Salesforce Agentforce pricing: credits, conversations, and real budgeting

Salesforce Agentforce pricing at a glance

The figures above are from Salesforce's live Agentforce pricing page, checked September 8, 2026. Salesforce labels that page informational and tells buyers to confirm detailed pricing with a representative.

How Flex Credits turn into a bill

Flex Credits are the part worth understanding before you run a pilot. Salesforce defines an action as a function an agent executes: updating a CRM record, answering a product inquiry, summarising a case, or running a prompt or flow. Its pricing help article says one standard action uses 20 credits. At the published pack price, that is $0.10 per action.

That does not mean each customer question costs $0.10. An order-status request might need several billable actions. Count actual metered actions, not every reasoning step. The following forecast applies only when all counted actions use the 20-credit rate; Salesforce separately lists voice actions at 30 credits:

monthly requests × average actions per request × $0.10

For example, a workflow that averages three actions across 10,000 monthly requests is 30,000 actions, or $3,000 at the published action rate. It is only an illustration, not a Salesforce quote. Measure your actual action path rather than assume every request is the same.

Conversation pricing and user pricing solve different problems

Salesforce lists Conversations at $2 for customer-facing use. This model can be easier to model when your demand is mostly inbound service conversations, because the billable unit is visible to a support lead.

Flex Credits fit broader work: employee help, CRM updates, workflows, voice, and agents that take several steps. Salesforce also lists an Agentforce User License at $5 per user/month, and states that it requires Flex Credits. Treat a user license as access, not proof that every action is unmetered.

Three costs that do not fit neatly in the agent line item

1. Your base Salesforce estate. Agentforce runs inside the Salesforce environment you already license. Salesforce's own calculator excludes Salesforce license costs.
2. Data and integrations. The same calculator calls out Data 360 Flex Credits. External systems, custom actions, and data preparation can also require build work.
3. Implementation and operating time. Salesforce excludes implementation from the calculator. Include the people who will define permissions, test edge cases, review failures, and maintain the workflow.

A credit forecast without these three lines is not a business-case forecast.

A sensible way to budget an Agentforce pilot

Start with a contained job, such as routing a defined type of case or answering a bounded policy question. Pull a month of representative demand. Map the actions the agent must take, including lookup, write-back, and escalation. Then model low, expected, and peak volumes.

Use Salesforce's Digital Wallet and the product's reported usage to compare that model with live consumption. Keep a human approval or escalation path for actions that change records, issue refunds, or expose sensitive data. The useful success test is not “did the demo respond?” but whether the workflow resolves the right cases without creating costly follow-up work.

When a helpdesk teammate is the simpler comparison

If the job is support rather than broad CRM automation, compare an agent against the actual work it needs to join: your queue, knowledge sources, and escalation rules. eesel's AI helpdesk teammate is built for that narrower job. It works alongside the dashboard, but it can also be operated by a person, a script, or coding agents such as Claude Code, Codex, and Cursor through the eesel CLI.

Use the CLI for a different budgeting question: what did the teammate actually do during a planned support workload? With Node.js 18.17 or newer, scope the read-only inspection to the intended teammate:

npx @eesel/cli billing --agent "Customer support"
npx @eesel/cli activity --agent "Customer support"

eesel billing shows live billing state, while eesel activity shows the work behind it. Together they let a team compare a planned month of support requests with the observed teammate activity before changing scope. They do not establish an equivalent unit between Agentforce Flex Credits and eesel billing; each product meters its own service. If a team changes an eesel configuration after the review, use the supported command’s --dry-run first and have the authorised person approve the real update. The dashboard and CLI are two views of the same workspace, and the CLI documentation has the full reference.

The eesel Home page shows setup tasks and helpdesk, Slack or Teams, and chat-bubble connection options.

The bottom line

Agentforce pricing is legible once you separate access from usage. For a consumption model, count the actions in the work, not merely the conversations that start it. Add your Salesforce, data, and implementation costs, then run a scoped pilot with a measurable escalation rate.

Put an AI helpdesk teammate to work

If you want to test AI on the support work you already run in Salesforce, eesel can sit beside that workflow rather than asking your team to adopt a new queue. Connect the knowledge and systems the teammate needs, define the actions it may take, test the behavior, and keep approvals where a human decision belongs. Try eesel when you are ready to compare a ready-to-work support teammate with a broad agent platform.