ServiceNow Now Assist Pricing: AI Control Tower Pricing, AI Agent Costs and the Bill Nobody Models

Short answer: ServiceNow does not publish a price for Now Assist, AI agents or AI Control Tower. Every figure is quoted per contract, and the bill has two metered parts, a license tier uplift and a pool of assists that each AI action draws down. AI Control Tower itself is packaged into ServiceNow's AI offering rather than sold on a public card, and several analysts report it is included in the current tiers. The number most business cases leave out is the third cost: what your agents pay to vendors, which appears on no ServiceNow invoice and has no ceiling in the ServiceNow agent model.

If you are building the business case for ServiceNow AI agents this quarter, that last sentence is the one to take into the meeting. The first two costs are negotiable and visible. The third is neither, and it is the only one that can grow without anybody signing a quote.

What ServiceNow actually publishes about AI pricing

Less than you would hope, and it is worth being exact about it, because a lot of what circulates as "ServiceNow AI pricing" is one partner blog quoting another.

From ServiceNow directly: on 29 January 2025 ServiceNow announced AI Agent Orchestrator and AI Agent Studio as included at no additional cost for Pro Plus and Enterprise Plus customers. On 6 May 2025 it made AI Control Tower generally available. On 9 April 2026 it announced new packaging spanning AI assistance, agentic automation and fully autonomous operations, with AI Control Tower named as a built-in component. None of those three announcements contains a dollar figure, and ServiceNow's product pages do not carry a price list.

ServiceNow's own documentation is more useful on the meter than on the price. It confirms AI usage is measured in assists, and its Australia release notes state that AI usage measurement is moving from a 365-day look-back model to a 365-day burn-down model, with usage resetting at the contract anniversary date. In plain terms, you buy an annual pool, agents draw it down, and what is left does not roll forward.

ServiceNow AI Control Tower pricing

AI Control Tower does not have a public list price, and the most defensible thing to say is that it is packaged into the AI tiers rather than sold alone. Licensing analysts describe the 2026 tiers as Foundation, Advanced and Prime, replacing Standard, Pro, Pro Plus, Enterprise and Enterprise Plus, and say AI Control Tower appears in each. Fortune reported in May 2026 that ServiceNow was offering it free for a year to qualifying customers. Both are secondary reports, so ask your account team to put the entitlement in writing on the order form.

The practical point for a buyer is that AI Control Tower is rarely the expensive line. The console that inventories and governs your agents is cheap relative to what those agents consume and, if they can transact, what they spend. We measured what AI Control Tower can and cannot govern in our ServiceNow AI Control Tower breakdown, including the 84 agent columns in ServiceNow's own SDK that contain no money field at all.

ServiceNow Now Assist pricing: the tier uplift and the assists pool

Now Assist is priced as two things stacked on each other. The first is a higher license tier across the users who get AI, which analysts at Redress Compliance put at a 30 to 60 percent uplift on the underlying license. The second is the assists pool, which every AI action draws down regardless of which user triggered it.

The uplift is the predictable part: it scales with seats, and you can negotiate it like any other license. The assists pool is the part that surprises people, and CIO reported in May 2026 that customers find assist usage difficult to predict. The reason is structural. Assists scale with what agents do, not with how many people you licensed, and an agent that runs on a trigger does not need a person to be at work for it to consume.

The trigger problem, in ServiceNow's own words

ServiceNow's documentation describes the failure mode directly. A misconfigured trigger "can process the same incident five or more times per day, across dozens of records, for several consecutive days," and each activation "spawns a conversation and consumes assists, resulting in charges for activity that provides no customer value." That is a vendor describing its own meter running on waste, which is unusually candid and worth quoting in your business case.

ServiceNow built a kill switch for it, and its defaults matter for your budget. It flags a trigger that fires more than 5 times per record in 24 hours, across 25 distinct records, for 3 consecutive days. It ships in warn_only mode, where it emails a warning each day and never deactivates anything. In enforce mode it warns on days one and two and deactivates the trigger on day three. So even with enforcement on, a runaway trigger burns assists for at least two full days before it stops, and with the shipped default it never stops. Set it to enforce on day one of your rollout.

ServiceNow AI agent pricing: the cost that is not on the invoice

An AI agent that summarizes incidents costs you assists. An AI agent that orders things costs you assists plus whatever it orders. ServiceNow ships the second kind: its Procurement product recommendation agent can browse the catalog and create orders, and its ITAM Purchase order creation agent creates purchase orders. Agents you build in AI Agent Studio can be given a catalog tool, and agents on other platforms that AI Control Tower discovers can hold a corporate card or call a paid API.

That spend lands on your card statement or in accounts payable, not on the ServiceNow invoice, so it never appears in the Measure dashboards that track AI cost. And there is nothing in the shipped agent model to cap it. ServiceNow's one real dollar limit, the Approval limit on sn_shop_approval_rule, belongs to a human approver's job code on purchase requisitions. Spend that does not flow through a requisition never meets it.

This is why the business case needs a third column. A realistic model has the tier uplift (fixed, negotiated), the assists pool (variable, capped by the pool size) and agent payouts (variable, uncapped unless you add a control). The first two have a ceiling set by contract. The third has whatever ceiling you put on it yourself.

How to model ServiceNow AI costs for a business case

Use your own quote for the first two lines, because nobody outside your contract can give you real numbers, and build the third from your agents' actual permissions.

1. List every agent and its tools. AI Control Tower's inventory is the right starting point. Mark any agent with a catalog, purchasing, card, invoice or paid-API tool.
2. Price the tier uplift from your quote across the users who need AI, not the whole user base. This is where most negotiation room sits.
3. Size the assists pool from a pilot, not a brochure. Count trigger-driven agents separately, because they consume without human activity and are the source of most overruns.
4. Set the kill switch to enforce before go-live and put the assists threshold alerts on a named owner.
5. Put a number on agent payouts for every agent that can transact: a monthly budget per agent, the counterparties it may pay and the amount above which a person approves. If you cannot write those three numbers down, you do not yet have a cost model, you have a hope.

A caution on scope before you over-buy. If the main thing you want from ServiceNow AI agents is getting incoming requests to the right team, price that single use case against dedicated ticket routing software before you commit a whole tier uplift to it. ServiceNow earns its price when agents work across ITSM, HR, procurement and CMDB together, and it is expensive for one narrow job.

Putting a ceiling on the third cost

The first two costs have a ceiling written into the contract. The third needs one written into policy. That means a per-agent spend limit that aggregates across a day, a week and a month, a merchant-locked virtual card so an agent authorized for one vendor cannot pay another, an approval threshold that holds anything above your number for a named approver, and an audit trail that records which agent paid whom so finance can reconcile it. It sits underneath AI Control Tower rather than competing with it: ServiceNow governs whether the agent may exist, and the spend layer governs what it may pay.

If you are comparing platforms, the same three-cost structure applies elsewhere, and the pattern holds. Gemini Enterprise pricing has published seat prices and a consumption meter with a real cap on the Google bill, and still no per-agent budget. Agentforce bills in Flex Credits and alerts rather than refuses. Every vendor prices the platform and the meter. None of them prices, or caps, the money the agent spends.