Salesforce Agentforce Pricing Breakdown: $125-$650 Per User Reality Check

Q1. How much does Salesforce Agentforce actually cost in 2026? [toc=1. Four Buying Routes]

A Salesforce admin I spoke with last quarter had one line in a budget sheet that read "Agentforce: TBD". Her CFO wanted a ceiling by Friday. She had the pricing page open, and she still could not fill in the cell, because the page does not describe one price. It describes four ways to buy.

Agentforce has four buying routes, not one price. The Agentforce User License is $5 per user per month and requires Flex Credits bought separately. Add-on licences are $125 per user per month for Sales, Service, and Field Service, and $150 for Industries. Agentforce 1 Editions start at $550 per user per month with bundled org-level credits. Pure consumption is $500 per 100,000 Flex Credits, or $2 per conversation for customer-facing agents. All figures retrieved from salesforce.com/agentforce/pricing on 7 September 2026.

🧭 The four routes side by side

⚠️ Two corrections worth stating plainly

There is no $650 government tier on the current Agentforce pricing page, and $125 is not the entry point either. Both numbers circulate widely in secondary coverage. Neither survives a check against the live page on 7 September 2026. If you want the practitioner accounts behind those numbers, our Agentforce reviews analysis collects them.

The $5 Agentforce User License is missing from most articles on this keyword. That is the route many employee-facing deployments actually take. It is also the route with the biggest gap between what the licence line looks like and what the bill becomes, because it carries no credits of its own.

💰 The commitment question sits underneath the rate

Flex Credits sell three ways, and the mode changes your exposure more than the rate does. Pre-Purchase means you buy a balance upfront and draw it down. Pre-Commit means you commit to a volume. PayGo means you pay as you consume, with no upfront balance.

Per-conversation billing is different again. It is available Pre-Purchase through Digital Wallet only, and it is not available in the same org as Flex Credits. You pick one meter per org, not both.

✅ How to use this table on Monday

Find your row first, then read the rest of this article. Most Salesforce customers are choosing between the $5 licence plus credits, the $125 add-on, and an Agentforce 1 Edition. The consumption rows are not alternatives to those. They are the meter that runs underneath two of them. For the capability list that sits behind each route, see our breakdown of Agentforce for Sales features.

Q2. Why is Agentforce not really priced per user? [toc=2. Why Seat Prices Mislead]

Agentforce is metered on actions, not seats. Two of its four routes bundle or ignore seats entirely, and the bill moves with how many actions agents take. A fixed-headcount team running one predictable workflow can approximate a seat price and be roughly right. Two things break that: voice actions consume 30 credits rather than 20, and any workflow where one user triggers many actions.

📉 The number everyone quotes is the least useful one

The seat figure in every headline for this keyword is the least useful number in the model. That includes the headline this page carried previously, which put the range at $125 to $650. Neither end of that range holds up. A seat price cannot describe consumption billing, and no amount of range-widening fixes that.

I want to be fair about why the mistake is so common. Every buyer in this market learned to price software per seat, because CRM taught them to.

🕰️ What the per-seat habit cost teams

Under the old habit, you counted licences, multiplied, and negotiated once a year. The number was wrong sometimes, but it was wrong in a stable direction. Finance could hold it.

That habit produces a specific failure with agents. You size the deal on headcount, then discover the bill tracks work volume instead. A twelve-person team can generate the action volume of a forty-person team if the workflow is chatty. The same distortion shows up when teams stack tools, which is why revenue tech stack consolidation starts with unit economics rather than feature lists.

🔁 What metered agent billing changed

Metering moved the cost from a number you negotiate once to a number that changes every month. That is the real problem, and it deserves a straight answer rather than a reassuring one. Salesforce does publish its consumption rates openly, including the per-action credit cost on its Agentforce pricing documentation. That is more pricing transparency than most of this category offers, and it is worth saying out loud.

⚖️ When the seat approximation actually holds

The counter-argument is fair, so here it is properly. If you have fixed headcount, one well-understood workflow, and standard text actions only, you can treat Agentforce as a seat price and land close enough for a budget.

Two conditions break that approximation. First, voice: an Agentforce Voice action consumes 30 credits against 20 for a standard action, a 50% premium on every voice interaction. Second, action fan-out: any workflow where one user triggers many actions per session decouples the bill from headcount entirely.

Here is the arithmetic that shows it. One user running 200 standard actions a month costs 200 x 20 credits = 4,000 credits, or $20 at $500 per 100,000. The same user running 200 voice actions costs 6,000 credits, or $30. Same person, same month, 50% more spend, and the seat count never moved. The same seat-versus-usage tension shows up across the category, which our comparison of revenue intelligence platforms works through in detail.

Q3. What are Flex Credits and what does one action cost? [toc=3. Credit Arithmetic]

Flex Credits cost $500 per 100,000, or half a cent each. A standard Agentforce action consumes 20 credits, so $0.10. An Agentforce Voice action consumes 30 credits, so $0.15. Salesforce defines an action as updating a record, automating a workflow, or resolving a case. Credits sell as Pre-Purchase, Pre-Commit, or PayGo, and unused credits do not roll over. They expire at the order end date.

🔢 The rate card

A credit is just a unit of consumption. It is not a token, and it is not a message. The thing that consumes credits is an action, and Salesforce defines an action concretely: updating a record, automating a workflow, or resolving a case.

🎙️ The voice premium most articles flatten

Almost every competing page on this keyword prints a single figure of $0.10 per action. That is correct for standard actions only. Voice actions consume 30 credits, which is $0.15, and the difference compounds fast in any contact-centre-shaped workload.

The formula is worth writing down once, because you will reuse it:

(standard actions x 20) + (voice actions x 30) = credits consumed
credits consumed x $0.005 = cost

🧾 The three purchase modes change your exposure

Pre-Purchase means you buy a credit balance upfront and draw it down as agents work. Pre-Commit means you commit to a consumption volume over the term. PayGo means you pay for what you use, with no upfront balance to buy.

These are not tiers of the same thing. They are three different commitment shapes, and the right one depends on how confident you are in your volume estimate. If you cannot estimate volume yet, a PayGo start followed by a Pre-Commit at renewal is the lower-risk sequence. The entitlement side of this, which editions include what, sits in our Salesforce Einstein pricing tiers explainer.

⏰ The expiry line nobody budgets for

Flex Credits do not roll over. Unused credits expire at the order end date, which means an over-purchase is not a saving carried forward. It is a write-off.

There is one thing I will not guess at. Salesforce's documentation does not describe what happens when a credit balance hits zero, whether agents throttle, queue, or stop. Ask your rep to put the answer in writing, because it is an operational question, not a pricing one.

Q4. Is Agentforce billed per conversation or per action? [toc=4. Conversations vs Actions]

Both exist, but not together in one org. Per-conversation billing is $2 for a customer-facing agent session, available Pre-Purchase via Digital Wallet only. Flex Credits bill per action at $0.10 standard or $0.15 voice. Because they are alternative meters rather than tiers, the choice is arithmetic: sessions under roughly twenty standard actions are cheaper on credits, and longer multi-action sessions are cheaper at a flat $2.

🔀 Two meters, one org

On a quote, these will look like two line items you could mix. They are not. Per-conversation pricing runs through Digital Wallet on a Pre-Purchase basis, and it is not available in the same org as Flex Credits.

That single constraint matters more than either rate. You are choosing a metering model for the whole org, and changing it later is a contract conversation, not a settings change. Deployment sequencing, including when that decision gets locked, is covered in our Agentforce implementation guide.

💸 What flat conversation pricing cost teams

Under a flat per-conversation rate, a session that resolved in one step cost the same as one that took fifteen. Teams with simple, high-volume interactions paid for depth they never used. The rate was easy to forecast and frequently unkind to the simplest workloads.

That is the trade the flat rate makes. Predictability in exchange for paying the same price for a one-step answer and a twenty-step one.

🧮 The break-even, worked once

Flex Credits changed the shape by billing the work rather than the session. So the comparison becomes arithmetic you can run in one line.

$2.00 divided by $0.10 per standard action = 20 actions

Below roughly twenty standard actions per session, credits are cheaper. Above it, the flat $2 wins. With voice actions at $0.15, the break-even drops to about thirteen actions, because each action costs more.

These volumes are illustrative. I am using round numbers to show the method, not to predict your usage. The failure mode across this entire category is a confident annual total built on an invented action count, and it is worth refusing to do that.

🎯 Which shape sits on each side of the line

Short, transactional sessions favour Flex Credits. Think status lookups, single record updates, or a quick qualification question that resolves in three or four actions.

Long, multi-step sessions favour the flat $2. Think a customer-facing case that involves retrieval, several updates, an escalation check, and a summary. Voice-heavy workloads push you toward the flat rate faster, because the 30-credit action rate erodes the credit advantage. Our roundup of Agentforce alternatives and competitors covers what the same workload costs elsewhere.

✅ How to decide before the quote

Pull thirty real interactions from your current process and count the discrete steps each one takes. That step count is your action proxy. If the median sits well under thirteen for voice or twenty for text, credits are the cheaper meter for you. If it sits above, the flat rate is doing you a favour. Teams already running AI agents for RevOps usually have this step count sitting in their workflow logs already.

Q5. Which licence route fits your team: $5 metered or $125 unmetered? [toc=5. Licence Route Decision]

A VP of Sales said this to a RevOps lead I was working with: "Just give everyone Agentforce." The RevOps lead priced it at $125 per seat for 60 reps and came back with $90,000 a year. Nobody had asked how often a rep would actually use it.

The $5 Agentforce User License includes no credits. Every action draws down Flex Credits bought separately. The $125 add-ons for Sales, Service, and Field Service, and $150 for Industries, are unmetered for the covered workflows on an Enterprise-level edition and an annual commitment. The crossover is roughly where a user's monthly actions, priced at $0.10 each, exceed the $120 gap between the two routes. All figures are from the Salesforce Agentforce pricing page, retrieved 7 September 2026.

💰 Why "buy $125 for everyone" usually overspends

The add-on licence is unmetered, which means a heavy user costs the same as a dormant one. That is great for your top five reps and wasteful for the other fifty-five.

Most sales teams have a long tail of light users. In my experience, the distribution is never flat, and the average hides it. The same pattern drives most of the waste our guide to reducing sales tech stack costs works through.

🔎 What the $5 route actually is

The Agentforce User License gives employees access to Agentforce inside Salesforce for $5 per user per month. It ships with zero credits attached. Every action those users trigger draws down a Flex Credit balance you buy separately at $500 per 100,000.

That condition is the whole story. The licence line is small, and the meter under it is not capped by default.

🧮 The crossover, worked

The gap between the two routes is $120 per user per month. Divide that by the standard action rate:

$120 divided by $0.10 per standard action = 1,200 actions per user per month

Below roughly 1,200 actions a month, the $5 licence plus credits is cheaper. Above it, the $125 add-on wins. For voice-heavy users at $0.15 per action, the threshold drops to about 800 actions.

For a sales team, translate that into something you can count. 1,200 actions a month is about 55 actions per working day, per rep. A rep doing agent-assisted research, CRM updates, and follow-up drafting on ten live opportunities can get there. A rep who opens Agentforce twice a week cannot get close. If you want the workflow inventory behind that count, our breakdown of AI agents for sales teams lists the tasks that generate actions.

⚠️ The one thing procurement misses

The $5 line passes procurement without a second look precisely because it is small. That is exactly when consumption surprises happen. A $3,600 annual licence bill for 60 users can sit on top of a credit bill several times its size.

So the useful question is not "which licence is cheaper". It is "which of our users are heavy, and how heavy". Split the team. There is no rule saying one route has to cover everyone.

❌ An honest limit on the $125 decision

Salesforce documents what the Agentforce for Sales agents can do, and the current documentation is the right place to check the capability list before you buy. What documentation cannot tell you is whether those agents will perform on your data, in your process. Our summary of Agentforce for Sales features maps the shipped capability list against real sales workflows.

Availability is not proof of effectiveness. I would not overcorrect the other way either, because the capability is real and shipped. I would simply refuse to pay unmetered pricing for a workflow nobody has run yet. Buy the $5 route for the pilot cohort, measure actual action volume for a quarter, and move the heavy users to $125 once you have the number.

Q6. What is included in Agentforce 1 Editions and the current lineup? [toc=6. Editions and Bundled Credits]

Agentforce 1 Editions start at $550 per user per month and bundle Flex Credits at org level, not per seat. The Agentforce pricing page describes 2.5M annual credits. The Sales pricing page shows Max including 2.75M. Both were retrieved on 7 September 2026 and they disagree, so verify the figure in your own quote. Sales Cloud now publishes as Enterprise Core $195, Advanced $395, and Max $550, with 500K, 1M, and 2.75M credits included per org per year.

📦 The lineup and what comes with it

The critical word in that table is "org". Bundled credits are a pool for the whole company, not an allowance per seat. Ten users and two hundred users draw from the same pool.

⚠️ A source conflict I am not going to resolve for you

The Agentforce pricing page and the Sales Cloud pricing page were both live on 7 September 2026, and they publish different bundled credit figures for the top tier. One says 2.5 million. The other says 2.75 million.

I could pick the higher number and sound confident. That would be worse than useless, because the difference is 250,000 credits, or 12,500 standard actions of headroom. Ask your rep to state the figure in the order form, and hold them to it.

🕰️ Why so many articles quote the wrong prices

A lot of secondary coverage still lists Enterprise at $165 and Unlimited at $330. Those are the retired lineup. Review platforms are a common source of the staleness, because their pricing pages update on their own schedule rather than Salesforce's.

There is a simple tell. If an article names editions called "Enterprise" and "Unlimited" without mentioning Core, Advanced, or Max, its figures predate the current lineup. That matters because every scenario built on a wrong input is wrong at the input. The companion entitlement question, which edition includes which AI features, sits in our Salesforce Einstein pricing tiers explainer.

🧮 What bundled credits are actually worth

Convert the pool into actions before you decide anything. At 20 credits per standard action, 500,000 credits is 25,000 actions a year. One million is 50,000. 2.75 million is 137,500.

Now divide by your headcount and your working days. For a 50-person org on Enterprise Core, 25,000 actions a year is about two actions per person per working day. That is a pilot allowance, not a deployment allowance.

✅ The rule I would apply

Bundled credits are genuine value for an existing Salesforce customer, and I do not want to talk anyone out of that. They are also the easiest thing in the model to over-read.

Treat the pool as a floor, not a budget. Estimate your action volume first, subtract the bundle, and price the remainder at $500 per 100,000. If the remainder is zero, you have found real value. If it is large, the edition upgrade was not the saving you thought. Buyers weighing that upgrade against a separate platform should read our comparison of Agentforce alternatives and competitors.

Q7. What does Agentforce actually require before you turn it on? [toc=7. Prerequisites Reality Check]

Less than most articles claim. Salesforce's Agentforce pricing page documents no Data Cloud requirement, no Einstein activation step, no sandbox licence, and no mandatory professional services. Edition is conditional: Salesforce Foundations is free but limited to builder and configuration tools without agent deployment, and the add-on licences formally require an Enterprise-level edition. Widely repeated prerequisite costs are third-party reported, not Salesforce-published.

📋 Documented, conditional, or unverified

❌ The claims I could not verify

Several figures circulate as if they were requirements. Data Cloud at $25 to $50 per user per month. A $24,000 per org per year multi-org fee. Forty hours of training per user. Mandatory sandbox licensing.

None of these appears as a documented Agentforce requirement on the pricing page as of 7 September 2026. That does not make them false. It makes them unverified, and there is a real difference between the two.

🧭 Three things people confuse

Getting this right removes most of the confusion in the category:

1. Enabling a service. Turning something on inside Salesforce that you already pay for.
2. Using included capacity. Consuming the credits bundled with your edition.
3. Buying more capacity. Purchasing additional Flex Credits or an add-on licence.

Most "hidden cost" claims are really category three being described as category one. Your configuration may genuinely need Data Cloud capacity for grounding, which means retrieving the right company data to answer with. That is a real cost when it applies. It is not a universal prerequisite. The data-architecture side of that question is covered in our guide to agentic AI implementation for RevOps.

⚠️ What Foundations does and does not let you do

Foundations is free and gives you builder and configuration tools plus 100,000 Flex Credits per org per year on an Enterprise-level edition. What it does not do is let you deploy agents in production without a paid route.

Saying Agentforce "cannot function on basic Salesforce editions" is too blunt. The accurate version is narrower: you can build and configure on Foundations, and you need a paid licence or credit purchase to deploy. Our overview of what Salesforce Agentforce is explains what sits inside each of those states.

✅ The Monday version of this section

Take this table into your next call with your Salesforce account executive. Ask them to confirm, in writing, which prerequisites apply to your specific configuration and which do not.

Then ask the one question the pricing page does not answer. What happens when your credit balance reaches zero? Salesforce's Agentforce pricing documentation does not describe throttling, queueing, or suspension behaviour, and I would not deploy an agent into a live customer workflow without knowing the answer.

Q8. What sits outside the rate card? [toc=8. Costs Outside the Rate Card]

Three cost categories never appear on the pricing page: implementation and configuration effort, the ramp before agents produce anything, and any Data Cloud capacity your grounding actually needs. All widely quoted figures for these are third-party reported rather than Salesforce-published, so treat them as ranges to test against your own quote, not as inputs to a total.

🔧 Implementation and configuration

Salesforce publishes no rate for professional services, data migration, integration work, or training. Third-party sources report implementation in the range of a few thousand dollars per agent, with deployment windows measured in weeks rather than days.

I am tagging every one of those figures as reported, not published. They are useful as a sanity check on a partner quote. They are not inputs you should add into an annual total and present to a CFO.

⏰ The ramp before anything works

The second cost is time. An agent that is licensed on day one is not producing on day one. It needs process documentation, tool permissions, and a few cycles of correction before anyone trusts its output.

This is the cost that never lands in a spreadsheet and always lands in a quarter. If you are building a business case, put the ramp in explicitly as months of delayed benefit rather than pretending it is zero.

For deployment cost and timeline in detail, our Agentforce implementation guide covers the sequencing. I am deliberately not rebuilding it here.

💸 Data Cloud grounding capacity

Grounding means giving an agent the right company data to answer from. Some configurations need Data Cloud capacity to do that well, and that capacity is billed separately from Agentforce.

Here is the honest position. The Agentforce pricing page does not document Data Cloud as a requirement. Independent coverage reports it as a significant real-world cost for full functionality, with figures that vary widely across sources.

Both of those things are true at once. Your answer depends on what your agents need to read, so ask the question in scoping rather than assuming either extreme. Teams sizing that dependency usually start with a CRM data quality audit before they scope credits.

⚠️ How to tell published from reported

Use a two-column test on any number you find, including the ones on this page:

• Published. It appears on salesforce.com or help.salesforce.com, and you can point to the page and the date you retrieved it.
• Reported. It appears in an article, a benchmark set, or a consultancy blog, and it describes someone else's deployment.

Reported figures are worth reading. They are not worth adding up. A total built from six reported ranges is a fiction with a decimal point on it.

✅ What to do with this before you sign

Ask your Salesforce account executive to quote implementation and any Data Cloud capacity as separate line items, not as a bundled "services" figure. Then ask what the quote assumes about your action volume, because that assumption is doing more work than any other number in the document. If you are weighing an in-house build against a bought platform, our build versus buy analysis for revenue AI frames the same trade.

For experience evidence on how deployments actually go, our Agentforce reviews analysis collects practitioner accounts. Read it alongside the rate card, not instead of it.

Q9. How do I estimate my annual bill before talking to a rep? [toc=9. Estimating Your Bill]

Estimate actions per user per month, multiply by 20 credits for standard actions and 30 for voice, then price at $500 per 100,000 credits. Add your licence route, subtract credits bundled with your edition, and wrap a Pre-Commit or spending cap around the result. Volume assumptions are the weak point, not the rates, so publish a low, expected, and high case rather than one number. The rates come from Salesforce's Agentforce pricing documentation, retrieved 7 September 2026.

💰 The CFO wants a ceiling, not a rate

Every finance leader I have worked with asks the same question about consumption software. Not "what does it cost per unit", but "what is the most this can cost me". Rates are easy to find. Ceilings require you to bound the volume.

That is the whole job of this section. You are not predicting usage. You are bounding it, then buying a mechanism that enforces the bound. If you want a structured way to model the return side of that bound, our revenue intelligence ROI calculator walks through the inputs.

❌ Why seat-based budgeting fails here

The old method was headcount times price. It produced a number finance could hold, and it was wrong in a stable direction.

With Agentforce, the same method produces a number that is wrong in an unstable direction. Two teams of the same size can differ by an order of magnitude in action volume, and nothing in the seat count tells you which team you are.

🧮 The five-step method

1. Count actions, not people. Pull thirty real workflows and count the discrete steps each one takes.
2. Split standard from voice. Multiply standard actions by 20 credits and voice actions by 30.
3. Price the credits. Multiply total credits by $0.005, which is $500 per 100,000.
4. Subtract the bundle. Remove the credits included with your edition, at org level, per year.
5. Add the licence line. Add $5 or $125 per user per month, depending on your route.

(standard x 20 + voice x 30) x $0.005 = consumption cost

🔢 Two illustrative scenarios

These volumes are invented for the arithmetic. Do not adopt them.

A 25-user team, 200 standard actions per user per month. That is 60,000 actions a year, or 1.2 million credits. Subtract a 500,000 credit bundle, leaving 700,000 credits at $3,500. Add licences at $1,500. Annual estimate: about $5,000.

A 150-user team, 400 standard actions per user per month. That is 720,000 actions a year, or 14.4 million credits. Subtract a 1 million credit bundle, leaving 13.4 million at $67,000. Add licences at $9,000. Annual estimate: about $76,000.

Now test the same team against unmetered pricing. 150 users at $125 a month is $225,000 a year. At 400 actions per user, metered is cheaper by a wide margin, which matches the 1,200-action crossover. Teams running this exercise across several tools usually find it doubles as a tech stack consolidation review.

⚠️ Compare incremental to incremental

The most common error in this category is a comparison table that puts one vendor's cheapest tier against Agentforce plus the entire Salesforce estate. Those totals are arithmetically consistent and completely meaningless.

If you already pay for Sales Cloud, the Salesforce licence is a sunk line. Your Agentforce decision is the incremental spend on top of it. Compare that incremental number against another tool's incremental number, or compare full estate against full estate. Do not mix the two. Our list of Agentforce alternatives and competitors is built on that same incremental basis.

What genuinely needs a quote: your bundled credit figure in writing, any Data Cloud capacity your grounding requires, and the behaviour when the balance hits zero.

Q10. What happens to your budget when the pricing model changes again? [toc=10. Budget and Contract Risk]

Salesforce has shipped three Agentforce billing models since 2024, moving from per conversation, to Flex Credits, to per-user editions, while Agentforce 1 Editions bill annually only. That combination puts re-forecasting risk on the buyer. Independent 2026 research also reports that most agent deployments never reach production, and that a meaningful share of those that do return negative ROI at twelve months. That is the strongest lever a buyer has for a shorter term or a re-pricing clause.

⏰ The renewal nobody plans for

The conversation I keep seeing goes like this. A RevOps lead builds a solid model in month one, signs a twelve-month deal, and then the metering model changes in month seven.

Their model is not wrong. It is measuring something the vendor no longer sells the same way. Nobody planned for that, because software pricing used to change once every few years, not twice in eighteen months.

🕰️ Three models in under two years

The $2 per conversation model arrived with the original launch. Flex Credits landed in May 2025 and shifted billing from the session to the action, as documented on the Salesforce Agentforce pricing page. Per-user editions and add-ons then layered seats back on top.

I want to be fair here. Each change moved toward finer granularity, which is generally better for the buyer. The problem is not the direction. It is that a signed annual commitment cannot absorb a change in the unit of billing.

📉 What the agent-ROI data says

Independent 2026 compilations of enterprise agent research report two numbers worth carrying into a negotiation. A large majority of agent deployments never reach production. And of those that do, roughly one in five shows negative ROI at the twelve-month mark.

I am tagging both as reported rather than published, because they come from aggregated industry research rather than a single primary dataset. Directionally, they match what I see. Most agent programmes stall on process documentation, not on model quality, which is the failure mode our guide to agentic AI implementation for RevOps is written against.

✅ Four asks that turn statistics into clauses

Base rates only matter if they change what gets signed. So bring these four to the table:

1. A spending cap. A hard ceiling on consumption, with a written description of what happens at the cap.
2. A re-pricing clause. If the metering model changes mid-term, you get to re-elect your route without penalty.
3. A stated rollover position. Credits expire at the order end date, so get any exception in writing.
4. A shorter first term. Twelve months is standard. Ask for a shorter initial term on the first agent workload.

The rollover ask matters more than it looks. An over-purchase is not savings carried forward. It is money that disappears on a date in your contract. Buyers running a formal evaluation should pair these asks with the checks in our AI CRM trust and governance evaluation guide.

💰 The counterweight, stated plainly

None of this means the pricing is unfair. For a company already deep in Salesforce, the credits bundled into Core, Advanced, and Max are real value that arrives with an edition you were buying anyway.

Salesforce also publishes its per-action credit cost openly, which is more than most vendors in this category do. My argument is narrower. Publish your assumptions, cap your exposure, and keep the term short until you have a quarter of real consumption data.

Q11. What compliance and governance costs land in the same budget cycle? [toc=11. Compliance and Governance]

From 2 August 2026, EU AI Act Article 50 transparency obligations apply to AI systems interacting with people, and the European Commission's final guidelines bring agents acting on a user's behalf into scope. Deployers must ensure the agent discloses that it is artificial and on whose behalf it acts, with penalties reaching EUR 15 million or 3% of worldwide turnover. Budget for disclosure changes to outbound email and voice flows alongside the licence line.

📅 What Article 50 requires, and from when

Article 50 of the EU AI Act (Regulation 2024/1689) is the transparency layer. It applies from 2 August 2026, and the Commission published final guidelines on transparency obligations in July 2026.

The guidelines are explicit that agents fall in scope where they interact with the people instructing them, or with other people while carrying out a task. Examples given include agents making bookings and handling correspondence, which is exactly the sales and service surface Agentforce sits on. The European Commission AI Act Service Desk confirms the 2 August 2026 application date for these rules.

🧾 The practical checklist

Disclosure is not one banner. It attaches to each surface where a person meets the agent:

• Outbound email. The recipient should be able to tell an agent wrote or sent it, and on whose behalf.
• Voice. Disclosure needs to happen at the start of the interaction, not in a footer nobody hears.
• Chat and web. Clear labelling that the counterpart is an AI system.
• Internal handoffs. Log which agent took which action, so you can reconstruct a decision later.

None of that is expensive on its own. It becomes a cost when it forces sequence rewrites, script changes, and legal review across every template your team already uses. Scope that work in the same quarter as the licence, not the one after. Mid-market buyers can borrow the control list from our mid-market revenue AI buyer guide.

⚠️ Where this hits the pricing conversation

No Agentforce pricing page connects disclosure duties to deployment cost. That is not a criticism of Salesforce specifically. No vendor pricing page in this category does it.

The gap matters because both bills arrive together. You license agents in Q1, deploy them in Q2, and discover the compliance work in Q3 when someone in legal reads the guidelines. Put a line in the budget for it now, even if the number is a placeholder.

💸 The internal half: spend governance

Regulatory governance has an operational twin, and consumption billing makes it mandatory. You need to know which agent spent what.

Three controls to specify before deployment:

1. Per-agent spend tracking. Attribution at the agent level, not just the org level.
2. Limits and high watermarks. Alerts before the balance moves, not after.
3. A pre-deployment estimate. A small proof of concept that sizes credit consumption before you scale the agent.

Oliv AI applies this same split internally, tracking spend per agent, running a pre-deployment proof of concept to size credit cost, and exposing settable limits and high watermarks. That is our own product description rather than an independent evaluation, and I would hold any vendor, including us, to showing it working in your org before you rely on it. The admin-side setup is documented in our RevOps implementation and admin guide.

The uncomfortable version of this section is simple. An agent you cannot attribute spend to is an agent you cannot govern, and an agent that does not disclose itself is a legal exposure with a licence fee attached.

Q12. If the bill moves monthly, should you just buy a flat seat price? [toc=12. Predictable vs Metered]

Sometimes. Metered and per-seat billing fail differently. Consumption punishes unpredictable volume, and seats punish light usage. Oliv AI is priced per seat and sits on top of the Salesforce record rather than replacing it, so the CRM stays the system of record while agent work bills on a line finance can forecast. The honest counterweight: a light-usage team can pay less under Salesforce's PayGo route than under any per-seat licence, and Oliv has no public review corpus yet, so a buyer needing reference-checkable proof will not find one.

💰 The question finance actually asks

The trigger is usually a board meeting. Someone asks what the AI line will be next quarter, and the honest answer is "it depends on how much the agents work".

That answer is true and unusable. It is also the reason this question gets asked in every Agentforce evaluation I have been near.

⏰ What metering costs you in effort

The rate is not the expensive part. The forecasting is. Someone has to watch consumption weekly, reconcile it monthly, and explain variance to finance quarterly.

For a small RevOps team, that is real hours against a person who was already stretched. I would rather spend those hours on pipeline hygiene, but the meter does not care what I would rather do. That trade is the same one our comparison of revenue intelligence platforms keeps running into.

⚖️ The structural difference, without a total

A per-seat layer changes the shape of the bill, not just its size. You know the number when headcount changes, and only when headcount changes.

That is the entire structural argument, and I am deliberately not attaching a saving percentage to it. Any number I quoted would depend on your action volume, which neither of us knows yet.

⭐ Where a layer on top of Salesforce fits

Oliv AI connects to the Salesforce record the team already keeps, reads meetings, email, calls, and connected data sources against it, and runs named agents, Olivia as orchestrator and Oliver as agent builder, on plain-English SOPs with per-agent approval gating and spend tracking. Oliv states that it maintains a context graph across opportunities, accounts, and contacts, which resolves which meeting belongs to which opportunity in messy CRMs. That is our own claim, supported by the actions the agents take and the inputs they require, not by third-party evaluation. The difference relevant to this page is the billing shape: seats rather than actions. I am not publishing a total or a savings figure here, because our canonical price list is being reconciled. The agent surface itself is documented in our overview of Oliv AI agents for sales teams.

❌ The concessions I owe you

Three of them, stated plainly. A light-usage team can genuinely pay less under Salesforce's PayGo route than under any per-seat licence, and PayGo exists precisely for that team. The credits bundled into Core, Advanced, and Max are real value if you are already buying the edition. And Oliv has no public review corpus yet, so if you need reference-checkable proof from strangers, you will not find it with us today.

If you already run Salesforce and just need the cost model, everything above should be enough to build your estimate. If you want to see what an agent layer looks like running on your own opportunity data, book a demo and bring one messy account with you.