Here is the definitional point first, because the whole negotiation turns on it. ServiceNow prices its agentic AI in assists, a consumption unit. As of 2026 guidance, a small Now Assist action costs in the region of 25 assists and a large agentic action in the region of 150 assists, with overage charged as top-ups once your committed pool runs out. Confirm those unit rates against your own order form, because they vary by deal size and by SKU mix. The number you sign is not a seat count; it is a forecast of machine activity, and vendors forecast it generously. The wider context for this shift sits in our 2026 guide to agentic AI ITSM pricing and negotiation.
We model the assist pool against your real pilot data and write the caps that hold. Book an AI cost control review.
What exactly is a ServiceNow assist?
An assist is the billing unit for Now Assist and the broader agentic stack ServiceNow is bundling from April 2026, which includes Now Assist, the Moveworks capability it acquired for $2.85B, Workflow Data Fabric and the AI Control Tower. A single user prompt can consume one assist or many, depending on whether the action is a summary, a generated knowledge article, or a multi-step agentic workflow that calls several tools. The buyer-side problem is that one visible click can trigger a chain of invisible actions, and each one draws down the pool. You are not buying answers; you are buying executions.
How fast does assist consumption actually scale?
Work the math, because the vendor will not show it to you in this form. Suppose 200 fulfillers each trigger 10 small Now Assist actions a day. At roughly 25 assists per small action, that is 200 by 10 by 25, which is 50,000 assists a day, or about 1.05 million assists across a 21-day working month. Now add one automated deflection workflow that fires 2,000 large agentic actions a day at roughly 150 assists each: that single workflow adds 300,000 assists a day, dwarfing the human usage. The lesson is blunt. Agentic automation, not headcount, is what blows the pool. A committed pool sized off seat count is sized off the wrong variable.
| Activity | Volume per day | Assist rate (approx) | Assists per day |
|---|---|---|---|
| Human small actions | 2,000 | ~25 | 50,000 |
| Agentic deflection workflow | 2,000 | ~150 | 300,000 |
| Knowledge generation | 300 | ~150 | 45,000 |
| Indicative daily total | 395,000 | ||
Rates above are indicative 2026 figures for illustration; confirm the exact unit costs and any per-SKU variations against your order form before you rely on them.
What should you negotiate before you sign?
The cap is built from four clauses, not one discount. Each closes a gap the vendor leaves open by default.
- Size the pool from a measured pilot. Run Now Assist for 30 to 60 days, log actual assist consumption by action type, and commit to a pool grounded in that data rather than the vendor's adoption model. A pool set from a slide is a pool set too high.
- Negotiate rollover on unused assists. If you commit to a yearly pool, unused assists should carry forward at least one quarter. Without rollover you are paying for a peak you may hit only seasonally, and the vendor keeps the difference.
- Cap the overage rate in writing. The committed-pool price is the price you negotiated; the overage rate is the price you did not. Fix the per-assist overage rate for the term so a busy quarter does not get charged at an undiscounted list rate.
- Add a true-up review, not an auto-escalator. Vendors prefer that overage rolls into a larger committed pool at renewal. Replace that with a joint review point where both sides re-size the pool against actual data, so a one-off spike does not become a permanent floor.
How does this fit the wider AI tier change?
ServiceNow is also consolidating its packaging, folding Standard, Pro, Pro Plus and Enterprise into Foundation, Advanced and Prime, with legacy SKUs going end-of-sale on 1 July 2026 and AI bundled into the new tiers from April 2026. That means the assist pool no longer sits beside your platform license as a clean add-on; it is increasingly inside the tier you select. Read the consumption cap together with the tier choice, covered in our companion piece on negotiating the new Foundation, Advanced and Prime AI tiers. The same discipline applies across vendors, set out in how to cap agentic AI consumption in ITSM contracts.
Our gated ServiceNow Now Assist Cost Control Guide includes the pilot worksheet and the assist-pool sizing model behind this sequence.
What does this protect against at renewal?
Without a written cap, three things happen at the next renewal. Your overage from the prior year becomes the baseline for a larger committed pool. The undiscounted overage rate you tolerated becomes the reference price. And the tier that bundled your AI gets repriced as a whole. The cap, the rollover and the true-up review are what keep all three from compounding. Across ITSM engagements the firm averages a 30% reduction, and on AI lines that reduction comes far more from sizing and caps than from a headline percentage off list.
Frequently asked questions
- What is a ServiceNow assist?
- It is the consumption unit for Now Assist and agentic AI. A small action costs roughly 25 assists and a large agentic action roughly 150, so the bill follows activity, not seat count. Confirm the unit rates on your order form.
- How do I cap Now Assist consumption before signing?
- Size the pool from a measured pilot, negotiate rollover on unused assists, cap the overage rate in writing, and add a true-up review so a spike does not auto-escalate the committed pool.
- Why does consumption matter more than seats?
- Seat pricing is predictable; consumption is not. One agentic workflow can fire thousands of large actions a day and outgrow the pool, and overage top-ups favour the vendor. Put the cap in the contract.
Cap the assists before signature.
We size the pool from your pilot, write the rollover and overage caps, and set the true-up review. Fixed fee or gainshare.
Book an AI cost control review →