Salesforce raced out of the gates when launching Agentforce 2.0 in October 2024.
In doing so, it beat out the likes of Microsoft, SAP, and ServiceNow, hoping businesses would establish its platform as a foundation for developing AI agents.
The initial $2 per conversation pricing model was simple to digest, and it delivered value across several use cases, like AI agents that handle frontline customer conversations.
Yet, as brands have started to experiment more with Agentforce, they’ve developed numerous AI agent prototypes that deliver differing levels of value. Some offered strong ROI, others didn’t.
Moreover, as the use cases expanded, the previous model caused confusion. After all, if an AI agent is running tasks like building a quote, for example, a conversation-based approach just doesn’t fit.
As such, Salesforce needed a more flexible, tangible Agentforce pricing model.
Bill Patterson, EVP of Corporate Strategy at Salesforce, believes the business has delivered this with its new and improved pricing strategy.
"We've been listening to our customers and developing a new framework called action-oriented pricing," he told CX Today.
This model doesn’t charge for small talk or general fluff like: "How was your day?" It charges based on the actual work performed by the agent, what we call "taking action".
Indeed, with the action-based model, businesses pay for the tasks agents complete, not for abstract technical metrics like compute, storage, or large language model (LLM) calls.
Each Agentforce action costs 20 credits (or $0.10), with Flex Credits sold in packs of 100,000 for $500.
Unlike traditional cloud-specific credits - like for Sales Cloud, Service Cloud, etc. – businesses can use a Flex Credit to power AI agents that work across all Salesforce clouds.
That extra flexibility allows customers to scale up or down as needed, regardless of whether their agent is used for sales, service, or marketing.
"Many customers want agents that can do all those things, and the Flex Credit supports that by being a universal currency across our platform," summarized Patterson.
Yet, perhaps most crucially, a Flex Credit scales from the simplest to the most complex use cases.
Think of it like powering a home with electricity. Businesses don't pay differently to fuel their stove versus their refrigerator; they all draw from the same grid.
Similarly, Salesforce provides an AI agent capacity to enterprises without forcing them to navigate complex pricing tiers.
Yet, Salesforce isn’t giving up on its conversation-based model. Instead, customers can opt for the pricing model that delivers a better ROI for their specific use cases.
Key Additions to the Pricing Model: Flex Agreement & a Digital Wallet
As companies deploy AI agents, they reimagine how their teams work and roles change. In these cases, they could end up with unused Salesforce licenses.
Via its Flex Agreement, Salesforce allows customers to convert those excess license costs into Flex Credits, which they can spend on AI agents. That gives customers more flexibility to reallocate budgets between human and digital labor.
"Many companies go through cycles of surge hiring and firing to meet market demand," said Patterson. "Instead, companies can use digital labor to handle peak demands. This balances their workforce more efficiently.
Previously, they might have purchased software licenses to handle those peaks. Now, they can convert unused licenses into digital labor capacity.
The feature is a positive addition to the pricing model, as it gives customers the confidence to invest in Salesforce with fewer concerns about overbuying.
Additionally, Salesforce can encourage more brands to try Agentforce, as they can do so for "free" if they haven't scaled as quickly on the vendor's CRM apps as anticipated and have unused capacity. That's a common occurrence.
Lastly, there’s also a Digital Wallet that gives companies visibility and control over how many actions their agents perform each month.
Moreover, the Wallet allows organizations to measure AI agent effectiveness, monitor usage, and even set limits if needed.
