What Prime actually gates, what the Control Tower is for, and the public AI target that turns your adoption into discount points. Three knowledge checks along the way, and 3 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. 3 times in the session the frame splits and a senior licensing analyst gives the view from inside real ServiceNow 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 nine, and this is the last of the AI sessions before we close module two. We have done the tiers, we have done the migration, we have done the meter. Today, the layer sitting on top of all of it. Agents. What Prime genuinely gates and what it does not, what an agent actually costs once you count everything rather than just the licence, and then something I have been saving, because it changes the posture of the whole negotiation. ServiceNow has told its investors, publicly, with dates attached, how much AI revenue it has to book and by when. Once you know those numbers, the AI conversation stops being a product pitch you are receiving and becomes a transaction where you happen to hold something they urgently need. That reframing is worth more than any technique in this course, and it is available to every one of you for the price of reading an earnings call. Three checks, homework, let's go.
Five objectives. First, separate the layers, telling apart assisted AI, agentic workflows, and fully autonomous agents, because those three get spoken about as one thing and they sit at different prices. Second, price an agent properly, across its licence, its pool burn, and the governance it demands, rather than just the seat on the quote. Third, read their targets, so you can explain ServiceNow's published AI commitments and why they change what your adoption is worth in the room. Fourth, trade deliberately, selling the attach, the logo, and the reference for pool size, rate caps, and clause protection rather than handing them over for goodwill. And fifth, govern the agents, keeping an inventory with owners, scopes, and ceilings, because agents are the first thing on this platform that spends your money without a person present. That last point is genuinely new. Everything before this course covered was consumed by humans. Agents are not.
Four numbers, and none of them are about your estate. One point five billion dollars. That is the AI ACV target management raised from one billion, and told investors remains on track for the end of calendar twenty twenty six. Thirty percent. The share of total ACV that has to come from AI by twenty thirty, published alongside a thirty billion plus subscription revenue goal. That is not an aspiration in a keynote, it is a mix ratio the board has put in front of the market. Seven hundred fifty million. Now Assist ACV disclosed at roughly that level in the first quarter of twenty twenty six, up from six hundred million at the end of twenty twenty five, so you can see the ramp and you can see how steep the remainder has to be. And four point eight billion in cloud infrastructure commitments through twenty thirty, which is the cost side, already contracted, already spent in commitment terms. Put those together. Several hundred million dollars of net new AI ACV has to land in signed contracts inside a small number of quarters, against infrastructure that is already paid for. Your rep carries a fragment of that number, and that fragment has your name on it. Let's hear what that means practically.
Guest analyst clip. Most buyers read the AI pitch as enthusiasm. The rep is excited, the deck is glossy, the demo is impressive, and it all reads as a company that believes in its product. And they do believe in it. But that is not why the pitch is arriving in your renewal with that particular intensity, and understanding the difference changes how you sit in the chair. The intensity is there because AI attach is a disclosed metric with a public number and a public date, and disclosed metrics behave differently inside a company than ordinary product goals. They cascade. The number is committed to the market, it becomes a regional target, then a territory target, then a line on your account executive's plan, and every one of those people is now looking for signed AI attach before a specific date. So when your rep pushes AI hard, that is not a preference. That is a constraint they are operating under. And constraints are tradeable in a way preferences are not. The practical consequence is straightforward. Something you may have been treating as an annoyance, the pressure to adopt AI, is actually the strongest card in your hand this cycle, because you can supply the thing they are measured on. The only mistake is to supply it for free, which is what happens when a buyer just says yes to the tier and never realises what they handed over.
Constraints are tradeable in a way preferences are not. That is the sentence. And notice the trap identified at the end, the buyer who simply says yes to the AI tier has supplied the scarce thing and received nothing specific in exchange, because they never understood they were holding anything. Every buyer in this position gets asked for AI attach. Only some of them get paid for it. Now let's be precise about the product, because you cannot trade well on a layer you do not understand.
Three levels of AI, four rows on the slide. Assisted, that is summaries, suggestions, drafted replies, Virtual Agent, and the defining characteristic is that a person is doing the work and the AI is helping. Any tier, including Foundation, low burn per action. Agentic workflow, where a task chains several actions and completes without step by step direction, available below Prime in workflow form, medium to high burn per task. Autonomous agent, where the agent decides and acts across a task end to end, on a schedule or a trigger, no human in the loop. Prime only, high burn, several times an interactive prompt. And custom AI skills, building your own rather than consuming what ships, also Prime only, with burn depending entirely on what you build. Now the note underneath, which you have heard from me twice already and will hear once more. The Control Tower and Workflow Data Fabric sit under all of these and ship in every tier. So what does Prime sell you? The bottom two rows. Autonomy, and the right to build your own skills. That is the product boundary, and where it applies it is genuinely worth paying for.
So what does the Prime gate actually gate? It gates autonomy, agents acting end to end, and that is a real boundary. It gates skill building, creating your own skills rather than consuming shipped ones. What it does not gate. It does not gate the platform, Control Tower and Data Fabric ship everywhere, so governance tooling is never a reason to buy Prime, it is a reason to use what you already own. It does not gate assistance, summaries and suggestions and Virtual Agent all run on Foundation, and I want you to sit with that one because it has commercial consequences. Most of the day to day value in the demos people fall in love with lives below the gate. The thing that made your leadership excited is quite often available at the entry tier. And the last point, the gate applies per population. It is a licence boundary, not an architectural one, so the team that runs agents needs Prime, and the people whose tickets those agents touch do not. That distinction is the entire first check.
Knowledge check one. You want forty service desk engineers running autonomous agents against incidents raised by six thousand employees. Who needs Prime? A, all six thousand and forty, because the agents touch everyone's records. B, the forty engineers who operate the agents. C, nobody, agents are bundled in every tier now. Or D, the forty engineers plus any approver in the workflow. Pause here, and ask whether the gate is about who is touched or who operates.
The answer is B, the forty engineers. The gate follows the operating population, not the touched population, and that is exactly the same logic as the fulfiller test from session two. What does this user do, not what happens near them. Answer A is uniform Prime wearing an agentic costume, and run the arithmetic at benchmark rates, roughly a hundred dollars a month of difference across six thousand people, that is a seven figure annual error, and it is an error I have watched get proposed with a completely straight face. Answer C confuses assisted AI, bundled everywhere, with autonomy, which is not. And answer D drags approvers up a tier for approving, which is the precise mistake session two spent an entire session unpicking, resurfacing here because agents make it sound new. It is not new. It is the same question.
What does an agent actually cost? Four axes, and only the first is on your quote. The licence, Prime for the operating population, visible and negotiable, and it is what everybody argues about because it is what everybody can see. The pool burn, several times an interactive prompt per run, on a schedule rather than a working day, and this is the line that grows after signature, quietly, because nothing about it appears in the negotiation you just finished. The governance, owners, scopes, ceilings, review, which is a real internal cost in somebody's time and is the only thing standing between axis two and unbounded compounding. And the value test, credits per resolved interaction, measured before you scale, which is session eight's discipline applied to agents specifically. Now the note, and it is the most important sentence on this slide. An agent that resolves work nobody valued is worse than no agent, because it converts an idle process into a metered one. Before automation that low value queue sat there costing you patience. After automation it costs you money, per action, forever. Automate what is worth automating, not what is easy to automate. Second check.
Knowledge check two. An agent pilot automates a high volume, low complexity queue and completes forty thousand tasks in a quarter. Everyone is delighted. What must you check before scaling it? A, nothing, forty thousand automated tasks is self evidently a success. B, credits per resolved interaction against the value of the work being resolved. C, whether the engineers enjoyed using it. Or D, whether ServiceNow will discount Prime further at that volume. Pause, and ask what volume proves and what it does not.
The answer is B. Volume proves the agent works. It does not prove the agent pays, and those are completely different claims. High volume low complexity queues are the easiest thing in any estate to automate and frequently the least valuable, which is exactly why they get picked for pilots, they demo beautifully. Scale one without the unit economics and you have multiplied a metered cost against work your business was tolerating perfectly happily for free. Answer C matters for adoption and not at all for economics. And answer D is my favourite wrong answer in this session, because it is the sophisticated mistake, negotiating price on a rollout you have not yet justified. That is precisely the sequence that produces consumption commitments outrunning value, and it feels like good procurement while you do it. Justify first. Then negotiate.
Now the trade, and this table is the commercial heart of the session. Left column, what they need. Middle, why. Right, what to ask for instead of giving it away. They need AI attach on your contract, because it books against a public target with a published date, so ask for pool size, a capped overage rate, and rollover. They need a recognizable logo on AI, because named customers de risk the story for the market and for the next buyer in your industry, so ask for discount points on the tier and price protection across the migration. They need a reference or a quote, which is reusable and therefore worth more to them than a single deal, so ask for clause protection, the definition freeze, swap rights, true down. And they need multi year AI commitment, because it counts twice, against this year's target and against the twenty thirty mix ratio, so ask for the longest price hold and uplift cap you can get, plus exit flexibility. And the note underneath, which I want to be very clear about. Nobody is asking you to like AI, or to pretend to want it. You are being asked to supply something with a public deadline attached, and things with deadlines are worth more than things without.
Guest analyst clip. Let me give you the shape of a trade that worked, because the abstract version sounds cynical and the real version is quite collaborative. Financial services customer, renewal in the fourth quarter, and the account team wanted two things badly. They wanted AI attach on the contract and they wanted the customer to speak at an event, because a regulated financial institution saying yes to autonomous agents is worth a great deal to a vendor selling into that sector. The customer's instinct was to refuse both, on the grounds that they were being used. We took a different view. Both of those things were nearly free for the customer to supply. They genuinely did want agents, for one specific team, about sixty people in major incident. And their CIO was entirely willing to speak, he enjoys speaking. So we sold both, explicitly, as line items in the negotiation. Here is your attach, here is your reference, and here is what they cost. What came back was a larger assist pool than the standard allocation, a capped overage rate for the full term, rollover on unused capacity, and Prime restricted to the sixty people who needed it. Not one dollar of headline discount changed hands, and the customer was better off by a wide margin over the term. That is the lesson. The valuable currency in an AI negotiation is often not price at all. It is the terms on the meter, and those are exactly what a seller under attach pressure is most willing to give.
No headline discount moved, and the customer came out substantially ahead. Notice how much easier that trade was because the customer genuinely wanted agents for a real team. You are not being asked to fake enthusiasm. You are being asked to notice that the thing you were going to do anyway has a market value this quarter, and to charge for it. Which brings us to how you sell it without buying a trap.
Five rules for selling the attach without buying a consumption trap. Attach small, commit narrow, because a genuine agent deployment on a named population is real AI ACV for them and does not require estate wide Prime, and your job is to stop those two being bundled into one decision. Trade for the meter, not the badge, taking pool size, rate caps, and rollover over headline discount, because the discount is one year and the meter is every year. Price the reference separately, since a quote or a case study or a conference appearance is a distinct asset, and the common error is including it silently in a deal you already agreed, which converts an asset into a courtesy. Time it against their calendar, because the target is annual and public, so the value of your attach genuinely rises as their year closes, and session thirty one covers that calendar properly. And write it down, because whatever you trade for has to survive the renewal, which means it belongs in the order form with the session four clause set and not in the warmth of a relationship with a rep who may be promoted next quarter. Third check.
Knowledge check three, and it is the whole session in one decision. ServiceNow offers you a deep first year discount if you commit the estate to Prime and appear in a customer story. What is the strongest counter? A, accept, the discount plus the goodwill is good value. B, refuse the story and negotiate the discount alone. C, offer the story and a narrow agent attach, and trade for meter terms and caps instead of a one year discount. Or D, accept the Prime commitment but push for more discount. Pause, and ask which of these assets is scarce and which cost is permanent.
The answer is C. Work the two halves separately. The reference is scarce for them and nearly free for you, so give it, deliberately, at a price. Estate wide Prime is permanent and expensive for you and merely convenient for them, so refuse that specific shape while still supplying genuine AI attach on a narrow population, which is the thing they actually need booked. Answers A and D both buy a structural, compounding cost with a temporary discount, and sessions four and seven have already priced that trade out for you, it loses by year three. Answer B is the proud mistake, refusing the story on principle, which throws away an asset that costs you almost nothing and could have bought clause protection you will still want in year four. The general rule, and write this one down. Sell what is cheap to you and dear to them. Keep what is dear to you and merely convenient for them.
Finally, the agent inventory, and I said at the start that agents are the first thing on this platform that spends money without a person present, so they need the same register you now keep for roles, products, and discovery scopes. Every agent named, what it does, which population operates it, what triggers it, what it is allowed to touch, because an unnamed agent is an unowned meter. A ceiling per agent, expected runs and expected burn per period with a threshold that pauses for review, which is session eight's per wave ceiling at agent resolution. A value line per agent, credits per resolved interaction reported monthly, and agents that fail that test get retired rather than expanded, which requires the political willingness to switch off something that was announced as an innovation. And a non production rule, agent testing schedules capped and owned, because the test harness burns the same pool as the live agent and answers to nobody by default. That closes module two. Session ten folds all of it, the tiers, the map, the meter, and the agents, into one entitlement baseline. One more clip before we do.
Guest analyst clip. I will leave you with the thing I think most estates will get wrong over the next two years, because it is already visible in the early ones. Agents are being deployed the way custom applications were deployed fifteen years ago. Enthusiastically, locally, by capable people solving real problems, with no central register and no cost line attached. And we know exactly how that story ends, because we lived it with custom tables and App Engine, where a build decision became a licence decision that nobody discovered until an audit. Agents will be worse, for one specific reason. A custom table sits there costing you an entitlement. An agent runs, on a schedule, consuming a metered resource, every day, whether or not anyone is getting value from it. The failure mode is not a compliance finding at renewal, it is a monthly invoice that grows without a decision. So the register is not bureaucracy, it is the cheapest control available, and the moment to establish it is before you have forty agents, not after. If you take one operational action from this entire module, make it that. A list, an owner per line, and a burn ceiling per line. It fits on one page today. In two years it will not, and by then it will be the most valuable page your platform team maintains.
It fits on one page today. That is the whole argument for doing this now rather than later, and it is the same argument as the user file and the product inventory, established while the list is short. Let's recap and close module two.
The agentic layer, in three sentences. Prime gates autonomy and custom skill building and nothing else, because assisted AI, the Control Tower, and Workflow Data Fabric all ship in every tier, and the gate follows the operating population rather than everyone the agents happen to touch. An agent is priced on four axes while only the licence appears on the quote, so the burn, the governance, and the credits per resolved interaction decide whether automation pays or simply meters work nobody valued. And ServiceNow has published AI revenue commitments with dates attached, which makes your attach, your logo, and your reference genuine currency, best spent on pool size, rate caps, and clause protection rather than on a single year of discount. Next session closes module two by folding everything, the tiers, the population map, the meter, and the agent inventory, into one entitlement baseline. The document you carry into a renewal.
Homework, about an hour. One, inventory the agents, every agent and agentic workflow running or planned, with operating population, trigger, and scope, and treat every blank as the finding it is. Two, size the Prime population, how many people genuinely operate agents against how many the current or proposed quote puts on Prime, and write both numbers down side by side, because that comparison is your session seven map applied to the most expensive tier. Three, test one agent, credits per resolved interaction for your busiest agent or workflow, against your own honest estimate of what that resolution is worth. Four, list your assets, what you can offer that counts against their AI target, attach, logo, reference, multi year, and price each one before anybody asks you for it, because pricing under request is always worse. And five, read the calendar, note their year end, the thirty first of December, against your renewal date, and mark where those two dates create pressure on them rather than on you.
Further reading, five guides. The AI ACV quota leverage trade is today's trade in full, the published targets, the four assets you own, and what each is worth in discount points, and if you read one thing from this session read that. The Now Assist strategy guide covers the agentic roadmap from the buyer's side and where the value genuinely sits below the Prime gate. The Now Assist pillar has the role stack and adoption shape behind agent economics. The AI pricing guide covers how the lines are priced underneath the tier bundling. And the insurance case study takes agents and assists from pilot to production with the governance that kept the meter honest. That is session nine, and module two is nearly done. Build the agent inventory, price your assets, and I will see you in session ten for the baseline.