Conversation based pricing, Flex Credits, and sizing an agent programme honestly. Three knowledge checks along the way, and 4 clips from a senior licensing analyst.
This is a taught session, not a talking head. The instructor works through analyst grade slides, and three times the video stops on a question with four options on screen. Pause, commit to an answer, and the next slide explains which option is right and why each of the others is wrong. 4 times in the session the frame splits and a senior licensing analyst gives the view from inside real SAP negotiations, and the instructor picks the clip apart when the slides return.
The full narration of this session, section by section, for reading and reference. Guest analyst clips are marked.
Welcome back. Session thirteen, Agentforce. Two sessions ago we met a meter that runs on rows. Today we meet one that runs on conversations, and it introduces a problem we have not had before: you cannot count the unit yourself. A user is a person you can list. A row is a number in a source system. A conversation is an event the platform decides has occurred. So today: the unit and why its definition belongs in your contract, Flex Credits and what the shared pool costs you, when a bundled edition is a discount and when it is a floor, and how to size an agent programme from containment rather than from ambition. Three knowledge checks. Let's begin.
Five objectives. First, name the unit you are buying, which is not a seat and not a row but a conversation, defined in the vendor's documentation. Second, interrogate that definition: where a conversation starts, when it ends, and what an abandoned session costs you. Third, read the Flex Credit pool, which is a shared consumption currency and a flexibility you pay for in expiry. Fourth, size from containment rather than volume, because total case volume is the wrong number and what the agent actually handles is the right one. And fifth, ask for the terms that matter: a written definition, a testing exclusion, a ramp, and a capped rate for the overage.
So, the unit. Four things about it. It is vendor defined, because the boundaries of a conversation live in product documentation rather than in your contract. You cannot audit it, since there is no independent record and their meter is the only meter in the room. Volume is a guess, because nobody knows how many conversations an agent will hold until it is live. And value is per case, which is the saving grace, because the unit maps to a deflected contact you can price. Let me take the second of those, because it is genuinely new territory for most buyers.
Guest analyst clip.
That last point is the one to hold onto. Unlike a credit, an agent conversation has an obvious business comparator: what the same contact costs when a human handles it. That comparator is the strongest thing you have here, and we will come back to it.
Right, five questions to put in writing before signature. Where does it start? A page load, a greeting, or the first user message, and you want the first user message rather than the agent's opening. When does it end? Idle timeouts can create two conversations out of one, so press for a stated window and a resumed session counting once. Abandoned sessions: a user who opens and leaves may still bill, and you want no charge where the agent did no work. Escalated to a human: the agent failed and you may still be billed, so ask for it reduced or excluded when the agent hands off. And testing: pre launch work can consume real volume, so get a non production exclusion or a test allocation. None of these are unreasonable, none are granted by default, and every one is easier to win before signature than after the first invoice.
First knowledge check. Which of these is the most expensive definition to leave unwritten? A, whether an escalated conversation still counts. B, the colour of the chat widget. C, which channel the agent runs on. D, how many agents you configure. Pause here and pick an answer before you continue.
A. Escalation is the failure case, and in the first year a large share of conversations escalate, so leaving it unwritten means paying full rate for the agent's unsuccessful attempts on top of the human handling you still pay for. B is not a commercial question. C matters for design and rarely changes the unit price. And D is free, because configuring agents costs nothing while running conversations through them is the meter, and confusing the two is how programmes get sized on entirely the wrong number.
Flex Credits next. One pool, spendable in several directions. It is a shared currency, so one balance is drawn on by agent actions and by other consumption in the platform. That flexibility has a price, because you pay for the optionality and the pool still expires at the end of the term. Actions vary in cost, so not every agent step is the same and the rate card matters as much as the balance. It hides attribution, since a single pool makes it much harder to see which programme consumed what. So split the reporting: even where the pool is shared, report consumption by use case from the first month, because you will need that split long before anybody asks you for it. Now, bundles.
Guest analyst clip.
Utilisation is the whole question. So to put it plainly: the bundled edition is a per user price with an allotment of consumption included, sold as simplicity. It is a discount if you use it, because at high utilisation the blended rate beats buying the pieces separately. And it is a floor if you do not, because below the allotment you have prepaid for volume that expires quietly.
And note that the per user half persists regardless. The seat charge continues whether or not a single conversation happens, which means a stalled agent programme on a bundled edition is still costing you the user licences every month while the consumption you paid for goes unused.
Second knowledge check. Your service desk handles two hundred thousand contacts a year. What is the right basis for sizing an agent programme? A, two hundred thousand, since every contact could reach the agent. B, the share the agent will actually contain, evidenced by a pilot. C, the vendor's benchmark containment rate for your industry. D, whatever volume the bundled edition includes. Pause here before you continue.
B. Total contact volume is the size of the opportunity rather than the size of the purchase, and early containment is typically a fraction of it while the agent learns your content and your edge cases. A buys the ceiling on day one. C is somebody else's data about somebody else's content, offered by a party with an interest in the number being high. And D reverses the exercise entirely, letting the packaging decide the programme, which is precisely how a bundle stops being a discount and becomes a floor.
So here is the sizing method, and the first step is not a Salesforce question at all. One, start with contact volume by channel, from your own service platform, by month, for the last full year. Two, identify the addressable slice, meaning the intents the agent can genuinely handle with the content you actually have rather than the content you intend to write. Three, apply a measured containment rate, from your pilot, on your content. Four, add the failure traffic, because escalated attempts still consume unless you negotiated otherwise. And five, ramp it across the term, since containment improves with tuning so the volume curve should rise rather than start flat. Let me say why step three carries all the weight.
Guest analyst clip.
Not a benchmark and not an aspiration. And notice what that does to the negotiation: a containment rate you measured is evidence, and a containment rate from a deck is a forecast made by the party being paid. Those two things should not carry equal weight in a commitment.
So, the containment rate itself, which is the single number the whole business case rests on. It is measurable: conversations resolved without a human, over conversations started, nothing exotic. It starts low, because early containment on unfamiliar content is usually well below the number in the deck. It improves with work, meaning content, intents and guardrails, which is effort you must resource or the curve stays flat. It caps the value, since the saving is contained contacts times the cost of a human contact and that is the ceiling on the whole programme. And measure it monthly, because it is also the number that tells you whether to expand or to stop.
Five failures. Sized on total volume, so a commitment covering every contact against an agent that will contain a fraction of them. The definition left to documentation, meaning a unit that can be reinterpreted in a document the vendor maintains alone. Testing billed as production, with months of tuning consuming the volume you bought for customers. No content owner, because containment is a content problem and nobody was resourced to fix the content. And escalations paid twice, where the agent's failed attempt is billed and the human handling is paid for as well.
Last knowledge check. Six months in, containment is half the assumption. What is the first move? A, buy more conversations to cover the shortfall in savings. B, find out whether it is a content problem or a scope problem, then fix the cause. C, switch the agent off and write the programme down. D, wait for the vendor's next model release to improve it. Pause here and pick an answer before you continue.
B. Low containment has two common causes and they need opposite responses: missing or poor content, which is fixable with effort you can resource, or intents the agent was never suited to, which means narrowing the scope rather than improving it. A spends more on a mechanism that is not working. C throws away a programme before diagnosing it, when the escalation data you already hold usually names the cause in an afternoon. And D outsources your business case to a release note. Let me describe the page that keeps you ahead of this.
Guest analyst clip.
Monthly, and it answers one question. Conversations against commitment, consumed and plotted monthly, with the projected date the allowance runs out. Containment rate as a trend rather than a month, because one month tells you nothing about direction. Escalation reasons ranked, so the top five intents that failed, which is your work list for next month. Cost per contained contact, meaning consumption cost over contained conversations, set against your human cost per contact. And the decision line: expand, tune, or narrow, written down every month by the person who owns the budget, because an agent programme without a monthly decision drifts for a year and then gets cancelled in one meeting.
Three sentences. Agentforce prices a unit you cannot independently count, defined in documentation the vendor maintains, so the five definitional questions belong in your order form: where a conversation starts and ends, and how abandoned sessions, escalations and testing are treated. Size the programme from containment rather than contact volume, because total contacts describe the opportunity while contained conversations describe the purchase, and early containment on your own content is usually well below the number in the deck. And the business case rests entirely on the containment rate, which is measurable monthly and improves only with content work somebody has to be resourced to do.
Homework before session fourteen, about ninety minutes. One, pull contact volume by channel, twelve months, by month, from your service platform rather than from a summary deck. Two, find the conversation definition, the current documentation, and check whether your contract references a version of it. Three, calculate your human cost per contact, fully loaded, because this is the comparator the whole business case needs. Four, get your containment rate: if you are live, measure it, and if you are not, note plainly that you do not have one yet. And five, list the definitional asks, the five from earlier, written as the clauses you will table at renewal.
Five guides, all on redresscompliance dot com. Salesforce Agentforce licensing twenty twenty six covers the pricing mechanics, the units and the terms that matter. Salesforce Agentforce pricing twenty twenty six shows what the conversation and credit rates look like in practice. Agentforce one Edition, is it worth it, works through when the bundle is a discount and when it is a floor. Is Agentforce worth the price covers containment, cost per contact and where the return actually comes from. And Salesforce AI pricing negotiation sets out the asks that get accepted on the AI lines, and the order to table them in.
That is session thirteen. The thing to take away is that this is the first Salesforce unit you cannot count for yourself, so the definition is the deal, and the containment rate is the business case. Next time we sort the AI you already own from the AI you are being asked to buy again: Einstein across the clouds, what your edition genuinely includes, and what arrives as a separate line. See you then.