Since its launch in October 2024, the Salesforce ecosystem has become saturated with Agentforce marketing, with Salesforce rebranding its World Tour events under the “Agentforce” moniker, and shifting virtually all its product advertising toward Agentforce. With all the buzz about autonomous agents, Salesforce has built high expectations for its agentic AI platform. However, Agentforce has been met with a mixed reception, ranging from genuine enthusiasm to feelings of betrayal, and everything in between.
Agentforce is nothing short of an existential battle for Salesforce. It knows that it needs a credible foothold in the agentic AI space, lest it become relegated to the position of data provider for external AI services. While there is no end in sight for customer experience platforms like CRMs and CDPs, autonomous agents may very well make some of their functions redundant. Additionally, agents promise more direct business impact compared to traditional SaaS solutions. Some experts even say agents could cause the decline of SaaS altogether.
The battle for agentic supremacy is a perilous one for Salesforce. On one hand, it cannot afford to become sidelined. However, its laser focus on AI risks alienating its loyal customer base that is accustomed to its more traditional low-code workflow automation tools. In order to appease its investors, Salesforce must show rapid growth for Agentforce, but it remains to be seen if the ecosystem is ready to adapt in time. Recently, Salesforce has taken action to address the barriers to Agentforce adoption – let’s cover these pain points and Salesforce’s answers to them.
Rigid and Complex Pricing
Pain Point: Customers have had limited flexibility with the Agentforce Conversations SKU, which charged per conversation and lacked granularity.
Salesforce’s Response: Flex Credits introduce an action-based pricing model for Agentforce.
One of the main gripes with Agentforce has been its “$2 per conversation” pricing model. With this model, Salesforce sought simplicity and ease of communication, but ended up limiting flexibility and obscuring costs. What is now called “Agentforce Conversations SKU” charges per conversation, which fails to reflect the value of resolution. With the model, you pay for each conversation regardless of whether they are short, long, meaningful, or useless. This model lacks granularity, making it hard for customers to budget accurately or scale their AI agent usage efficiently.
Salesforce addresses this issue with the introduction of Flex Credits, a usage-based pricing model that charges based on specific agent actions rather than broad conversation counts. This shift gives customers a more transparent, predictable, and scalable approach to pricing.
Conversations that lead nowhere no longer come with a cost. Instead, you pay for value delivered by AI agents. While it is true that complex conversations can now exceed the $2.00 mark, the average cost per conversation will likely be significantly lower in most cases.
Lack of Real-Time Usage Visibility
Pain Point: Customers struggled to monitor and manage their usage effectively.
Salesforce’s Response: Digital Wallet now provides real-time tracking of Flex Credit consumption, agent action usage patterns, and overage alerts.
The lack of real-time visibility on credit consumption has made it difficult to forecast costs and optimize usage. While Digital Wallet showed the credit consumption status, it was not able to estimate future credit consumption. This meant that establishing the ROI for Agentforce use cases amounted to trial and error and guesswork.
Salesforce addresses this issue with updates to the Digital Wallet, a centralized dashboard that provides live tracking of Flex Credit consumption. This tool now shows agent actions, usage patterns, and credit balances in real time. Additionally, overage alerts and consumption breakdowns allow organizations to proactively manage their AI deployments and avoid unexpected cost spikes.
Limited Internal Use Cases for AI Agents
Pain Point: Customers wanted to deploy AI agents internally (e.g., for employee support) but lacked the tools.
Salesforce’s Response: Agentforce for Employees enables deployment of AI agents within Salesforce and Slack to power use cases like FAQs, onboarding, and seller support.
Surprisingly, one of Agentforce’s blind spots thus far has been limited agent operability within the Salesforce Platform. Initially, use cases focused on customer service agents operating in customer-facing situations, and a few sales-related ones, such as SDR and sales coach agents. Up until now, enhancing employee productivity and knowledge sharing has been difficult with out-of-the-box solutions. These constraints have limited Agentforce’s ability to replace manual and rigid rule-based workflows.
Salesforce addresses this gap with the launch of Agentforce for Employees, which allows customers to deploy new employee-facing AI agents within the Salesforce Platform and Slack. These agents can now handle a variety of internal tasks, including answering FAQs, assisting sellers, and onboarding new hires.
Made generally available along with Flex Credits, this feature opens up new opportunities for organizations to enhance internal operations using AI. With internal agentic use cases, Salesforce extends Agentforce’s usefulness into the core of daily operations.
Concerns About Overages and Billing Surprises
Pain Point: Customers feared unexpected charges due to usage spikes that exceed allotted credits.
Salesforce’s Response: Flex Credits imposes no overage penalties; customers continue to be billed at the contracted rate.
With the Conversations SKU, customers feared exceeding the pre-bought credits they were allotted, as this brought overage charges. Where standard conversations were priced at $2.00, the overage cost was around $2.50 per conversation, exceeding the limit. A 20% cost increase is significant, especially at high volumes. In the new Flex Credits pricing model, customers are billed at their contracted rate for any overages, with no penalties, and can track their consumption in real time through the Digital Wallet.
This dashboard not only displays current usage and remaining balances but also includes overage alerts to help customers stay within their limits.
Unfortunately, Digital Wallet does not have a “kill switch” option to freeze agents, external LLM calls, or agent actions automatically when a threshold is reached. Now that Digital Wallet is able to produce real-time insights on credit consumption and predict future trends, it should not be technically difficult to achieve such a solution.
At the time of writing, it is unclear if Salesforce has this functionality on their roadmap or if this is intentional. Be it as it may, this is a functionality that should alleviate the worries of customers and partners alike.
Prohibitively High Cost Compared to Competitors
Pain Point: Customers perceived Agentforce’s Conversations SKUs’ pricing a magnitude higher than some of its competitors.
Salesforce’s Response: Flex Credits offer more granular pricing and significantly lower cost in simple, high-volume conversations.
Under the conversation-based model, customers are charged a flat fee of $2 per conversation, regardless of the complexity or number of actions performed. This is in contrast with Zendesk’s outcome-based model and OpenAI’s token-based pricing, both of which fall below Agentforce’s price point. With Flex Pricing, customers now pay based on actual usage; specifically, 20 Flex Credits per action, which equates to just $0.10 in spend.
This model becomes more cost-effective when conversations involve fewer than 20 actions, which is the break-even point compared to the old pricing. Flex Credits are sold in increments of 100,000 for $500, giving customers the flexibility to scale their usage as needed. Moreover, Salesforce Foundations includes 100,000 Flex Credits by default, allowing customers to test agentic POCs with minimal risk.
Incurring Costs During Development
Pain Point: Customers are upset that they pay for actions already during the build and test phases of a project.
Salesforce’s Response: Testing Center offers a controlled environment for testing agent actions for free, minimizing costs before deployment.
Salesforce hasn’t advertised the fact that all actions cost credits – whether they are done in production or sandbox environments. This applies to Data Cloud credits and Einstein Requests as well, amplifying the cost, especially when unstructured data or custom AI models are used.
Salesforce’s answer has been to run tests within Agentforce’s Testing Center as much as possible, as these tests do not cost credits. However, proper user acceptance testing (UAT) and system integration testing (SIT) require that agents be deployed within sandbox orgs to mimic production use as closely as possible. In practice, Agentforce does incur costs during test and build phases, even when Testing Center is used extensively.
Frustration Over the “Fine Print”
Pain Point: Customers have come to mistrust Salesforce’s claims of simple pricing due to concerns over hidden details that impact overall cost.
Salesforce’s Response: Unfortunately, it falls to Salesforce partners to help their customers understand the implications of different cost factors.
When simplicity in pricing is used as a key selling point of a product, the expectation is set high. In reality, both the Conversations SKU and Flex Credits SKU have notable caveats that customers should be aware of. For instance, Agentforce uses Data Cloud credits and Einstein Requests in addition to agents’ conversations or actions.
Another surprise for most users is that with Flex Credits, they are actually maxed out at 10,000 tokens per action. This means that a single action passing a prompt of 25,000 tokens to an LLM would be considered as 3 actions instead of one.
When an agent has multiple actions with lengthy exchanges, these charges add up drastically. While this information is publicly available, it isn’t exactly customer-centric design to place crucial cost implications within bulleted lists in seemingly mundane Help articles. This is an area where added transparency would help calm customer and partner nerves significantly.
Final Thoughts
Salesforce’s responses shown in this article are profound, tackling some of the major barriers to adopting and scaling Agentforce. Considering that it is still early days, Salesforce shows that they too are learning and adapting to the evolving AI landscape. While Salesforce’s actions are commendable, the true challenges of Agentforce may be even more difficult to address.
Arguably, the biggest issues behind slow AI adoption are foundational ones: weak or non-existent data governance frameworks, outdated development and change management methodologies, siloed decision making, reliance on legacy automations and systems, among others. These are not the kinds of issues Salesforce can solve on behalf of its clients, but it can be part of the solution. The problem is that these changes take time and strategic commitment, which undermine Salesforce’s rush for agentic dominance.
Comments: