Now Assist, bundled into the tier and billed past the pool
ServiceNow bundles Now Assist into Foundation, Advanced, and Prime from 9 April 2026, then meters usage against a bundled pool: once the pool is spent, top up packs bill per unit. Bundling did not make the AI free; it moved the cost from a line you could see to a meter you have to model, and the true up is where the difference gets billed.
Prepared by Redress Compliance · August 7, 2026 · ServiceNow advisory. Based on the estates benchmarked through the 2025 to 2026 packaging change.
Executive summary
The pool is a budget, not an entitlement. Every tier ships a finite pool of assist capacity, Foundation the smallest and Prime the largest, covering a defined volume of generative and agentic actions before overage begins.
Across our benchmarks, overage landed at 15 to 30 percent of tier spend, a 22 percent median, once Now Assist and agents were in real use: the tier price was predictable and the meter was not.
Agents burn the pool at a different order of magnitude.
An autonomous agent completing a task consumes several times the capacity of an interactive prompt, running many actions per task, so consumption rises with agent adoption rather than seat count, and seat based forecasting understates the meter every time.
The forecast that works is an actions model: prompts per user, agent tasks times an action multiplier, plus the load nobody counts.
Development instances share the production meter. Usage in sub production and dev instances draws on the same pool as production, which makes heavy agent testing the most common source of unexpected overage: teams exhausted pools without a single production user noticing.
The dev and test line belongs in the consumption model before the pool size is agreed, because it is in the invoice either way.
Four terms cap the variable half of the bill. The pool sized to a modeled year of usage, the per unit overage rate capped for the term, rollover so unused capacity carries forward, and an annual spend ceiling that triggers a review rather than an invoice.
Buyers who pinned the rate and rollover before signing avoided the worst true up shocks; buyers who trusted the bundle met the meter at renewal.
What draws on the pool, and how fast
| Source | Relative burn | Usually modeled? |
|---|---|---|
| Interactive prompt | Low: the small end of the range | Yes, and usually the only thing modeled |
| Agentic workflow task | Medium to high: multiple actions per task | Rarely |
| Autonomous agent run | High: several times an interactive prompt | Rarely |
| Dev and sub production usage | Variable, and unbounded by production adoption | Almost never |
Seat forecasts understate the meter by construction. The pool depletes by actions, and actions scale with automation, not headcount: an agent on a schedule runs whether anyone watches, and a test harness in dev runs harder than any user.
A consumption model that includes agent runs and non production load, compared against the bundled pool, is what separates a forecast from a surprise true up.
The bundling, and what it changed
From 9 April 2026, Now Assist ships inside every tier rather than as a separate add on, with the pool scaling up the Foundation, Advanced, and Prime ladder.
The framing that follows, AI is now included, so stop worrying about the meter, is exactly backwards in the benchmarks: bundling lowered the anxiety and raised the bill, because the visible add on line became an invisible meter that only surfaced at true up.
15 to 30 percent above tier spend once agents ran.
The tier decision itself, which pool you start from and what the migration between tiers costs, sits upstream of the consumption question, and the 2026 pricing pillar carries the full packaging map.
The ServiceNow 2026 pricing playbook
The consumption model built in: the Now Assist meter, the overage math, the tier pool sizes, and the four terms that cap the variable half of the bill.
Get the white paper →The four terms, and the governance that protects them
- The pool: sized to a modeled year of usage, interactive, agentic, and non production included, not to the tier's default allowance.
- The unit rate: capped for the contract term, because the overage rate and the pool size together decide the variable half of the bill.
- Rollover: unused capacity carrying forward rather than forfeiting, the clause that makes conservative sizing safe.
- The ceiling: an annual spend cap that triggers a conversation rather than an invoice, the backstop when the model was wrong.
Governance after signing protects what was negotiated: AI Control Tower monitoring consumption against the pool, internal alerts firing before overage begins rather than after, and the dev and test load reviewed on the same dashboard as production.
A capped rate with no monitoring is a cap discovered at true up.
- Percentile standing for your exact deal size and industry, from real closed transactions
- Scenario simulation before the call: test alternative terms and see the financial impact of each
- A negotiation playbook, talking points, and a two page executive brief on day one
What we saw across the packaging change, 2025 to 2026
Across the ServiceNow estates Morten Andersen and the team benchmarked through 2025 and 2026, consumption overage was the line item that surprised buyers most:
Above tier spend once Now Assist and agents were in real use, against a tier price everyone predicted correctly.
Buyers who pinned the unit rate and rollover up front avoided the true up shocks the defaults delivered.
The pattern is the enterprise AI meter's standard arc, played out on the ServiceNow estate: a consumption construct introduced beside the subscription, forecast on interactive intuition, and repriced by the automation and the test harnesses nobody modeled.
The deeper Now Assist capability and packaging treatment sits in the Now Assist pillar, and the unit level renewal defense, the reclaims and caps that fund the wider ServiceNow negotiation, in the renewal playbook.
Your first five moves
- Confirm the bundled pool size for your tier, because the pool and the rate together are the variable bill.
- Build the actions model: prompts per user, agent tasks with a multiplier, and a realistic dev and test load, compared against the pool.
- Negotiate the four terms before signing: pool size, capped unit rate, rollover, and the annual ceiling that forces a review.
- Set AI Control Tower alerts before overage begins, with dev and sub production on the same dashboard as production.
- Recheck the model as agents ship, because consumption scales with automation, not headcount. The ServiceNow practice runs the meter with you.
Frequently asked questions
Is Now Assist free in ServiceNow's 2026 packaging?
No. From 9 April 2026 it bundles into Foundation, Advanced, and Prime rather than selling separately, but each tier includes only a finite pool of assist capacity, and once the pool is spent, top up packs bill per unit.
Included up to a limit is not unlimited, and the true up is where the difference gets billed.
What counts against the Now Assist pool?
Interactive prompts, agentic workflow tasks, and autonomous agent runs, plus usage in development and sub production instances, which draws on the same pool as production.
The non production line is the most common source of unexpected overage, because test harnesses burn capacity no production dashboard shows.
How much does Now Assist overage cost?
The per unit rate is quote specific, but the benchmarked outcome is consistent: overage landed at 15 to 30 percent of tier spend, a 22 percent median, once Now Assist and agents were in real use.
The rate and the pool size together decide the variable half of the bill, which is why both belong in the negotiation.
Why do autonomous agents burn so much assist capacity?
An agent completing a task runs many actions, consuming several times the capacity of one interactive prompt, and scheduled agents run regardless of who is watching.
Consumption therefore rises with agent adoption rather than seat count, and any forecast built on seats understates the meter by construction.
How do we cap Now Assist consumption costs?
Pin four terms before signing: the pool sized to a modeled year of usage, the per unit overage rate capped for the term, rollover of unused capacity, and an annual spend ceiling that triggers a review rather than an invoice. Then govern with AI Control Tower alerts set to fire before overage begins.
Do development instances really use the production pool?
Yes: sub production and development usage draws on the same bundled meter as production, so heavy agent testing can exhaust the pool without a single production user noticing. Model the dev and test load explicitly before agreeing the pool size, and monitor it on the same dashboard afterward.
