Agentforce is priced by consumption, not by seat. Every conversation and action draws against a credit balance. That model can be efficient or it can run away. The difference is governance set before go live.
Salesforce Agentforce prices by consumption rather than by seat. This guide explains conversations, Flex Credits, the Data Cloud dependency, and the buyer side controls that cap cost.
Agentforce is priced by consumption. You pay for what the agents do, measured in conversations or actions, rather than for a fixed number of user seats.
The published Agentforce pricing now spans four buying units. A customer facing conversation costs $2. Flex Credits cost $500 per 100,000. A stand alone Agentforce User License is $5 per user per month.
Editions sit above those meters. Agentforce 1 Editions start from $550 per user per month and bundle 2.5 million Flex Credits per organization each year. Add ons for Sales, Service, and Field Service run $125 per user per month, and Industries $150.
Salesforce set out this shift in its May 2025 flexible pricing announcement. The mix matters more than any single rate, because the same workload can land on very different meters.
Agentforce buying units, list price
| Buying unit | List price | What it buys |
|---|---|---|
| Conversation | $2 each | One customer facing session |
| Flex Credits | $500 per 100,000 | Metered actions, 20 credits each |
| Agentforce User License | $5 per user per month | Employee access that draws credits |
| Sales or Service add on | $125 per user per month | Unmetered employee facing usage |
| Agentforce 1 Editions | from $550 per user per month | Bundle with 2.5M credits per year |
Seat pricing is predictable because it is capped by headcount. Consumption pricing is uncapped by design, so the cost follows usage wherever it goes.
A seat that a user forgets to open still costs the same. A meter that agents call thousands of times a day compounds. The budgeting reflex from Sales Cloud does not carry over.
The finance implication is a variable cost line where there used to be a fixed one. Plan for a monthly true up, not an annual seat count, and give the owner authority to throttle agents.
Define precisely what triggers a billable conversation or action with Salesforce. Ambiguity in that definition is where consumption cost quietly inflates.
A conversation covers one customer facing session. An action, at 20 Flex Credits, covers one unit of agent work such as a record update or a case resolution. Voice actions cost 30 credits. Get these definitions in writing.
Ask which unit each workload should sit on before signing. A high volume customer channel may favor conversations, while a back office automation estate favors Flex Credits. The wrong default can double the run rate.
Editions can beat raw metering for heavy internal use. Agentforce 1 Editions bundle 2.5 million credits a year, which absorbs steady employee automation without a separate block purchase.
Flex Credits are a prepaid balance that agents draw down per action, giving finer control than whole conversations. They suit estates running many small automated actions.
Credits are bought in blocks of 100,000 for $500, which sets one credit at half a cent. A standard action consumes 20 credits, so $0.10. A voice action consumes 30 credits, so $0.15.
An action is a discrete unit of agent work: updating a record, running a workflow, or resolving a case. The Agentforce platform meters each draw, so monitoring the burn rate is essential from day one.
Enterprise Edition customers receive 100,000 Flex Credits at no cost through Salesforce Foundations. That free block funds a pilot, but it disappears quickly once agents run at production volume.
Watch the blended rate, not the headline. A block looks cheap at half a cent, but an agent that fires ten actions to close one case spends $1.00 before any conversation charge applies.
Credits fund every agent uniformly. There is no per agent cap by default, so one misconfigured automation can drain a block, which is why burn alerts matter more than the unit price.
Flex Credit action costs at list
| Item | Flex Credits | Cost at list |
|---|---|---|
| Standard action | 20 | $0.10 |
| Voice action | 30 | $0.15 |
| Credit block | 100,000 | $500 |
| Foundations free tier | 100,000 | $0 for Enterprise Edition |
Seat pricing versus consumption pricing
| Dimension | Seat model | Agentforce consumption |
|---|---|---|
| Cost driver | Headcount | Conversations and actions |
| Predictability | High, capped by seats | Lower, follows usage |
| Scaling risk | Step changes at thresholds | Continuous growth with volume |
| Main control | User provisioning | Volume governance |
Choose per conversation for customer facing volume you can count, and Flex Credits for many small employee facing actions, because the two meters reward different workloads.
A conversation at $2 is simple to forecast when each customer session maps to one billable unit. High deflection support desks often model cleanly this way, because ticket volume is already tracked.
Flex Credits win when a single request triggers several actions. At 20 credits per action, three actions in one interaction cost $0.30, cheaper than a $2 conversation only while the action count stays low.
The trap is mixing models without doing the arithmetic. Run both rate cards against your real workload before you sign. The cheaper unit on paper can be the dearer one in production.
Blended estates are common. Many buyers run customer facing conversations on the $2 unit and internal automations on Flex Credits, then reconcile both against one budget line each month.
Salesforce offers three buying models: Pre Purchase upfront, Pre Commit in arrears against a baseline, and PayGo with no commitment. Match the model to how confident your forecast is.
The free Digital Wallet tracks consumption across these models. Use it, because a commitment sized to a hopeful forecast converts directly into paid credits you never burn.
Agentforce relies on Data Cloud to ground agents in your data, and Data Cloud is itself priced by consumption, so adopting Agentforce usually adds a second meter to the bill.
Salesforce now markets Data Cloud as Data 360. Budget Data Cloud alongside Agentforce from the start. Treating it as a later add on is how the total cost outruns the original business case.
Grounding is not optional. An agent with no data context returns generic answers, so the pressure is always to ingest more, which runs the Data Cloud meter harder.
Data Cloud charges for ingesting and processing data, metered in credits of its own. The more context your agents need, the more that meter runs.
Profile unification, segmentation, and real time data streams each carry their own credit draw. An Agentforce rollout that broadens grounding sources will pull Data Cloud spend up with it.
Model Agentforce and Data Cloud as one consumption line, not two. A business case that budgets agent actions but ignores the grounding cost behind them will understate the true run rate.
Ask Salesforce to quantify expected Data Cloud consumption for your grounding design. A footprint modeled up front is far easier to govern than one discovered on an invoice.
The standard pitch is that consumption pricing is fairer because you only pay for what you use, so there is no need to worry about overbuying. We disagree. In the Agentforce and Data Cloud commitments we have advised, the open ended meter combined with optimistic volume forecasts produced bills well above the seat based tools they replaced, because usage rose faster than anyone projected. The buyer side move is to treat consumption pricing as a budget you must actively govern, pilot to measure true volume, and commit credits to a measured floor rather than a hopeful forecast. Pay as you go is only cheap when you control the go.
Source: Redress Compliance advisory engagement file, 2024 to 2025.
Consumption pricing is not cheaper or dearer than seats. It is less forgiving. The meter rewards governance and punishes optimism.
Forecast from measured pilot volume, not a planning estimate, and build the model bottom up from actions per interaction times interactions per period.
Start with volume you already track: cases, chats, or calls per month. Multiply by the actions each one triggers, then by the credit cost per action. That yields a monthly credit run rate.
Add a contingency band. Early forecasts commonly undershoot, so a defensible model carries a base case plus an upside where adoption and action counts both rise.
Then convert credits to dollars at $0.005 per credit and compare against the per conversation alternative. The cheaper meter for your actual action density becomes obvious once the numbers sit side by side.
Illustrative monthly consumption model
| Input | Base case | Upside case |
|---|---|---|
| Interactions per month | 10,000 | 16,000 |
| Actions per interaction | 3 | 4 |
| Monthly actions | 30,000 | 64,000 |
| Flex Credits used | 600,000 | 1,280,000 |
| Cost at list | $3,000 | $6,400 |
Illustrative arithmetic at list price, not engagement data.
Refresh the forecast monthly against Digital Wallet actuals. Consumption is a live number, so a static annual estimate drifts within a quarter once new agents and channels come online.
Retries, long multi step reasoning, and broad channel exposure each raise actions per interaction. Instrument these early, because a rise from three to four actions per interaction lifts the bill by a third.
Volume governance caps Agentforce cost more than any rate negotiation: control which agents run, on which channels, and watch the burn. Three controls keep consumption in line.
Run a measured pilot to learn true conversation volume before buying a large credit block. The free 100,000 credit Foundations block is enough to gather real numbers.
A pilot converts guesswork into a run rate you can defend to finance. Skipping it means sizing a commitment against a forecast no one has tested.
Guardrails decide which agents can spend before they ever run. Set them in configuration, not in a policy document.
Name an owner for the consumption line. Without one, credit burn becomes everyone and no one, and the first sign of overrun arrives on an invoice rather than a dashboard.
Size any credit commitment to measured volume, not to an optimistic forecast. A commitment you overshoot is discounting you never earned.
Commit to the floor you will certainly use, then buy incremental blocks as real volume proves out. That keeps leverage on your side at renewal.
Negotiate a rate hold on the credit price, rollover of unused credits, and a written definition of a billable action before you commit to any block.
The list rate of $0.005 per credit is a starting point, not a fixed cost. A volume commitment should earn a discounted rate held for the term, so a mid term list increase cannot reprice your run rate.
Unused credits are the quiet leak. Push for rollover or a true up rather than use it or lose it, so an optimistic first year does not strand paid balance.
Definitions carry the risk. Get action, conversation, and voice action defined in the order form, and cap how a definition change can affect credit draw during the term.
The Flex Agreement lets you shift budget between user licenses and Flex Credits. Negotiate that flexibility explicitly, because reallocating between human and digital labor is where the model can actually save money.
Put a benchmark clause in writing. The right to reprice against prevailing market rates at renewal protects you when Salesforce discounts the same units more deeply for later cohorts.
Tie any multi year commitment to service levels and an exit ramp. If adoption stalls, you want the ability to step down committed volume rather than fund credits the estate will never draw.
Contract protections to secure
| Protection | Why it matters | Buyer ask |
|---|---|---|
| Rate hold | List price can rise | Fix the credit rate for the term |
| Credit rollover | Forecasts miss both ways | Roll unused credits forward |
| Action definition | Definitions drive the meter | Define units in the order form |
| Flex Agreement | Priorities shift mid term | Allow budget reallocation |
Early deals show that the meter, not the rate card, decides the bill, and that grounding cost and volume growth are the two lines buyers underbudget most.
The pattern follows the pricing structure. Because the credit rate falls only through negotiated commitments, unmanaged rollouts pay full list on volume that scaled past the plan.
Data Cloud grounding is the second surprise. Buyers who scoped agent actions but not the ingestion behind them saw a second meter appear that was never in the first business case.
Cohort timing matters. Early adopters often signed before the Flex Credit rate card matured, so a renewal is the moment to reset terms against the current published structure.
Grounding scope is the swing factor. Two buyers with identical agent counts can post very different bills purely because one ingests far more context into Data Cloud than the other.
The buyers who fared best treated the free Foundations block as a metering experiment, sized a floor from it, and locked a rate hold before scaling. Governance, set first, did the heavy lifting.
The consistent lesson is sequence: pilot, instrument, then commit. Buyers who inverted that order, committing first and measuring later, carried the widest gap between forecast and invoice.
What to do next breaks into seven concrete moves, taken in order.
Agentforce is priced by consumption. You pay for what agents do, measured in conversations or actions, rather than for a fixed number of seats. A conversation has historically been priced around two dollars.
Flex Credits are a prepaid balance that agents draw down per action, giving finer control than whole conversations. They suit estates running many small automated actions.
Because consumption pricing is uncapped by design. Unlike a seat model bounded by headcount, the cost follows usage, so volume growth flows straight to the bill.
In practice yes. Agentforce relies on Data Cloud to ground agents in your data, and Data Cloud is itself priced by consumption, adding a second meter to the cost.
Run a measured pilot before committing. Early forecasts commonly undershoot real volume, so live pilot data is far more reliable than a planning estimate.
Volume governance. Limiting which agents run, on which channels, and monitoring credit burn with alerts does more than any rate negotiation.
Not before a pilot. Size any commitment to measured volume rather than an optimistic forecast, so you commit to a floor you will actually use.
Seat pricing is predictable and capped by headcount. Consumption pricing is less forgiving, growing continuously with usage, so it rewards governance and punishes optimism.
A Flex Credit costs half a cent, sold in blocks of 100,000 for $500. A standard action uses 20 credits, so $0.10, and a voice action uses 30 credits, so $0.15.
Enterprise Edition customers receive 100,000 Flex Credits at no cost through Salesforce Foundations. That block funds a measured pilot but is consumed quickly at production volume.
Consumption math, Flex Credit sizing, the Data Cloud dependency, governance controls, and the buyer side moves across the Salesforce estate.
Used across more than five hundred enterprise engagements. Independent. Buyer side. Built for procurement leaders running the next renewal cycle.
Agentforce moves Salesforce from seats you can count to actions you must govern. The rate card is the easy part. The governance is the whole game.
