all insights

Agentforce pricing explained: Flex Credits, conversations, and the real cost math

Two Agentforce quotes for the same agent can differ by a factor of four depending on which meter you pick. Here's how conversations and Flex Credits actually price out, what the invoice leaves off, and how to build a year-one number you can defend.

Agentforce pricing explained: Flex Credits, conversations, and the real cost math — article illustration

Ask three people what Agentforce costs and you’ll get three answers. One quotes $2 per conversation. Another quotes $0.10 per action. A third says their AE mentioned a per-user licence and an edition bundle with credits included. All three are right, which is exactly the problem: Agentforce has had two consumption meters, two per-user models, and an edition bundle in the space of about eighteen months, and most cost estimates we see mix units from more than one of them.

The stakes aren’t trivial. On the same workload, the conversation meter and the Flex Credit meter can produce year-one numbers that differ by 3–4x. And the metered agent usage is often not even the biggest line — Salesforce’s own pricing examples carry a footnote saying the math does not include other costs like Data 360 credits, which is where knowledge indexing and grounding actually get billed.

This post walks through the pricing model as it stands in early 2026: how each meter works, where the crossover between them sits, the adjacent costs that don’t appear on the Agentforce rate card, and a worked year-one estimate with every assumption labelled. Prices change — treat the official pricing page as the source of truth and this as the map.

Two meters, one agent: conversations vs. Flex Credits

Agentforce launched at Dreamforce 2024 with a single, simple meter. Pricing started at $2 per conversation, with standard volume discounts. One customer engagement, one charge, regardless of what the agent actually did. Salesforce’s own product leadership later described the trade-off bluntly: a conversation cost $2 whether the customer asked what time a store opens or asked the agent to troubleshoot a complex mechanical problem.

That flat rate is easy to budget and brutal on simple, high-volume use cases. So in May 2025 Salesforce introduced a second consumption meter: Flex Credits, which price the individual actions an agent executes rather than the interaction as a whole. An action is a specific function the agent performs on the platform — updating a record, retrieving order details, answering a question from knowledge, running a custom prompt or Flow.

Both meters still exist, but they are not interchangeable and they are not combinable. The pricing page FAQ is explicit: Flex Credits and Conversations will not be supported in the same org. If you’re on conversation pricing and want to move, you can — but you’ll need to swap all of your existing conversation SKUs, not blend the two. Conversations are also positioned for customer-facing agents only, and are sold on a pre-purchase basis only, while Flex Credits cover customer-facing, employee-facing, and voice use cases.

One definitional note that matters for budgeting: a conversation is metered per engagement session, not per message. The exact session boundaries (including time-based cutoffs) live in Salesforce’s product terms rather than on the pricing page, so if you’re modelling the conversation meter seriously, get the definition in writing from your AE before you extrapolate.

How Flex Credits are consumed and priced

The mechanics are simpler than the name suggests. As of early 2026, per the official pricing page:

The complexity of the request is what drives cost. Salesforce’s illustrative example: an SDR-style inquiry that answers a product question and books a meeting might execute 3–6 actions, costing $0.30–$0.60 — versus a flat $2 on the conversation meter. Its pricing-page worked examples run the same way: an order-status request is 2 actions ($0.20), a case-management assist is 3 actions ($0.30), a field-service scheduling flow is 5 actions ($0.50).

Three buying structures sit on top of the credit itself. Pre-Purchase means paying upfront for a set amount of usage across the contract term, and it’s where the deepest discounting lives. Pre-Commit is a committed baseline billed monthly in arrears, with a true-up at term end if you land under your commitment. PayGo is pure pay-as-you-go, billed monthly in arrears with no upfront commitment. Per the pricing FAQ, PayGo and Pre-Commit apply to Flex Credits only — conversation pricing, the add-ons, and Agentforce 1 Editions don’t qualify.

Two FAQ answers deserve more attention than they get. First, unused Flex Credits do not roll over into subsequent subscription terms — over-buying on a pre-purchase deal is money gone. Second, there’s no overage penalty: exceed your entitlement and you’re simply billed your contracted rate monthly in arrears. That asymmetry should shape your negotiation. Buying slightly under your forecast and absorbing overage at the contracted rate is usually safer than buying ahead of demand you haven’t proven.

There’s also a free on-ramp. When Flex Credits launched, Salesforce announced that customers on Enterprise Edition or above can get 100,000 Flex Credits at no cost through Salesforce Foundations. The exact allocation has shifted over time and varies by region, so check the current Foundations page — but the point stands: a proof of concept shouldn’t cost you licence money. Consumption is tracked in Digital Wallet, which is provided free and gives per-action granularity on the Flex meter. Turn its alerts on before go-live, not after your first surprising invoice.

Per-user licences: the third model most estimates ignore

Consumption pricing suits customer-facing agents, where volume scales with your customers rather than your headcount. For employee-facing agents the math often inverts, and Salesforce sells flat-rate options for exactly that reason. As of early 2026, the pricing page lists:

  • Agentforce add-ons at $125 per user per month for Sales, Service, and Field Service — unmetered Agentforce usage for those employees.
  • Agentforce Industries add-ons at $150 per user per month for Industries clouds.
  • Agentforce 1 Editions from $550 per user per month, bundling the add-on with a pool of 2.5 million Flex Credits per org per year.
  • An Agentforce User License at $5 per user per month — a lightweight option that provides access for every employee but still meters usage against Flex Credits, with access to a limited set of CRM objects.

The break-even arithmetic is worth doing per persona. At list, $125 per month buys the equivalent of 1,250 standard actions ($0.10 each). A service rep leaning on an agent all day clears that easily; a casual internal user never will. This is where the Flex Agreement matters — Salesforce’s contract structure that lets you convert user licences into Flex Credits, or Flex Credits into user licences as priorities shift. If you’re unsure which side of the break-even a team sits on, negotiate the swap right up front rather than guessing perfectly.

Where the crossover sits: the 20-action rule

Because a standard action is $0.10 and a conversation is $2, the crossover is clean arithmetic: 20 actions per conversation. Below that, Flex Credits are cheaper for the same workload. Above it, the flat conversation rate wins. For voice, at 30 credits per action, the crossover drops to roughly 13 actions.

ConversationsFlex Credits
List price$2 per conversation$500 per 100,000 credits ($0.10 per standard action)
Metering unitThe engagement sessionEach action executed (20 credits; voice 30)
ScopeCustomer-facing agents onlyCustomer-facing, employee-facing, voice, prompts
Buying modelsPre-Purchase onlyPre-Purchase, Pre-Commit, PayGo
Idle interactionsCharged per session regardlessNo action, no charge
Cheaper whenConversations routinely exceed ~20 actionsTypical conversations stay under ~20 actions

So which side of the line do real agents land on? Salesforce’s own published examples all sit between 1 and 5 actions per interaction. In practice, most well-scoped service agents we see resolve interactions in single-digit actions — which is why, for typical deployments, Flex Credits price out meaningfully cheaper on identical traffic.

The honest caveat: action counts are a design property, not a constant. An agent with vague topics and overlapping actions will chain more calls per request than a tightly scoped one. Retrieval retries, multi-step troubleshooting flows, and agents that orchestrate other systems all push the count up. Before you commit to a meter, run a pilot on the free Foundations credits, watch the per-action telemetry in Digital Wallet for a few weeks, and let measured actions-per-conversation — not the demo script — pick your model.

The costs the pricing page won’t total for you

The Agentforce line is rarely the whole invoice. Four adjacent costs show up in almost every real deployment, and Salesforce flags the first one itself in the fine print of its pricing examples: the examples do not include other costs like Data 360 credits or other consumption services.

  • Data 360 (formerly Data Cloud) consumption. Grounding an agent in your data — the thing that makes it useful — runs on a separate credit pool with its own rate card. Data 360 credits list at $500 per 100,000, and can be transferred between Data 360 and Agentforce; there’s also a profile-based model at $240 per 1,000 profiles per year ($420 for the enterprise tier). Ingestion from core Salesforce connectors is free, but downstream operations are not.
  • Knowledge indexing. Agents that answer from your knowledge base first need that content processed into a searchable index. On the published Customer Data Cloud rate card, unstructured data processing is metered at 60 credits per MB in production — and the same document notes Data Cloud credits don’t roll over either. Rates vary by operation (queries are cheap at 2 credits per million rows; profile unification is 100,000 credits per million rows), and Salesforce has been folding Data 360 into the Flex Credit framework, so verify against the current rate sheet before you model it.
  • Integration and build work. Every action an agent takes maps to something underneath — a Flow, an Apex invocable, an API callout. If the order system, the knowledge base, or the entitlement logic isn’t already exposed cleanly on the platform, someone has to build that, and it’s routinely the largest single line in year one.
  • Tuning and monitoring. Agents aren’t launched; they’re raised. The first 90 days of transcript review, topic refinement, and grounding fixes is real ongoing effort, whether it’s your team’s time or a partner’s.

None of these appear on the Agentforce pricing page as dollar figures, and any estimate that omits them isn’t an estimate — it’s the software subtotal wearing a trench coat.

Building a year-one estimate you can defend

Here’s the pattern we recommend: verified list prices on every rate, explicit labelled assumptions on every volume, and honest “scope this, don’t guess” entries for services. The volumes below are hypothetical assumptions for illustration — swap in your own traffic. Every rate is list price as of early 2026 and linked to its source.

Scenario (assumed): a customer-facing service agent on web chat, handling 2,500 conversations per month at an average of 5 actions per conversation, grounded in roughly 1.5 GB of knowledge content.

Line itemVerified rateAssumption (yours will differ)Year-one estimate
Agent usage — Flex Credits20 credits/action; $500 per 100k credits2,500 conversations/mo × 5 actions = 250k credits/mo$15,000
Same traffic on the conversation meter$2 per conversation2,500 conversations/mo$60,000
Knowledge indexing — Data 36060 credits per MB processed; credits $500 per 100k~1.5 GB indexed once + ~150 MB/mo of updates ≈ 200k credits~$1,000
Other Data 360 consumption (queries, transforms)Varies by operation — check the current rate sheetLight grounding workloadAllowance: model from rate card
Implementation & integrationn/a — services, not licencesActions wired to Flows/Apex, security review, testingScope-dependent: get a quote, don’t guess
Tuning & monitoringn/aHeaviest in the first 90 days, then steady-stateScope-dependent

Under these assumptions the software subtotal lands around $16,000 for year one on Flex Credits — versus $60,000 for identical traffic on the conversation meter, which is the crossover math made visible. At 5 actions per conversation you’re paying $0.50 per interaction instead of $2. Push the average to 20 actions and the two models converge; past that, conversations win.

Three practices keep an estimate like this defensible. State every volume as an assumption and give the sensitivity (“at 4,000 conversations a month this line becomes $24,000”). Use list prices and then note that standard volume discounts apply — discounts are upside, not a plan. And never present the software subtotal as the project cost; the integration and tuning lines are real even when they’re someone’s salaried time. If you want to pressure-test your own volumes and payback period, our Agentforce ROI calculator runs this same math interactively.

Pricing is a design decision, not a procurement detail

The most useful reframe we can offer: with Agentforce, your architecture is your bill. On the conversation meter, cost is fixed per interaction and your incentive is deflection volume. On Flex Credits, every action your agent takes is a metered event — which means topic scoping, action design, and retrieval quality stop being purely technical concerns and become line items. A sloppy agent that takes eight actions to do a three-action job costs you 2.6x more, forever, at any volume.

That cuts the other way too. The teams that treat the meter as a design constraint — narrow topics, deterministic actions, grounding that resolves in one retrieval instead of three — end up with agents that are simultaneously cheaper and better, because action count and answer quality are both symptoms of the same discipline.

So run the sequence in order. Prove the use case on free Foundations credits. Measure real actions-per-conversation in Digital Wallet. Pick the meter the telemetry supports, negotiate the swap rights and the overage rate rather than the sticker, and budget the Data 360 and build lines from day one. Do that and Agentforce pricing stops being a moving target and becomes what it should have been all along: a forecastable operating cost you chose with your eyes open. That’s the position you want to be negotiating from — and the position your CFO will actually sign off on. This is the kind of modelling we do with clients before any SKU gets signed, as part of our Agentforce practice.

Understanding the basics

What is the difference between Agentforce conversations and Flex Credits?

Conversations are a flat consumption meter: $2 list price per customer engagement session, regardless of complexity, available for customer-facing agents on a pre-purchase basis only. Flex Credits meter the individual actions an agent executes — 20 credits (about $0.10) per standard action, 30 for voice — and cover customer-facing, employee-facing, and voice use cases. The two can’t be combined in the same Salesforce org, so you pick one consumption model and can switch later by swapping SKUs.

How much does a Flex Credit cost?

Flex Credits list at $500 USD per pack of 100,000 as of early 2026, which works out to half a cent per credit. A standard Agentforce action consumes 20 credits ($0.10), and an Agentforce Voice action consumes 30 credits ($0.15). Credits are fungible across actions, prompts, translations, and voice actions, and they’re the same unit sold for Data 360 consumption. Always confirm current rates on Salesforce’s official pricing page, since packaging has changed several times.

Do unused Flex Credits roll over?

No. Salesforce’s pricing FAQ states that unused Flex Credits do not roll over into subsequent subscription terms, and the Data Cloud rate card applies the same rule to Data 360 credits. There’s no overage penalty, though — if you exceed your entitlement, usage is billed at your contracted rate monthly in arrears. In practice that means it’s usually safer to buy slightly under your forecast and absorb overage than to pre-purchase credits you might forfeit.


Trying to turn Agentforce list prices into a budget your CFO will believe? Talk to us — we build these estimates for a living.

Keep reading

All insights