The shift from per-seat to consumption pricing changes who carries the risk. A seat is a known, fixed number you can right-size before signing. An assist pool is a forecast, and a forecast you have not built yourself is a forecast the vendor builds for you, sized to the optimistic adoption curve in the business case rather than to measured use. This article sits under our 2026 guide to agentic AI ITSM pricing and negotiation, and its job is narrow: explain the unit, then give you the clauses that contain it.
What is an assist, and how does it differ from a token?
An assist is ServiceNow's billing unit for Now Assist. It is an abstraction sitting on top of token consumption: the underlying model still processes input and output tokens, but you are charged in assists so the meter reads in actions rather than in raw model usage. As a working guide, a small action draws about 25 assists and a large agentic action about 150, though the exact draw varies by skill and by how much context the action pulls in. Tokens are the engine; assists are the fare. You negotiate the fare.
The distinction matters because an agentic action is not one call. When an AI agent reads a ticket, queries the CMDB, drafts a change, and posts an update, each step consumes, and the assist count reflects the whole chain. The vendor will demonstrate a single tidy summary at 25 assists. Your production traffic will include the long agentic chains, and those are where the pool drains.
How does the assist meter turn into a bill?
You commit to an annual pool of assists. Usage inside the pool is covered by the subscription. Usage above the pool is an overage, charged per assist at a rate that, unless you have fixed it in the contract, defaults to list at the moment you breach. The structure rewards the vendor twice: once when you over-commit to a pool you do not fully use, and again when you under-commit and pay overage at an unnegotiated rate. The table below shows the three positions and who they favour.
| Position | What it looks like | Who it favours |
|---|---|---|
| Over-committed pool | Large prepaid assist block sized to the business case | Vendor: you pay for assists you never draw, and they rarely roll over |
| Under-committed pool | Small block, frequent overage | Vendor: overage bills at list unless you fixed the rate |
| Pool sized to a measured pilot | Block set from real consumption data, with a fixed overage rate and true-down rights | Buyer: you pay close to what you use and cannot be surprised on the rate |
There is no neutral default here. Whatever number you do not set, the vendor sets, and the vendor sets it in their favour. The buyer-side move is to refuse to commit a pool until you have measured one, which is the discipline laid out in how to model Now Assist consumption before you commit.
Book an AI cost control review.
We size the assist pool from your own pilot data, fix the overage rate, and write the cap into the contract. Fixed fee or gainshare.
Book an AI cost control review →What are the steps to control Now Assist consumption?
Run this as a procurement sequence, in order. Skipping a step hands the vendor the gap.
- Run a metered pilot first. Turn on Now Assist for a defined group for a fixed window and capture assists by skill. Do not commit an annual pool until you have a baseline drawn from your own traffic, not the vendor's demo.
- Project from real chains. Weight your forecast toward the long agentic actions that draw about 150 assists, not the 25-assist summaries the sales motion leads with. Model the mix you will actually run.
- Fix the overage rate in the contract. An overage clause that points to list price is a blank cheque. Set the per-assist overage rate as a contracted number, held flat for the term.
- Negotiate a not-to-exceed ceiling. Cap total annual AI spend regardless of consumption, so a runaway agent or a bad rollout cannot uncap your budget.
- Secure true-down and rollover. Win the right to reduce the committed pool at renewal and to roll a portion of unused assists, so an over-sized first year does not lock in forever.
- Require monthly consumption reporting by skill. You cannot govern what you cannot see. Make per-skill assist reporting a contractual deliverable so you spot the draw before the invoice does.
This is the same sequence we apply across consumption-priced AI, whether the meter reads in assists, credits or tokens. The wider control set, including how caps are drafted, is in how to cap agentic AI consumption in ITSM contracts, and the cost of the uplift specifically is broken down in what the Now Assist AI uplift actually costs.
Our gated 2026 ITSM Negotiation Benchmark Report includes the consumption-pricing benchmarks and the assist-pool worksheet behind this method.
What will the vendor argue, and what do you counter with?
ServiceNow will argue that consumption pricing is fairer because you only pay for what you use. That is true only if the pool is sized to use and the overage rate is fixed; without those two terms, you pay for what you committed and then pay again at list for what you used beyond it. The counter is not to reject consumption pricing, it is to insist the pool is set from measured data and the rate is locked. With the April 2026 packaging move and legacy SKUs going end-of-sale on 1 July 2026, the vendor also has a clean reason to reopen your commercial model, and a reopened model is a chance to set these terms rather than inherit them.
Across more than 500 engagements and over $420M in ITSM contract value negotiated, the pattern holds: the buyers who control AI cost are the ones who measured before they committed and fixed the rate before they signed. The ones who absorbed the uplift treated the assist pool as a line item rather than a forecast they owned.
Frequently asked questions
- What is a Now Assist assist?
- The unit ServiceNow meters Now Assist in. A small action draws roughly 25 assists, a large agentic action about 150. You buy a pool, and usage above it bills as overage, so assists, not seats, drive the cost.
- How do I cap Now Assist consumption?
- Size the pool from a measured pilot, set a not-to-exceed annual ceiling, fix the overage rate in the contract, win true-down and rollover at renewal, and require monthly per-skill consumption reporting.
- Are assists bundled into the new ServiceNow tiers?
- From April 2026 AI is bundled into Foundation, Advanced and Prime, with legacy SKUs end-of-sale on 1 July 2026. Bundled still means a metered allowance with billable overage, so confirm the included pool on your order form.
What to do next
Do not commit an assist pool from a slide. Run a short metered pilot, build your own forecast from the long agentic chains, then take the fixed pool, fixed overage rate and not-to-exceed ceiling to the table together. When you want that run as a structured engagement, it goes through our AI cost control service on fixed fee or gainshare.
Stop the meter before it starts.
We measure your Now Assist consumption, size the pool to it, and lock the overage rate into the contract.
Book an AI cost control review →