Buyer's guideSalesforce Agentforce Center of Excellence

Agentforce pricing at program scale: forecasting and controlling consumption

Agentforce costs follow usage, not headcount, so a program with several agents needs a forecast built from process volumes and agent design rather than seat counts. This guide explains the commercial models Salesforce lists today, how to estimate conversations and actions per agent, how to show costs to business units, which controls catch overruns early and which design choices waste consumption. Prices change, so treat the figures here as dated examples.

Reviewed 10 October 2026 7 min read

On this page

Why agent costs do not behave like seat licenses
Salesforce's Agentforce commercial models, as currently listed
Estimating volume agent by agent
A hypothetical forecast for three agents
Showback or chargeback across business units
Consumption controls to switch on before go-live
Design patterns that inflate consumption
Questions to settle with Salesforce and finance before scaling
Questions and answers
Sources

Why agent costs do not behave like seat licenses

A seat license costs the same whether a user logs in once a month or all day. Agent consumption moves with demand and with design: a busy week in the contact center, a new channel or a change that makes an agent call one more action per request all show up in the bill. Two agents serving the same volume can cost very different amounts.

That makes budgeting closer to cloud infrastructure than to software licensing. Finance needs a forecast with explicit assumptions, product owners need to see how their design choices affect cost, and the program needs monitoring that flags drift early. The CoE is the natural owner of all three, because it sees every agent at once.

Salesforce's Agentforce commercial models, as currently listed

Salesforce lists three main ways to pay for Agentforce. Product names, units and terms change, so check the current pricing page and your contract before relying on any row.

Aspect	Flex Credits	Conversations	Per-user add-ons and editions
What is metered	Each action an agent performs¹.	Each conversation¹.	Each licensed user per month¹.
List price when checked	$500 per 100,000 credits, with a standard action using 20 credits and a voice action 30¹.	$2 per conversation¹.	Sales, Service, Field Service and Operations add-ons at $125 per user per month; industry add-ons at $150¹.
Positioned for	Any agent, including employee-facing and multi-step workflows	External, customer-facing agents¹.	Predictable budgeting for named users inside Salesforce apps
Buying options	Pre-Purchase, Pre-Commit or PayGo¹.	Not offered with Pre-Commit or PayGo¹.	Not offered with Pre-Commit or PayGo¹.
Combining models	Not supported alongside conversation pricing in the same org¹.	Not supported alongside Flex Credits in the same org¹.	Max Editions bundle the add-on with an annual Flex Credits allocation per org¹.

Figures are US dollar list prices from Salesforce's pricing page, checked on 10 October 2026¹. Contracted rates and regional prices differ.

Estimating volume agent by agent

Define the unit of work
For each agent, name the event that starts an interaction: an inbound chat, a case, a lead or an internal request. Use the same definition in the forecast and in monitoring.
Output
Unit definition
Pull historical volumes
Take a year of volumes for that unit from Salesforce reports, by month and channel, so seasonality shows. Exclude channels the agent will not serve.
Output
Baseline volume
Estimate the share the agent handles
Agree an assumption for the proportion of volume the agent will engage with, plus a low and a high case. Widen it only when pilot data supports it.
Output
Engagement scenarios
Measure actions per conversation
Count actions in sandbox evaluation runs and pilot transcripts instead of guessing, and note the spread between simple and complex requests, including retries.
Output
Actions-per-conversation range
Convert to the billing unit
Multiply engaged volume by actions per conversation and by the credits per action type; for conversation pricing, multiply engaged volume by the conversation rate.
Output
Monthly consumption forecast
Add testing and growth
Include the batch tests and evaluation runs each release needs, agents planned later in the year and growth in channel volume.
Output
Program forecast

A hypothetical forecast for three agents

Hypothetical example

For example, a software company might plan three agents. A customer service agent is expected to engage eight thousand chats a month at an average of four actions each, or thirty-two thousand actions. An internal sales assistant might handle two thousand requests at three actions each, and a field-service scheduling agent one thousand requests at five actions each.

That totals forty-three thousand actions a month. At the listed 20 Flex Credits per standard action, the program would need 860,000 credits a month before testing and growth¹, which finance then prices at the contracted rate.

If a design change pushed the service agent from four actions per chat to six, its consumption would rise by half with no change in customer demand. That is why actions per conversation is the number to watch.

Showback or chargeback across business units

If
The program is in its first year and volumes are still uncertain.
Then
Use showback: report each unit's consumption monthly without moving budget.
Charging teams for an unstable forecast discourages the pilots the program needs.
If
Several units run production agents with stable volumes.
Then
Move to chargeback by agent, and allocate the CoE's shared costs by a published key.
Teams manage what they pay for, and a published key avoids arguments about overhead.
If
One agent serves several units, such as a shared service desk.
Then
Allocate by the unit recorded on each conversation or case, not by headcount.
The usage data already identifies the requester, so allocation can follow actual use.
If
Credits are bought centrally in advance.
Then
Set internal allowances per unit and review draw-down every quarter.
Unused Flex Credits do not roll over into the next subscription term¹, so over-allocation is a real loss.

Consumption controls to switch on before go-live

0 of 7 checked

Threshold alerts in Digital WalletDigital Wallet supports configurable alerts as usage approaches key thresholds²; route them to the program owner and the finance partner.
A weekly per-agent review at launchDigital Wallet reports consumption down to agent, action and channel²; review the top consumers each week until patterns settle.
Test usage tracked separatelyConfirm how sandbox and test runs are metered under your contract and keep them apart from production figures.
Retry limits on failed actionsCap how often an agent may retry before it escalates to a person.
Early escalation for out-of-scope requestsHand off requests the agent cannot serve instead of letting it ask clarifying questions repeatedly.
A consumption estimate in every release recordCompare actual actions per conversation with the estimate after each release.
A named owner for the forecastOne person reconciles forecast, actuals and contract balance every month.

Design patterns that inflate consumption

Over-granular actions

Early signalOne request triggers a chain of small lookups.

MitigationCombine steps that always run together into one Flow or Apex action, so the agent invokes one action instead of several.

Agent work that should be automation

Early signalThe agent sends routine notifications or updates fields on a schedule.

MitigationMove deterministic, triggered work to Flow outside the agent and keep the agent for requests that need reasoning.

Vague scope

Early signalMany conversations end in clarification loops or off-topic answers.

MitigationTighten subagent descriptions and instructions, and test with ambiguous requests before release.

Untracked evaluation runs

Early signalConsumption jumps in weeks with no change in customer volume.

MitigationSchedule and budget batch tests, and record them so they are not mistaken for production growth.

Stranded committed credits

Early signalThe pre-purchased balance is far ahead of draw-down late in the term.

MitigationBuy in line with measured demand, weighing the cost of overage, billed at the contracted rate monthly in arrears¹, against credits that expire unused.

Questions to settle with Salesforce and finance before scaling

0 of 6 checked

Which commercial model each org will useThe choice binds every agent in that org, customer-facing and internal alike.
The contracted rate and buying modelPre-Purchase, Pre-Commit or PayGo, and how any true-up at term end works.
How sandbox and test usage is metered
Multipliers for voice and other action typesSalesforce refers buyers to its Agentforce rate card for these¹.
Who receives alerts and invoicesAnd who can act on them the same day.
The allocation key for shared program costs

Questions and answers

How should we budget an Agentforce pilot differently from production?

Budget a pilot for learning: a fixed allowance covering expected conversations plus repeated evaluation runs, with an alert well before it runs out. The pilot's main output for finance is the measured engagement rate and actions per conversation, which replace the assumptions in the production forecast. Avoid committing large volumes until those two numbers are known.

Do test runs in sandboxes consume credits?

Ask your account team how sandbox and testing usage is metered under your contract, because terms can differ and change. Either way, put evaluation runs in the forecast, since each release should re-run its test set, and keep test consumption visible separately so it is not mistaken for growth in production demand.

Can one agent use conversation pricing and another Flex Credits?

Not within the same Salesforce org: Salesforce states that Flex Credits and conversation pricing are not supported together in one org¹. Companies with several orgs can choose differently per org. Within one org, choose the model that suits the whole agent mix, because customer-facing and employee-facing agents will share it.

What happens if we exceed our Agentforce entitlement?

Salesforce's pricing page says there is no overage penalty: usage beyond the entitlement is billed at the contracted rate, monthly in arrears¹. The practical risk is to the budget rather than a surcharge, which is why threshold alerts and a monthly reconciliation of forecast, actuals and balance matter.

Sources

Agentforce Pricing — Salesforce · checked 10 October 2026
Optimize Your Consumption with Digital Wallet — Salesforce · checked 10 October 2026

Pressure-test your Agentforce consumption forecast

Share the agents you plan, their expected volumes and the commercial model under discussion. We will reply with the assumptions we would test first and the controls to set before launch.

Review my forecast

Why agent costs do not behave like seat licenses

Salesforce's Agentforce commercial models, as currently listed

Estimating volume agent by agent

Define the unit of work

Pull historical volumes

Estimate the share the agent handles

Measure actions per conversation

Convert to the billing unit

Add testing and growth