgrades Salesforce, Cites Lack of Evidence for Agentforce Momentum
- Salesforce: Record First Quarter Fiscal 2027 Results
- Salesforce Customer Story: Grupo Falabella
- CIO: Salesforce Unveils AI Help Agent With Pay-Per-Resolution Pricing
CRM & ERP Enterprise Technology Expert and Entrepreneurial Executive with 20+ years of leading CRM, ERP, Customer Experience, and Block-chain initiatives and projects across internal and customer facing technologies. Proven success in closing large deals in Pre Sales customer facing engagements and deploying enterprise wide CRM & Customer Experience solutions internationally and domestically.