How Now Assist counts assists and where to check them

Usage of ServiceNow Otto (the AI experience that includes Now Assist) is drawn from the pool of assists in your contract. The first thing admins new to ServiceNow’s AI tend to struggle with is what consumes how much, and where to check it.

With the Brazil release approaching, people are also asking whether Brazil changes how usage is counted. The short answer: I found nothing in the Brazil release notes that changes how assists are counted. The most detailed material on counting, where to check, and settings that prevent overuse is a set of community posts written by ServiceNow employees between June and July 2026, so this article is built on those. That material assumes releases up to Australia Patch 3.

What an “assist” counts#

Now Assist meters consumption in a unit called the assist, not in tokens. An assist is generated each time a Now Assist skill or an agentic workflow runs.

• A simple skill such as case summarization consumes 1 assist
• An agentic workflow that uses multiple actions consumes 25 assists or more

For an agentic workflow, consumption depends on how many tools a single run uses. The actions of the orchestrator and communicator agents are not counted.

Tools used in one run	Tier	Assists consumed
0–4	Small	25
5–8	Medium	50
9–20	Large	150

Skills, Virtual Agent topics and conversational catalog items that run as tools inside a workflow do not consume extra assists on their own. What gets billed is the agentic workflow run itself. The material gives an example where cutting a workflow from 9 tools to 8 drops consumption per run from 150 to 50.

Your contracted pool of assists is managed as a 365-day allowance and resets every year on your contract anniversary. Usage in non-production environments and testing in AI Agent Studio draw from the same pool as production. However, ServiceNow’s official documentation says assists used on demo instances are excluded from the consumed total. In other words, if you try AI features on a non-production instance while validating your Brazil upgrade, that usage comes out of your contract pool. Usage on a demo instance does not.

Does Brazil change it: what I could confirm#

I checked three sets of release notes on ServiceNow Docs: “New features and products in Brazil”, “Changes to Brazil features and products” and “Upgrade information for all Brazil features and products”. I searched them for terms such as assist.

I found nothing saying the way assists are counted, the consumption per run, or the size of the contract pool will change. These are the six items related to AI usage or entitlements:

• The Now Assist and Moveworks experiences are renamed ServiceNow Otto (changes). The Vulnerability Response entry in the upgrade information explains that Otto is the brand for the new AI experience and will be reflected in product names, then says “Your product entitlements remain unchanged. Check your entitlements to determine your access to specific features.”
• AI licensing has three tiers, Foundation, Advanced and Prime, and the license determines which app features, generative AI skills, agentic workflows and AI agents you can use (both new features and changes)
• Security Incident Response gains an AI ROI Summary dashboard whose metrics include total assists consumed and assists per resolved incident (new features)
• Older AI models remain usable in already published skills, agents and workflows, but can no longer be used for new development or configuration. Details are in a Now Support Knowledge Base article (KB3066214). It requires a login, so I have not read it for this article (changes)
• In AI Control Tower’s security settings (Settings > Rules and templates > Security), the defaults for Sampling rate and Max skill calls change, and upgrading updates them to the new defaults (changes). The release notes do not mention any connection to assists
• App summary generation moves from a skill-based approach to an AI agent. Installing the App summary plugin brings the App Summary AI agent with it, but it is disabled by default to avoid unexpected charges. An admin must enable it in AI Agent Studio before users can generate summaries (changes)

ServiceNow’s community announcement also says the rename “does not affect your licensing, entitlements, or existing access”, and that no new purchase is required. That announcement is about the naming changes in the Community, so check how your own contract is handled with your ServiceNow representative.

The August and September 2026 releases of AI Control Tower add features to track and allocate AI consumption costs, including ServiceNow AI Assist consumption. Breakdowns by department and user are limited to ServiceNow AI agents and Anthropic AI. The same post says many of the new features are available only in the new interface.

This check covers those three lists only. I have not checked individual product release notes or contractual terms.

Where to check usage#

You can check usage in the following places, each at a different level of detail.

1. Subscription Management (Subscription Management > Subscriptions > Now Assist usage): shows total contracted, consumed and remaining assists across instances, with a per-instance breakdown
2. Account level entitlements (a tab in Subscription Management): shows the action count, assist ratio and total assists per skill and agent name. According to the official documentation, total assists are calculated by multiplying the action count by the assist ratio
3. Now Assist Admin > Analytics: a per-instance dashboard for usage, skill performance, results from skills you built yourself, and more. It is a Store app, and the material says it requires no additional licensing beyond your Now Assist subscription
4. AI Agent Studio > Analytics > Assist Consumption: assist consumption by AI agents
5. AI Control Tower: has a consumption dashboard, and is also where the cost allocation mentioned above happens

The material notes that the breakdown in Account level entitlements can reveal that most of the pool is going not to frequently used lightweight skills but to rarely used complex agents. If you want to know consumption per skill, this is where to check your own numbers.

Raw consumption data is recorded in tables. sys_gen_ai_usage_log holds the assist count for each call, and sys_generative_ai_log holds a reference to the case or task that ran the skill. Joining the two lets you trace which record caused a given consumption. Check the tier of an agent run in sn_aia_tool_execution_tier.

The material also covers operational caveats. If an assist calculation fails, an hourly job recalculates it. Metering is processed in batches outside the instance, so records lag slightly. Cloning the usage tables from production to development skews your analysis, so check that they are excluded in your clone settings.

Settings that prevent overuse#

System properties let you set up alerts, shutoffs and limits. The properties below are said to be available from Zurich Patch 10 and Australia Patch 3 (June 2026). The material I read does not say how they behave in Brazil.

Spike alerts (sn_aia_property table)

Property	Purpose	Default
alert.assist_spike_hours_to_check	Interval of the job that checks for spikes (hours)	3
alert.assist_spike_usage_percentage_threshold	Increase over the previous check that triggers an alert	0.5 (50%)
alert.assist_spike_usage_threshold	Minimum number of assists before an alert fires	5,000

Shutting off runaway triggers (kill_switch.*)

kill_switch.mode can be off, warn_only or enforce. The default is warn_only, which sends a warning every day. With enforce, it sends warnings on days 1 and 2 and disables the trigger on day 3. Four conditions decide when it fires: the maximum number of times a trigger fires on the same record in one window (kill_switch.max_fires_per_window, default 5), the number of records that exceed that maximum (kill_switch.min_distinct_records, default 25), the window length (kill_switch.window_size, default 1440 minutes), and the lookback period (kill_switch.consecutive_windows_duration, default 4320 minutes).

Recursion protection (recursive_check.*)

Executions with the same conditions can create records up to 50 times per 15 minutes (default) and update records up to 5 times per 15 minutes. Combining “on record create” and “on update” triggers can cause loops, so the material recommends reviewing your trigger configuration.

Request limits

sys_one_extend_rate_limit_rules lets you limit the number of requests to generative AI features or LLM providers, per instance or per user. To limit agent runs, set the Capability Definition to AIA ReAct Engine.

The material recommends keeping consumption alerts on in non-production environments as well, because unexpected spikes happen in development and testing.

If you want to set a cap per team or per person and stop usage once it is exceeded, a separate post shows how to build this with out-of-the-box features only. It builds a report on sys_gen_ai_usage_log, sends warnings at 50%, 75% and 90% of the cap, and removes the skill’s role from anyone who exceeds it. When the period rolls over, the role is restored automatically. Stopping individuals requires that the role is granted directly to the users in question. If they inherit it from a shared role or a group, exclude a suspension group through the skill’s applicability condition. The post advises against modifying the roles that out-of-the-box skills reference. It also explains that while Now Assist analytics and Subscription Management can report consumption, there is no feature that acts automatically when a team or person is consuming heavily.

What to check before upgrading#

Here is what to check about assist consumption around your Brazil upgrade.

1. On your validation instance, confirm spike alerts are enabled before trying AI features. Validation usage also comes out of your contract pool
2. The Sampling rate and Max skill calls in AI Control Tower’s security settings are updated to the new defaults on upgrade, so check the values afterward
3. Record your current consumption per skill in Subscription Management. It gives you a baseline to compare against after the upgrade
4. Confirm that the AI models used by any new skills or agents are ones available in Brazil

I could not find contract pricing or the pool size per tier in the official material. Check anything contractual with your ServiceNow representative.

Press coverage also recommends checking your contract. A CIO article (May 12, 2026) reports that contracts include a set number of assists and that customers can buy more if needed. It also says usage and bills are hard to predict because the number of assists a single agentic interaction uses varies.

An analyst at Info-Tech Research Group advises finding the “line of demarcation where the meter starts running” in your contract. The analyst also raises the risk that an agent stuck in a retry loop keeps consuming assists.

A TechTarget article (April 10, 2026) reports that from April 9, products are offered in Foundation, Advanced and Prime tiers. The AI platform layer that used to be purchased separately is included in these tiers.