A8gent
A8gent
·
Pricing

ServiceNow Now Assist Pricing: What an Assist Is

Now Assist meters in assists. ServiceNow documents the consumption rules precisely, publishes the rate card only as a PDF, and no dollar price per assist anywhere.

ServiceNow Now Assist Pricing: What an Assist Is visual
AI Agent Starter Kit cover

Free download

AI Agent Starter Kit

Workflow picker, ROI worksheet, 40-prompt pack, and the first-agent rollout playbook. The 1-hour version of the entire A8gent system.

Workflow picker
ROI worksheet
Prompt checklist
Team rollout plan
Open it now

No spam. Worksheet + prompt checklist + rollout plan delivered instantly.

Contents
  1. Fast answer
  2. Quick facts
  3. What this covers
  4. Why it matters
  5. Where to go next
  6. Outcomes
  7. Step by step
  8. The assist is a real unit with real rules and no published price
  9. The token rule multiplies, and the failures are free
  10. Negotiating when there is no list price
  11. At a glance
  12. Mistakes to avoid
  13. FAQ
TL;DRAnswer-first

Fast answer

Now Assist consumption is measured in assists. ServiceNow's own FAQ states that different skills and uses consume different numbers of assists, and publishes the per-skill rate card only as a PDF rather than as a page you can cite. What it does document precisely is the consumption behaviour of custom and third-party model calls: a call made through the Now Assist Skill Kit or the Generative AI Controller consumes 1 assist per call, and calls exceeding 1,000 output tokens consume additional assists proportionally, so a 3,000-token output charges 3 assists. It also documents what is free: for out-of-the-box skills using a NowLLM, a request that errors does not charge an assist, and that includes requests stopped by Now Assist Guardian guardrails. What ServiceNow does not publish anywhere is a dollar price per assist. The FAQ directs customers to contact their account representative for licensing. Verified 1 September 2026.

Facts last verified 2026-09

Quick facts

Consumption unit
The assist (checked 2026-09-01)
Per-skill rates
Vary by skill; published only in a ServiceNow rate card PDF
Custom and third-party LLM calls
1 assist per call
Long outputs
Calls over 1,000 output tokens consume additional assists; a 3,000-token output charges 3
Errors
No assist charged for OOTB NowLLM requests that error
Guardrail trips
No assist charged when Now Assist Guardian guardrails are triggered
Third-party provider charges
Billed separately by the provider under its own licensing
Price per assist
Not published; contact your account representative

On this page

What this page covers

A comparison visitor should understand the tradeoff, the best-fit scenario, and the next diagnostic tool to confirm the choice.

  1. 01What ServiceNow publishes
  2. 02How an assist is consumed
  3. 03The token rule that multiplies assists
  4. 04What is explicitly not charged
  5. 05Forecasting without a price
  6. 06What to ask your account rep

Why does this matter now?

ServiceNow has documented the consumption rules of Now Assist in more detail than most vendors document anything, and has published no price for the unit those rules meter. That combination is genuinely useful if you read it in the right order. You can model assist consumption accurately from skill mix and output length before you ever see a quote, which means the negotiation becomes about one number, the rate, rather than about an entire opaque model. Buyers who skip the modelling arrive at the call with nothing to anchor against.

Internal path

Where to go next from this page

These links are part of the A8gent learning and conversion path. Use them to move from concept, to diagnosis, to workflow build, to course.

Start with readiness

What you should be able to do after this

  • Understand what one assist covers
  • See how bring-your-own-LLM calls consume assists
  • Know which failures are not charged
  • Get the rate card and the price from your rep

How do you do it, step by step?

1. Inventory the skills you will actually turn on

Assist consumption varies by skill, so the first input is a list of the Now Assist skills you intend to enable and roughly how often each fires. ServiceNow states plainly that different skills and uses consume different numbers of assists, so a flat per-interaction assumption will not survive contact with your bill.

2. Get the rate card, which exists only as a PDF

The per-skill assist rates live in a ServiceNow PDF rather than on a citable web page, which is why no honest public page can reproduce them. Ask your account representative for the current Assists rate card document and check the version date on it, because the skill catalogue moves faster than the PDF does.

3. Model your custom and third-party model calls separately

Calls through the Now Assist Skill Kit or the Generative AI Controller consume 1 assist per call. That is a clean, countable rule and it makes bring-your-own-model workloads the easiest part of the forecast. Remember that the third-party provider bills you separately under its own licensing on top of the assist.

4. Apply the 1,000 output token rule to your long outputs

Calls exceeding 1,000 output tokens consume additional assists proportionally. ServiceNow's own example is a 3,000-token output charging 3 assists. Any skill that drafts long articles, summaries or code will therefore cost a multiple of what a short classification costs, and output length is a lever you control in the prompt.

5. Subtract the failures you are not charged for

For out-of-the-box skills on a NowLLM, requests that error do not charge an assist, and that includes requests where Now Assist Guardian guardrails trigger. The same no-charge treatment applies to third-party OEM and custom skill calls on the ServiceNow side, though the third-party provider may still bill you. Do not pad your forecast for failures.

The assist is a real unit with real rules and no published price

ServiceNow measures Now Assist consumption in assists, and states directly that different skills and uses consume different numbers of assists. That single sentence rules out the two forecasting shortcuts most teams reach for. You cannot assume one interaction equals one assist, and you cannot assume the skills your service desk uses cost the same as the skills your HR case team uses. The unit is real, but it is not uniform, and the variance lives in a rate card that is distributed as a PDF rather than published as a page anybody can cite.

That PDF distinction matters more than it sounds. A rate card on a public web page can be linked, dated and checked by anyone, including you at renewal. A rate card in a PDF handed over by an account team is a version you hold, dated at the moment you were given it, against a skill catalogue that ServiceNow expands continuously. If you take one operational habit from this page, make it this: when you receive the assists rate card, record its version and date, and ask for a fresh copy at every renewal rather than assuming the copy in your procurement folder still describes the product.

We are aware of the per-skill assist figures that circulate in third-party write-ups, the ones that assign a small number of assists to summarisation and a larger number to document question answering. We have not reproduced them, and the reason is simple: we could not verify them against ServiceNow's own published material, and a wrong per-skill multiplier in a forecast compounds across every invocation for the life of the contract. Where the vendor keeps a number in a PDF, the honest public page says so and points you at the PDF.

What ServiceNow does publish in citable form is the behaviour of custom and third-party model calls, and this is the most useful part of the FAQ for anyone building rather than buying off the shelf. A call made through the Now Assist Skill Kit or the Generative AI Controller consumes 1 assist per call. One call, one assist. That is a rule you can hold in your head and apply to an architecture diagram, which makes bring-your-own-model workloads the most forecastable part of the platform.

The token rule multiplies, and the failures are free

The 1-assist-per-call rule has an important qualifier: calls exceeding 1,000 output tokens consume additional assists. ServiceNow's own worked example is a 3,000-token output charging 3 assists. So the meter is not really per call, it is per call rounded up through 1,000-token output blocks, and the cost of a skill therefore scales with how much text it produces rather than with how hard it thinks. A classification that returns a single label and a long-form article draft are the same architecture and are not the same price.

This makes output length a genuine cost lever, and one that sits in your prompt rather than in your contract. A skill instructed to return a structured summary in under 200 words stays inside one assist. The same skill left to produce whatever length feels natural drifts across the 1,000-token boundary intermittently, which produces the worst kind of bill: variable, unexplained, and only visible in aggregate. Constrain output length in the skill definition and you have removed the largest source of forecast error in the whole model.

The other half of the consumption rules runs in your favour and is frequently missed. For out-of-the-box skills leveraging a NowLLM, ServiceNow states that if the request results in an error it does not charge an assist, and that this includes cases where Now Assist Guardian guardrails are triggered. The same no-charge treatment applies on the ServiceNow side to third-party OEM calls and custom skills, though the third-party provider may still bill you for the underlying call under its own licensing.

That guardrail clause is worth dwelling on because of what it does to behaviour. If blocked requests were billable, there would be a quiet financial pressure to loosen guardrails, which is precisely the wrong pressure to put on a governance feature. Because they are free, you can tune Guardian toward safety without a cost conversation attached. When you build your forecast, do not pad it for error rates and guardrail trips; on the ServiceNow meter they are zero, and padding for them inflates the volume you take into a negotiation.

Negotiating when there is no list price

ServiceNow publishes no dollar price per assist. The FAQ directs customers to contact their account representative to learn about licensing for Now Assist, and as of 1 September 2026 we found no public per-assist rate anywhere in ServiceNow's own material. That is entirely normal for an enterprise platform sold through an account team, and it is not in itself a problem. It becomes a problem only when a buyer treats the absent price as an absent model and turns up to the call with nothing modelled.

The stronger position is to arrive with an assist volume you computed yourself. Take your skill inventory, apply the rate card the account team supplies, add one assist per custom or third-party model call, and multiply anything with long output through the 1,000-token rule. Now you have a number with a method behind it, which changes the shape of the conversation. Instead of asking what Now Assist costs, you are asking what the rate is on a volume you can defend, and asking whether it steps down as that volume grows.

There is a useful comparison available on the Microsoft side of the fence, and it is worth making explicitly. Copilot Studio publishes its credit pricing and its per-action credit rates on a public page, so you can price a conversation before you speak to anyone. ServiceNow does not. That is not automatically a verdict on value, since a negotiated enterprise rate can easily beat a published list rate, but it is a real difference in how much work you have to do before you know what you are buying, and it is a fair thing to raise in a competitive evaluation.

Finally, remember that the assist meter is not your only Now Assist cost. If your skills call third-party models, that provider bills you separately under its own licensing on top of the assists ServiceNow charges, and ServiceNow says so in the same FAQ. A forecast that captures assists and stops has captured one of two invoices. The total cost of ownership question here spans both, plus the platform licensing the skills sit on, and it is the combined figure that should be compared against anything else.

At a glance

ServiceNow Now Assist consumption rules, read from the Now Assist FAQs on 1 September 2026

ItemWhat ServiceNow publishesEffect on your bill
Unit of consumptionNow Assist consumption is measured by assistsAll forecasting is in assists, not conversations
Per-skill ratesDifferent skills and uses consume different numbers of assistsNo flat per-interaction assumption is safe
Rate card locationPublished as a PDF, not a citable web pageRequest it from your rep and record its version date
Skill Kit / Generative AI Controller call1 assist per callCountable directly from your architecture
Output over 1,000 tokensConsumes additional assists; 3,000-token output charges 3 assistsConstrain output length in the skill definition
Request errors (OOTB, NowLLM)No assist chargedDo not pad the forecast for failures
Now Assist Guardian guardrail tripNo assist chargedGuardrails carry no metered cost
Third-party LLM provider feesCharged separately by the providerA second invoice outside the assist meter
Price per assistNot publishedRate must come from your account representative

What mistakes should you avoid?

  • Assuming every Now Assist skill costs the same number of assists
  • Quoting per-skill assist figures from a blog post rather than the ServiceNow rate card PDF
  • Forgetting that outputs over 1,000 tokens consume additional assists
  • Forgetting the third-party LLM provider bills separately on top of the assist
  • Padding the forecast for guardrail trips and errors that ServiceNow does not charge for
  • Going into a licensing call without a modelled assist volume to anchor the rate against

FAQ

How much does a ServiceNow assist cost?
ServiceNow does not publish a dollar price per assist. Its Now Assist FAQ directs customers to contact their account representative to learn about licensing. As of 1 September 2026 we could find no public per-assist rate from ServiceNow, so we do not print one.
What is an assist in Now Assist?
The consumption unit for Now Assist. ServiceNow states that Now Assist consumption is measured by assists and that different skills and uses consume different numbers of assists. The per-skill rate card is published as a PDF rather than as a public web page.
How do bring-your-own-LLM calls consume assists?
One assist per call for custom skills using external models through the Now Assist Skill Kit or the Generative AI Controller. Calls exceeding 1,000 output tokens consume additional assists, so ServiceNow's example of a 3,000-token output charges 3 assists. Your third-party provider also bills you separately under its own licensing.
Am I charged for failed Now Assist requests?
No, on the ServiceNow side. For out-of-the-box skills leveraging a NowLLM, a request that results in an error does not charge an assist, and ServiceNow states this includes cases where Now Assist Guardian guardrails are triggered. Third-party providers may still charge for the underlying call.
How do I forecast Now Assist cost without a published price?
Model the volume, then negotiate the rate. Count expected invocations per skill, apply the rate card your rep supplies, add one assist per custom or third-party model call, and multiply long-output skills by the token rule. That gives an assist total you can take into the licensing conversation as an anchor.

Sources & further reading

Done for you

Want this built for your business?

Tell us the workflow. We scope, build, and ship the agent with guardrails and the numbers to prove it worked. The scoping call is free.

Free scoping call. You own the code.

Request a scoping call

Was this page helpful?