AI Licensing Mechanics

Salesforce AI Credits: How the New Consumption Model WorksFlex Credits, per-conversation pricing, per-user licences — three models running simultaneously on the same product, and how to decide which one is right for your organisation

Salesforce’s AI pricing is no longer a simple line item. Since May 2025, the company has introduced three distinct pricing models for Agentforce — Flex Credits (per action), conversation pricing (per interaction), and per-user licences (unlimited usage per seat) — all running concurrently. This guide explains how each model works mechanically, what actions actually cost, how the Digital Wallet tracks consumption, and how to model which approach minimises your total AI spend.

Updated February 202615 min readFredrik Filipsson
¶ This article is part of our Salesforce Einstein Agentforce AI Cloud Cio Playbook series — the independent reference for enterprise salesforce licensing, contract strategy, and cost optimisation.
$0.10
Cost per action under Flex Credits (20 credits per action)
$2.00
Legacy cost per conversation under the original pricing model
$500
Minimum Flex Credit purchase (100,000 credits)
100K
Free Flex Credits for Enterprise Edition+ customers via Salesforce Foundations

Three Pricing Models, One Product: The Current State

Salesforce launched Agentforce in late 2024 with a single pricing model: $2 per conversation. By mid-2025, that model had been joined by two additional approaches. As of early 2026, all three run simultaneously — but they cannot coexist within the same Salesforce org. You must choose one model for your entire deployment. Understanding the mechanics of each is essential before committing, because switching models mid-contract requires renegotiating your entire Agentforce agreement.

ModelUnit of MeasureBase CostLaunchedBest Suited For
Per ConversationComplete interaction (any complexity)$2.00 per conversationOctober 2024Simple, high-volume customer-facing chatbots with predictable interaction patterns
Flex CreditsIndividual actions within interactions$0.10 per action (20 credits)May 2025Multi-use-case deployments where action complexity varies significantly
Per-User LicenceMonthly seat with unmetered usage$125/user/month (add-on)Summer 2025Employee-facing AI usage where per-seat budgeting is preferred

The evolution is significant. The original $2/conversation model was simple but blunt — a three-action interaction checking store hours cost the same as a twenty-action interaction troubleshooting a complex technical issue. Flex Credits introduce granularity by charging per action. The per-user licence reintroduces predictability by removing consumption tracking entirely for a fixed monthly fee. Each model has fundamentally different cost dynamics, and the right choice depends on your specific usage patterns, volume, and organisational appetite for variable versus fixed costs.

Flex Credits: The Mechanics in Detail

Flex Credits are Salesforce’s primary consumption-based pricing model for AI. The system works on a straightforward hierarchy: you purchase credit pools, credits are consumed by actions, and actions are the atomic unit of work that an AI agent performs.

What Is an Action?

An action is a specific function that an AI agent executes on the Salesforce platform. Each action consumes 20 Flex Credits ($0.10). Actions include updating a customer record, summarising a case, answering a product enquiry, executing a custom prompt, running a flow, retrieving knowledge articles, scheduling an appointment, or any other discrete task the agent performs. The critical concept: a single customer interaction typically involves multiple actions. An agent resolving a customer enquiry might perform 3–10 actions in the process — identifying the customer, retrieving their history, querying the knowledge base, composing a response, updating the case record, and logging the interaction. Each of those is a separately metered action.

Credit Consumption by Use Case

Salesforce provides illustrative examples of credit consumption by scenario. The actual consumption for your deployment will vary based on agent design, complexity of workflows, and how many steps each interaction requires.

Use CaseTypical Actions per InteractionCredits ConsumedCost per Interaction
Simple FAQ response (knowledge lookup + answer)2–340–60$0.20–$0.30
Case management (identify customer, pull history, add comment)3–460–80$0.30–$0.40
Product enquiry with recommendation4–680–120$0.40–$0.60
Field service scheduling (find customer, check availability, book slot, confirm)5–7100–140$0.50–$0.70
Complex troubleshooting (multi-step diagnosis, record updates, escalation)8–15160–300$0.80–$1.50
AI SDR qualification (enquiry analysis, product comparison, meeting scheduling)3–660–120$0.30–$0.60

Compare these figures against the legacy $2/conversation model: for simple interactions (2–4 actions), Flex Credits are 70–85% cheaper than conversations. For complex interactions (10+ actions), Flex Credits approach or exceed $2, making them comparable to conversation pricing. The crossover point is approximately 20 actions per interaction — anything below that favours Flex Credits; anything above favours the flat $2 conversation rate. The critical question for your organisation: what is your average number of actions per interaction? If you do not know, you are not ready to choose a pricing model.

Purchasing Flex Credits

Flex Credits are sold in packs of 100,000 credits at $500 per pack (the pre-purchase price). Salesforce offers three payment models for acquiring credits, each with different pricing and commitment dynamics.

Payment ModelHow It WorksPricingBest For
Pre-PurchaseBuy a fixed block of credits upfront for the full contract term. Usage draws down from the balance.Most favourable rate (from $500/100K credits; deeper discounts at volume)Organisations with predictable, consistent AI usage that can forecast annual consumption
Pre-CommitCommit to a minimum annual spend; pay monthly based on actual usage against that commitment.Better than pay-as-you-go; lower than pre-purchaseOrganisations that want a cost ceiling but actual consumption flexibility month to month
Pay-as-You-GoNo upfront commitment. Pay monthly for actual credits consumed.Highest per-credit rate (premium over pre-purchase)Pilots, proof-of-concept deployments, organisations with highly unpredictable usage

The gap between pre-purchase and pay-as-you-go pricing is significant enough to warrant careful modelling. Organisations that can accurately forecast their annual consumption save 20–35% by pre-purchasing versus paying monthly. However, unused pre-purchased Flex Credits do not roll over into subsequent subscription terms — use-it-or-lose-it dynamics mean that overcommitting on pre-purchase creates waste. The optimal strategy for most enterprises is to pre-purchase a conservative base volume (80% of projected consumption) and pay-as-you-go for peak periods that exceed the base.

The Free Credits Trap

All customers with Enterprise Edition or above can get 100,000 Flex Credits for free through Salesforce Foundations. This sounds generous but is a calculated adoption strategy. 100,000 credits fund approximately 5,000 simple interactions or 1,500 complex ones — enough to build a proof of concept and create organisational dependency, but nowhere near enough for production workloads. The free credits are designed to get you started, not to fund your AI programme. Budget for commercial credit purchases from Day 1 if you plan to go beyond experimentation.

Per-Conversation Pricing: The Legacy Model

The original Agentforce pricing model charges a flat $2 per conversation regardless of complexity. A conversation encompasses the entire customer interaction from start to finish, regardless of how many actions the agent performs. The simplicity is the appeal: if your agent handles 10,000 conversations per month, your cost is $20,000. No credit tracking, no per-action metering, no consumption anxiety.

Where conversation pricing still wins: Organisations with high-volume, high-complexity customer-facing use cases where the average interaction involves 15+ actions. In those scenarios, $2 per conversation is cheaper than the equivalent Flex Credit cost. Customer service deployments for technical products (where troubleshooting flows are long and multi-step) and field service coordination (where scheduling involves numerous system queries) are the clearest use cases where conversation pricing maintains an advantage.

Where conversation pricing loses: Simple interactions — FAQ deflection, order status checks, appointment confirmations — where 2–4 actions suffice. Paying $2 for a 3-action interaction that would cost $0.30 in Flex Credits is a 567% premium. Organisations with a mix of simple and complex interactions overpay significantly on the simple ones under conversation pricing.

Critical constraint: Flex Credits and conversations cannot coexist in the same Salesforce org. You must choose one model for all Agentforce use cases. Organisations on the legacy conversation model can transition to Flex Credits, but doing so requires swapping all existing Agentforce SKUs to the new model — a contract renegotiation, not a simple switch. Speak with your AE about the migration path, and use our contract negotiation advisory to ensure the transition does not introduce unfavourable terms.

Per-User Licences: The Fixed-Cost Alternative

Launched in Summer 2025, Agentforce per-user licences offer unmetered employee-facing AI agent usage for a fixed monthly fee per seat. This model eliminates consumption tracking entirely — users can interact with Agentforce without worrying about credit depletion or conversation counts.

Licence TypePriceWhat’s IncludedMetering
Agentforce Add-On$125/user/monthEmployee-facing Agentforce on Enterprise/UnlimitedUnmetered internal AI usage
Agentforce 1 Sales$550/user/monthFull Sales Cloud + Agentforce + 1M Flex Credits/org/yearInternal: unmetered. External: credits
Agentforce 1 Service$550/user/monthFull Service Cloud + Agentforce + Data Cloud + Digital EngagementInternal: unmetered. External: credits

Per-user licences address the core objection that enterprise procurement teams have to consumption models: unpredictable costs. A CFO can budget $125 × 200 users × 12 months = $300,000/year with certainty. There is no overage risk, no consumption monitoring, and no month where the AI bill spikes unexpectedly. The trade-off is that you pay the per-user rate whether individual users consume heavy AI functionality or barely touch it.

The break-even calculation: At $125/user/month, the per-user licence costs $1,500/user/year. Under Flex Credits at $0.10 per action, a user would need to trigger 15,000 actions per year (1,250 per month, roughly 42 per working day) to reach the same cost. If your average user triggers fewer than 1,250 actions per month, Flex Credits are cheaper per user. If they trigger more, the per-user licence saves money. The complication is that usage varies dramatically across roles: heavy-use service agents might trigger 2,000+ actions per month while occasional sales users trigger 200. A blended model — per-user licences for heavy users and Flex Credits for light users — is not possible within a single org, making the choice a portfolio-level decision.

The Flex Agreement: Converting Between Models

Salesforce’s Flex Agreement is the commercial mechanism that allows organisations to shift investment between traditional user licences and Flex Credits as priorities change. Under a Flex Agreement, you can convert unused user licence seats into Flex Credit allocations, or convert Flex Credits into additional user licences for new teams adopting Agentforce.

This flexibility is strategically important because AI adoption does not follow a straight line. An organisation might start with 50 Agentforce per-user licences for its service team, discover that only 30 agents actively use AI, and want to convert the remaining 20 seats into Flex Credits for customer-facing bot deployments. Without a Flex Agreement, those 20 unused seats are simply wasted spend. With it, they become fungible credits that can fund a different use case.

The Flex Agreement is not a default contract feature — it must be negotiated explicitly. Conversion ratios (how many Flex Credits you receive per user licence dollar) are contract-specific and negotiable. Our recommendation: any enterprise entering an Agentforce agreement exceeding $200,000 annually should insist on a Flex Agreement as a standard contract term. The cost of not having one is paying for AI resources you cannot reallocate when usage patterns inevitably shift from your initial projections.

Digital Wallet: Monitoring and Governance

The Digital Wallet is Salesforce’s consumption management dashboard for Flex Credits. It is automatically provisioned (at no additional cost) for any customer with Flex Credit entitlements. The wallet provides near-real-time usage tracking, proactive threshold alerts, usage trend analytics, and the ability to tag credit consumption to specific agents, departments, or use cases.

What the Digital Wallet tracks: Total credits purchased versus consumed, consumption rate per day/week/month, consumption by agent (which AI agents are using the most credits), consumption by action type (which specific actions are most expensive), forecasted depletion date (when your credit pool will run out at current burn rates), and historical trends for capacity planning.

Governance best practice: Set alert thresholds at 50%, 75%, and 90% of your credit allocation. The 50% alert triggers a mid-period review to confirm consumption is tracking to plan. The 75% alert triggers a reforecasting exercise to determine whether additional credits are needed. The 90% alert triggers an escalation to prevent service interruption. Sandbox environments consume credits at a discounted rate, but they still consume credits — ensure your development and testing teams are aware that experimental agent configurations have real cost implications. Our Einstein and Agentforce CIO playbook includes a complete governance framework.

Decision Framework: Which Model Minimises Your Cost?

The choice between Flex Credits, per-conversation, and per-user licensing is not a philosophical preference. It is a financial calculation based on four variables: average actions per interaction, monthly interaction volume, ratio of simple to complex interactions, and the proportion of your workforce that will use AI actively.

Your ProfileRecommended ModelWhy
High-volume customer bots with complex troubleshooting (15+ actions/interaction)Per Conversation ($2)Flat rate cheaper than 15+ actions at $0.10 each
Mixed use cases with varying complexity (3–12 actions/interaction)Flex CreditsGranular pricing rewards simple interactions; only pay for what you use
Employee-facing AI for >40 agents using AI daily at high volumePer-User Licence ($125/user)Predictable budget; no consumption tracking; cost-effective at 1,250+ actions/user/month
Pilot/POC with uncertain volumesPay-as-You-Go Flex Credits or free Foundations creditsNo commitment; learn actual consumption patterns before locking in a model
Large multi-department deployment with shifting prioritiesFlex Credits + Flex AgreementPool credits across use cases; reallocate between user licences and credits as needed

Critical Run a Consumption Pilot Before Committing

No amount of theoretical modelling replaces actual consumption data from your own environment. Use the free 100,000 Foundations credits or negotiate a 90-day pay-as-you-go pilot to measure real action counts per interaction, actual monthly volumes, and peak versus steady-state consumption. Then use that data to model 12-month costs under each pricing model. Committing to a multi-year pre-purchase without this data is how organisations end up with either unused credits (wasted spend) or unexpected overages (budget surprises). Our contract flexibility assessment models all three scenarios.

Worked Cost Comparison: 10,000 Monthly Interactions

To illustrate the financial difference between models, here is a worked example for an organisation handling 10,000 Agentforce interactions per month with a mix of simple and complex use cases.

Assumptions: 60% simple interactions (average 3 actions each), 30% moderate interactions (average 7 actions each), 10% complex interactions (average 14 actions each). Total actions: 18,000 simple + 21,000 moderate + 14,000 complex = 53,000 actions per month.

Pricing ModelMonthly CostAnnual CostNotes
Per Conversation ($2 × 10,000)$20,000$240,000Fixed regardless of action count
Flex Credits (53,000 actions × $0.10)$5,300$63,60073% cheaper than conversations for this mix
Flex Credits (pre-purchase with 25% volume discount)~$3,975~$47,70080% cheaper than conversations
Per-User Licence (50 agents × $125)$6,250$75,000Fixed; no consumption anxiety; 69% cheaper than conversations

In this scenario, Flex Credits are the clear winner, saving $176,400 annually versus per-conversation pricing and $11,400 versus per-user licensing. However, the per-user licence provides cost certainty that Flex Credits do not — if interaction volumes doubled unexpectedly, the per-user cost remains $75,000 while the Flex Credit cost would jump to $127,200. The risk profile of your organisation determines which saving matters more: the guaranteed lower cost (Flex Credits) or the guaranteed budget certainty (per-user).

Now consider the same organisation but with predominantly complex interactions (average 18 actions each). Total actions: 180,000 per month. Flex Credits at $0.10 per action = $18,000/month ($216,000/year). Per conversation: still $20,000/month ($240,000/year). Flex Credits are still cheaper, but the gap narrows to 10%. At 20+ actions per interaction, the models converge — and the simplicity of per-conversation billing may justify the marginal premium.

Negotiation Strategies for AI Credit Agreements

Negotiate Flex Credit volume discounts aggressively. The published rate of $500 per 100,000 credits is the entry point, not the market price. Pre-purchase commitments of 1 million+ credits should target 20–35% discounts, bringing the effective per-action cost to $0.065–$0.08. Salesforce is in strategic growth mode for Agentforce (the product hit $540 million ARR by Q3 FY2026, growing 330% year-over-year) and has strong incentive to lock in large consumption commitments at favourable rates to demonstrate adoption momentum.

Insist on a Flex Agreement from day one. The ability to convert between user licences and Flex Credits is valuable insurance against forecasting errors. It should be a non-negotiable contract term for any agreement above $100,000 annually. Negotiate favourable conversion ratios — ideally 1:1 dollar value, where every $1 of unused user licence converts to $1 of Flex Credits.

Negotiate credit rollover or extended expiry. The default is that unused pre-purchased Flex Credits do not roll over. For large commitments, push for a 90-day grace period after contract end or for unused credits to roll over into the subsequent term. This protects against the use-it-or-lose-it risk that makes pre-purchasing expensive for organisations with variable consumption.

Bundle AI credits into your core renewal for maximum leverage. Purchasing Flex Credits as a standalone mid-term add-on gives your AE incremental revenue with minimal effort — they have no incentive to discount. Bundling credit purchases into your Service Cloud or Sales Cloud renewal creates a larger deal where Flex Credit concessions can be traded against core licence commitments. The best enterprise deals combine core licence renewal, Flex Credit pre-purchase, and Flex Agreement terms in a single negotiation.

Demand sandbox credit discounting. Development and testing environments consume Flex Credits during agent design and iteration. Salesforce offers discounted sandbox consumption rates — but the discount must be negotiated explicitly. Standard sandbox discount is approximately 50% of production credit rates. Push for 80–90% discounting on sandbox environments to avoid penalising your engineering team for building better agents. For comprehensive negotiation guidance, see our Salesforce AI and Data Cloud negotiation playbook.

Hidden Costs Beyond the Credit Price

The per-action credit cost is the visible pricing layer. Several additional costs sit beneath it that must be factored into total AI cost of ownership.

Cost 1 Data Cloud Requirement

Many Agentforce use cases require Data Cloud for data unification and grounding. Data Cloud is separately licensed with its own credit-based consumption model at approximately $1,000 per 100,000 credits. If your AI agents need to access unified customer profiles across systems, Data Cloud credits are an additional cost layer on top of Flex Credits. Budget for Data Cloud consumption at 30–50% of your Flex Credit spend for data-intensive use cases.

Cost 2 Agent Development and Maintenance

Building effective AI agents requires Salesforce developer and architect expertise. Agent design, prompt engineering, testing, and ongoing refinement represent a staffing cost of $50,000–$150,000 per major agent deployment. Salesforce’s Agentbuilder tool simplifies configuration but does not eliminate the need for skilled technical resources, particularly for complex multi-step workflows.

Cost 3 Token Limits and LLM Overhead

Each action has embedded token limits that control how much context the underlying large language model processes. For complex prompts that exceed default token limits, additional processing may consume more credits than the standard 20-per-action rate. Monitor the Digital Wallet for actions that consistently consume above-average credits — these may indicate prompt designs that are inefficient or exceed intended complexity.

Cost 4 Premier Support on AI Products

If your organisation purchases Premier Support, the 30% surcharge applies to your AI product spend as well as your core licence spend. On a $200,000 Flex Credit pre-purchase, Premier adds $60,000 at list. Negotiate Premier Support scope to exclude or cap the AI component, particularly during the early adoption phase when your support needs for AI are minimal. See our guide to maximising value from Premier Support.

Frequently Asked Questions

How much does one Salesforce AI action cost?+
$0.10 per action under the Flex Credits model. Each action consumes 20 Flex Credits from a pool purchased in blocks of 100,000 credits for $500. Volume discounts of 20–35% are achievable on pre-purchase commitments above 1 million credits, bringing the effective cost to $0.065–$0.08 per action. Sandbox actions are discounted separately.
Can I use Flex Credits and per-conversation pricing in the same org?+
No. Flex Credits and conversation-based pricing cannot coexist in the same Salesforce org. You must choose one model for all Agentforce use cases. Transitioning from conversations to Flex Credits requires swapping all existing Agentforce SKUs to the new model. Speak with your Account Executive about the migration path, and ensure the contract renegotiation does not introduce unfavourable terms on your core licences.
Do unused Flex Credits roll over?+
No, by default. Unused pre-purchased Flex Credits do not roll over into subsequent subscription terms. This creates use-it-or-lose-it dynamics that penalise over-commitment. Negotiate a rollover clause or extended grace period for large pre-purchase agreements. The optimal strategy is to pre-purchase conservatively (80% of projected consumption) and use pay-as-you-go for variable demand above the base.
How do I get the free 100,000 Flex Credits?+
Through Salesforce Foundations. All customers with Enterprise Edition or above can access 100,000 Flex Credits at no cost via the Salesforce Foundations programme. This allocation funds approximately 5,000 simple interactions — sufficient for a proof of concept but not for production workloads. Ask your Account Executive to activate Foundations credits if they have not already been provisioned.
When is the per-user licence cheaper than Flex Credits?+
When a user triggers more than 1,250 actions per month. The Agentforce Add-On at $125/user/month provides unmetered usage. At $0.10 per action under Flex Credits, 1,250 actions cost $125 — the breakeven point. Heavy-use service agents who interact with AI agents dozens of times per day often exceed this threshold, making the per-user licence more cost-effective. Light users who trigger fewer than 500 actions per month are 60%+ cheaper on Flex Credits.
What is a Flex Agreement?+
A contract mechanism that lets you convert between user licences and Flex Credits. If you over-provisioned per-user Agentforce licences, a Flex Agreement lets you convert unused seats into Flex Credits for other use cases (or vice versa). Conversion ratios are negotiable. We recommend insisting on a Flex Agreement for any Agentforce contract exceeding $200,000 annually, as it provides insurance against forecasting errors.

Salesforce Licensing Advisory — Explore More

Model Your Agentforce Costs Before You Commit

Three pricing models, consumption-based credits, per-conversation rates, and variable volume discounts create a licensing architecture where the wrong choice costs six figures annually. Redress Compliance models the complete AI cost structure for enterprise Salesforce deployments — with no Salesforce partnership, no reseller arrangement, and no commercial relationship of any kind.

← Back to Salesforce Knowledge Hub