What your edition already includes, and what arrives as a separate line. 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 fourteen, Einstein across the clouds. Last session we priced an agent. Today we do something less glamorous and, in my experience, worth more money: we sort out what you already own from what you are being asked to buy again. Salesforce has attached the Einstein name to a decade of features, some of which ship with your edition and cost nothing extra, and some of which run a meter. The brochure does not draw that line. So today: the two kinds of AI and the one test that separates them, what your edition actually includes, the four different things the word included can mean, the inventory that finds duplication, and where estates end up paying twice. Three knowledge checks. Let's begin.
Five objectives. First, separate the two kinds of AI: predictive features that ship with an edition, and generative features that meter. Second, say what your edition includes, specifically and by name, rather than accepting that AI is in there somewhere. Third, read the word included properly, because it usually means a capped allotment and the cap is the part nobody quotes. Fourth, build an AI inventory: every AI line you pay for, what it does, and whether something else already does it. And fifth, stop paying twice, because the most common AI overspend is buying a capability the edition already granted you.
So, two kinds of AI. Predictive: scoring, forecasting, recommendations, usually bundled at the higher editions. Generative: drafting, summarising, conversing, and almost always metered by consumption. Flat or metered is the only test that matters, meaning does using it twice as much cost twice as much. And same brand, because both are sold under one name and the brochure rarely draws the line for you. Let me explain why that single test is worth more than any amount of product knowledge.
Guest analyst clip.
So the first question in any AI conversation with an account team is not what it does. It is whether the line is flat for the term or moves with usage. Ask that first, before the demonstration, because after the demonstration everybody is talking about capability and nobody is talking about the meter.
Right, what the editions include, and treat this as the shape rather than your answer. At Professional there is little to none, and AI is largely an add on at that tier. At Enterprise you get a set of predictive features plus a modest generative allotment. At Unlimited, a wider predictive set and a larger allotment. And at the AI editions, a bundled consumption allotment on top of a higher seat price, which is the entire point of that tier. Where to confirm it: the order form first, then the Product Terms version it references, and the entitlement screen in the org as a cross check. The bundles move between releases, so anything you read in a guide, including that table, is the shape and not the answer.
First knowledge check. An account team says generative AI is included in your edition. What is the next question? A, which features exactly, and what is the allotment before it meters. B, when was it added to the edition. C, does it use the latest model. D, can we see a demonstration. Pause here and pick an answer before you continue.
A. Included nearly always means included up to a limit, and the limit is the commercial fact, so a question that does not ask for the number has not really been asked. B is history. C is a technical detail that changes with releases and rarely changes the bill. And D shows you the capability while telling you nothing about the entitlement, and demonstrations are exactly where the distinction between bundled and metered goes quiet.
So what arrives separately. Five lines that are their own purchase. Agent conversations, priced per conversation or through credits, as we covered last session. Consumption above the allotment, because the included generative volume runs out and what follows is metered. Data Cloud underneath it, since grounding AI in your data is a Data Cloud workload with its own credits, which is the connection people miss most often. Specialist AI products, where marketing, commerce and industry specific AI often sit outside the core bundle. And the enterprise AI agreement, an umbrella commitment that buys scale and locks a floor, which you should read as a commitment rather than as a discount. Now, that word included.
Guest analyst clip.
Four uses, and only one of them is what you heard. Genuinely bundled: flat for the term, no usage cap, turning it on costs nothing further. Included up to a limit, which is the common case, so a stated allotment with consumption pricing beyond it. Included in the tier above, which is true of the product family and not of your edition, and that is a different statement.
And included in the roadmap, meaning announced rather than available. Never accept a roadmap item as an entitlement, and if it matters to the deal, get it written as a commitment with a date, because otherwise you have bought a plan.
Second knowledge check. You are quoted a new AI add on. What is the cheapest thing to check first? A, whether the price is competitive against other vendors. B, whether your edition already entitles you to the same capability. C, whether the model is the newest available. D, whether the contract term aligns with your renewal. Pause here before you continue.
B. The cheapest negotiation is the one where you discover you already own the thing, and this happens more often with AI than with any other Salesforce line, because the naming shifts between releases and the buying centre is rarely the licensing team. A prices a purchase you may not need. C is a specification question. And D is genuinely a good question, and it comes second, because term alignment on an unnecessary line is tidy housekeeping around a wasted purchase.
Which brings me to the inventory, one table, and most estates have never built it. Every AI line you pay for, taken from the order forms rather than from memory, including anything bundled into an edition. What each one does, in a sentence a business person recognises rather than the product name. Flat or metered, which is the single most useful column and the one that predicts next year's surprise. Who uses it and how much, because unused AI is the easiest reduction you will ever propose. And what else does this, meaning overlap with another Salesforce line or with a tool the business already owns. Let me tell you what building one usually turns up.
Guest analyst clip.
Half of them will not be recognised. That is the normal result, and it is not a sign of a badly run organisation. It is what happens when a product family grows by acquisition and rebranding faster than any procurement function can track.
So, where the duplication usually hides. The edition already had it, meaning a capability bundled at your tier and bought again because it was renamed. Two clouds, one function, where Sales and Service each ship their own version of the same assistance. The add on and the bundle, where an AI edition was purchased while the standalone add ons stayed on the order form, which is the most expensive version of this and one of the most common. The platform outside Salesforce, where your organisation already licenses the capability from another vendor entirely. And the pilot nobody closed, a trial that converted to a paid line and never appeared in anybody's review.
Five failures. Included taken at face value, with no allotment written down, so the first metered invoice is the first real information anybody has. Bought by the business, meaning an AI line agreed in a product conversation that never reached the licensing team. No adoption measurement, so AI lines renew for years without anybody checking whether they are used. The edition upgrade justified by AI, where a whole tier is bought for one capability that could have been an add on. And the umbrella agreement read as a discount, when an enterprise AI commitment is a floor, and floors are hard to walk back.
Last knowledge check. Which AI line is easiest to reduce at renewal? A, a metered line running above its allotment. B, a per user AI line with measured low adoption. C, a capability bundled into your edition. D, an enterprise AI commitment mid term. Pause here and pick an answer before you continue.
B. A per user line with adoption data behind it is the cleanest reduction in the whole agreement, because the quantity is yours to set at renewal and the evidence is not arguable. A is running hot, so the conversation there is about rate and headroom rather than reduction. C is not a line at all, so there is nothing to remove without changing edition. And D is a commitment you signed, where mid term relief is possible and it is a concession you will pay for somewhere else. Let me set out the review that keeps this current.
Guest analyst clip.
Twice a year, and it is short. Read the inventory out loud, line by line, with the business owner present. Check adoption per line, active users against licensed users, and consumption against allotment. Re-test the duplication question, because the answer changes with every release and with every other purchase you make. Mark each line keep, reduce or remove, before the renewal window, so your position exists before the proposal arrives rather than being assembled in response to it. And keep the allotments in one place, so the word included never has to be reinterpreted from somebody's recollection of a conversation again.
Three sentences. The word Einstein covers two commercially different things, and the only test that matters is whether using the feature twice as much costs twice as much, because predictive features generally ship flat with an edition while generative features almost always run a meter. Included nearly always means included up to a limit, so the useful question is never whether AI is in your edition but which features exactly and what the allotment is before consumption pricing starts. And the largest AI saving available to most estates is not a better rate, it is discovering you already own the capability, which is what the inventory is for.
Homework before session fifteen, about two hours. One, list every AI line on your order forms, including anything described as bundled or included with an edition. Two, find the allotment for each, the number at which it starts to meter, and where you cannot find one, write unknown, because unknown is a finding. Three, mark flat or metered in one column, and that column will tell you where next year's surprise is coming from. Four, pull adoption per line, active users against licensed users, for the last quarter. And five, find one duplication: two lines doing the same job, or one line doing what the edition already granted.
Five guides, all on redresscompliance dot com. Salesforce Einstein AI licensing sets out what sits in the editions and what is sold alongside them. The Salesforce AI credits consumption model covers how the metered half behaves once the allotment is gone. Salesforce AELA, the AI enterprise agreement, covers the umbrella commitment and reading it as a floor rather than a discount. Salesforce Enterprise against Unlimited shows what the tier step actually buys, AI included. And Salesforce hidden costs covers the lines that arrive after signature and how to find them early.
That is session fourteen. The thing to take away is that with AI the cheapest negotiation is the one you avoid, and you avoid it by knowing precisely what your edition already entitles you to before anybody quotes you for it again. Next time we close module three by putting controls around all of this: governing AI and data spend, who owns what, and the monthly page that stops a surprise becoming an incident. See you then.