Agentic AI where your processes and your data already live
If you do not yet know which AI opportunities are worth pursuing, start with AI Use Case Discovery. If the question is who controls the AI you already run, start with AI Control Tower.
Ask in plain language and get the answer with the work already done.
ServiceNow AI gives your business AI skills, AI agents and agentic workflows that run on your live enterprise data, inside the processes you already operate.
Give an agent a job, a boundary and a stop condition, and it completes the work instead of handing you a draft.
The record
updated.
The approval
routed.
The case
closed.
Every action it takes is permission-scoped, logged and governed, so the people who have to approve AI can see exactly what it did.
AI agents, skills and agentic workflows running on your own business data
ServiceNow AI runs natively on the platform that already holds your requests, cases, orders, assets and approvals.
AI skills
Summarize, draft and classify inside the work.
AI agents
Take a job and complete it across systems.
Agentic workflows
Chain several agents toward one business outcome, with a person in the loop wherever you want one.
Ask a question in plain language across chat, voice, web and mobile, and get an answer grounded in your records rather than in general knowledge.
Search knowledge, catalog and case history in one place.
Ask for a number without building a report.
Have an agent triage an incoming request, gather what is missing, act in the connected systems and close the loop.
Let a workflow draft the quote, route the approval and update the order.
ServiceNow Otto
The experience layer where the asking happens.
Workflow Data Fabric, Knowledge Graph and Context Engine
Let the AI reach the records and systems that hold the answer.
AI Control Tower
Where you see and govern all of it. AI Control Tower
Tells you what your leave policy says.
Approves the leave, updates the roster and notifies the manager.
On ServiceNow the AI sits on the same platform as the process, the data and the permissions, so the action is available to it. That is the difference between AI that explains your business and AI that runs part of it.
What changes when AI works inside your platform
Work completes without a person carrying it between systems. You get faster answers for customers and employees, fewer queues waiting on someone to route them, and a record of every AI action in one place. Your people spend their time on the exceptions rather than on the handling.
A customer asks where their order is
Request understood
Systems read
Answer back, work updated
A service event happens
Impact assessed
Affected customers identified
Follow-up work created
A new demand arrives
Assessed for value and feasibility
Prioritized with evidence attached
No waiting for the next meeting
A quote needs pricing and an approval
Offer drafted
Approval routed
Pauses where a person must decide
Your AI estate stops being a rumor.
Every agent, model and connection sits in one inventory with an owner and a risk classification, which is what turns "we have AI somewhere" into something you can report on and extend. Consumption gets an owner too, so the AI usage already inside your license tier goes to work on purpose rather than surfacing at renewal.
Those are the gains worth measuring, and they get measured on your own processes and your own numbers.
Where ServiceNow Otto fits, and what is generally available now
One place to ask, covering conversational AI, enterprise search, voice agents in several languages and plain-language data questions. It brings together what ServiceNow previously offered as Now Assist, Moveworks and AI Experience. Every action it takes is governed and logged through AI Control Tower.
A specific set of AI skills, AI agents and agentic workflows, and that set depends on your release and your entitlements. Otto is the experience layer and the direction of the platform.
ServiceNow introduces Otto.
Otto can be experienced in ServiceNow EmployeeWorks and AI Control Tower.
Rollout across all products.
We check your instance before we design anything, so your plan is built on what you can switch on now, with the roadmap marked as roadmap.
What your license tier lets you build
ServiceNow repackaged its licensing in April 2026 into three tiers: Foundation, Advanced and Prime. AI is included in all three rather than bought separately, so most customers already own more AI capability than they use. The detail worth knowing is which tier lets you build your own.
| From ServiceNow’s documentation | Foundation | Advanced | Prime |
|---|---|---|---|
| Use and configure the supplied AI skills and agents | ✓ | ✓ | ✓ |
| Create net-new custom AI skills and agents | , | , | Prime only |
| Autonomous AI workforce | , | , | Prime only |
| AI Control Tower | ✓ | ✓ | ✓ |
| Manage AI assets outside ServiceNow | , | , | ✓ |
| Workflow Data Fabric | ✓ | ✓ | ✓ |
Workflow Data Fabric is included; its Zero Copy Connectors sit in a paid upgrade.
Assists, from one pool for the whole instance
Usage is metered in units ServiceNow calls assists. The pool belongs to the whole instance rather than to individual users, different actions consume different amounts, and testing outside production draws on the same pool.
Decide what kind of AI you intend to build before you choose a tier.
The tier decides whether you may build at all. Then give the usage pool a named owner and a monthly review before your first agent goes live. ServiceNow publishes no assist values, no pool sizes and no prices, so we work the mechanism through against your own estate rather than handing you an estimate.
Bring us one process you want AI to finish.
We will show you what it takes to make that happen on your instance, before anything reaches a design.
You are not locked to one model
ServiceNow supports its own model service alongside Azure OpenAI, Google Gemini and Anthropic Claude on AWS. You can choose the model for the whole instance, for a group of skills or for a single skill, and the policy that decides which model runs where is approved and audited in AI Control Tower.
| Supported model | Bring your own key | The whole instance | A group of skills | A single skill |
|---|---|---|---|---|
| ServiceNow model service | , | ✓ | ✓ | ✓ |
| Azure OpenAI | ✓ | ✓ | ✓ | ✓ |
| Google Gemini | ✓ | ✓ | ✓ | ✓ |
| Anthropic Claude Our preferred | ✓ | ✓ | ✓ | ✓ |
Claude is our preferred intelligence layer.
The architecture does not force every use case onto one model. Anthropic and ServiceNow have a documented partnership: Claude is the default model behind ServiceNow Build Agent and a preferred model across the platform, and Anthropic is the first design partner for ServiceNow Action Fabric.
Bringing your own keys is supported for Azure OpenAI, Claude on AWS and Gemini, so a regulated business can run models through its own cloud account.
Some supplied skills, agents and agentic workflows do not offer model choice.
A provider ServiceNow does not certify is a custom integration with its own translation layer rather than a setting.
We establish both against your instance before anything is designed.
The AI you already chose can do governed work in ServiceNow
ServiceNow Action Fabric opens the platform to AI agents built elsewhere. A workflow that runs when a person clicks through the ServiceNow interface can also run when an approved external agent asks for it, with no interface involved. Every action is identity-checked, permission-scoped and auditable through AI Control Tower.
The AI assistant your people already have open.
Most organizations have already bought AI, so this is the difference between adding another assistant and connecting the one you have to work that completes. Those actions draw on the same metered pool as the rest of your AI activity, which is one more reason to give the pool an owner.
Connect the data and the systems that live outside the platform → ServiceNow Integrations
How AI gets approved in your business, not just built
Autonomy is earned with evidence. Before an agent acts, you decide what it may see, which actions it may take, where a person must approve, and what happens when it fails. AI Control Tower logs interactions, enforces policy and can stop an agent. Usage thresholds cap what it can spend.
ServiceNow frames its AI as sense, decide, act, secure. We work it in the other direction, because that is the order that gets a program approved.
What the AI may see, may do and must ask about, with the approval thresholds, the containment behavior and the audit trail agreed before anything is switched on.
The AI can reach trusted data, and the data is good enough to act on.
It understands the business context well enough to make the call you would make.
It performs the work, inside the boundaries you set.
We design the happy path and the stop condition together.
An agent that goes wrong is contained and can be stopped. An agent that spends is capped.
Govern every AI asset in one place → AI Control Tower
Establish your regulatory obligations → EU AI Act Readiness
AI agents are only as good as the process underneath them
AI delivers when the process, the data, the permissions and the ownership underneath it are right, and putting those right is most of the value of the project.
A process the teams who run it have agreed.
Catalog, knowledge and configuration data the AI can trust.
Integrations to the systems that hold the rest of the answer.
Roles and permissions that match how your people work.
A license tier and a usage pool checked against what you intend to build.
A named owner for every production agent, because that is what keeps it working after go-live. Solution Support can carry that work with you.
We do not start with AI ideas.
A workshop where everyone lists what AI could do produces a wish list. We combine your business priorities with your real ServiceNow operational data to find where repetitive, high-value work actually happens, then assess each candidate for value, feasibility and risk. The platform’s own agent advisor mines your historical records for that evidence, so the shortlist comes from your process data rather than from the room.
We test before we trust.
Every agent intended for production gets these, designed with the workflow rather than added after go-live:
An agent working in a demo is not evidence that it is ready for production. One governed process runs end to end before the second one starts, so the pattern is proven rather than assumed.
Share of requests completed without a human handoff
Share of agent actions needing correction
Time to resolve for the targeted process
Consumption per completed outcome
Production agents with a named owner
Share of AI assets carrying a risk classification
Decide which AI opportunities are worth pursuing first → AI Use Case Discovery
We rebuilt our own company around this before selling it
We run on it.
ServiceNow and Claude are how we operate, document, manage knowledge and deliver work. We redesign our own processes around AI first, then productize what works. That is a different claim from "our consultants use AI", and it is why our advice comes from operating this rather than from reading about it.
We build the processes the AI has to work inside.
We implement the commercial, customer service, field service and project operations side of the same platform: Sales CRM, CPQ, CSM, FSM and Project Operations. An agent is only as good as the process and the data underneath it, and that is the part we do deepest.
We build and run agents.
In our own operations and in the environments we build, under explicit gates: scope approval, verification before anything is created, and a build log. An agent holds credentials and takes actions, so it gets an owner, a permission set and a review date like any other production component. AI-assisted knowledge work runs inside live client engagements.
Two platforms, on purpose.
ServiceNow provides the orchestration, the workflow and the governance. Claude provides the intelligence. We are a ServiceNow and an Anthropic Claude consulting and implementation partner, so one team designs across both instead of two suppliers meeting at your integration layer.
Independence is in the contract, not the pitch.
We are required to document the recommended system of record and the integration boundary, including where another supplier’s product is the better answer.
Senior practitioners, from the first workshop to go-live.
Founded in 2025 and built around people with 20+ and 18+ years in platform and business operations. You work with them directly: no account layer, no bench, no handover to a team you have not met.
Find out what your AI is ready to finish
An agentic AI readiness review looks at three things: one business process you want AI to complete, the state of the data and the platform underneath it, and what your license tier permits. You get a straight answer on what is possible now, what needs fixing first, and what is still roadmap.
Agentic AI
Software that decides what needs to happen next and then does it, within rules you set.
AI agent
One piece of AI given a job, a set of tools and a boundary.
Agentic workflow
A business objective handled by one or more AI agents working in sequence.
ServiceNow Otto
The AI experience layer: one place to ask, across chat, voice, web and mobile.
Action Fabric
The mechanism that lets an approved external AI agent trigger a ServiceNow workflow.
AI Control Tower
The governance layer: what AI is running, who owns it, what risk it carries and what it is allowed to do.
Assist
The unit ServiceNow uses to meter AI usage.
The questions buyers ask us.
What is the difference between generative AI and agentic AI?
›
Generative AI produces content: a summary, a draft reply, a suggested article. Agentic AI decides what should happen next and then performs it, across systems, within limits you define. The practical test is whether the work is finished at the end. A generative feature hands you something to use. An agentic workflow updates the record, triggers the approval and closes the task, and pauses where a person must decide.
Is ServiceNow Otto available now?
›
Partly, and the detail matters when you are planning. ServiceNow introduced Otto in May 2026 and states that it can first be experienced in ServiceNow EmployeeWorks and AI Control Tower, with rollout across all products in the year ahead. So Otto is the direction and the experience layer. What is generally available on a given instance today is a specific set of AI skills, agents and agentic workflows, which depends on the release and the entitlements in place.
Do we get all the AI capability with our license?
›
Most of it, and the build rights are the exception. Since April 2026 the three tiers (Foundation, Advanced, Prime) each include AI, so you already own capability you can put to work. Using and configuring the supplied AI skills and agents works at every tier. Creating custom AI skills and agents, and the autonomous AI workforce, require Prime. Usage is metered against a pool that belongs to the whole instance. How licensing fits your wider plan is covered in Roadmapping.
What does it cost?
›
We do not publish pricing. AI sits inside the license tier rather than beside it, usage is metered in assists, and ServiceNow publishes no assist values or pool sizes, so the answer depends on your tier, your estate and what you intend to build. We work it through with you rather than estimating.
Which AI models can we use?
›
ServiceNow supports its own model service alongside Azure OpenAI, Google Gemini and Anthropic Claude on AWS. Models can be selected at instance, skill group or individual skill level, governed in AI Control Tower. Bring-your-own-key is supported for Azure OpenAI, Claude on AWS and Gemini. Two limits: some supplied skills do not support model choice, and using an uncertified provider is a custom integration, not a setting.
Can we use the AI assistant we already have?
›
Yes, through ServiceNow Action Fabric. It exposes ServiceNow flows, playbooks, approvals and catalog items to approved external AI agents. A process that normally runs when a person clicks through the interface can be triggered by your assistant instead. Every action is identity-checked, permission-scoped and auditable. Those actions draw on the same metered usage pool as the rest of your AI activity.
What stops an AI agent doing something wrong?
›
Design, not hope. Before go-live, you define what data the agent may see, which actions it may take, where a human must approve, and what happens when it fails. How every agent is registered, owned and switched off is covered on AI Control Tower.
Where is our data processed, and does it stay in Germany?
›
This depends on the model provider, the region, the specific capability and how the connection is configured, so it needs a definite answer for your setup rather than a general reassurance. It is usually the first question a German security review asks. We establish it per solution, in writing, before anything is designed, and we tell you what leaves the instance and what does not.
Is ServiceNow AI the right place for us to start?
›
It is the strongest place to start when ServiceNow already runs a business process end to end, the work crosses systems and departments, and governance and auditability are hard requirements rather than preferences. If AI is already inside your license tier with no plan attached, you are paying for it either way. Where the process is undefined or two teams disagree about it, and where the catalog, knowledge articles and configuration data are poor, we fix that first, because stale knowledge and wrong configuration data produce confident wrong answers. We would rather tell you that in the first conversation than the third. And where the need is document and knowledge work with nothing to execute afterwards, the better answer is a Company Second Brain on Claude, which we also build. If nobody in your team has taken an agent into production before, a Forward Deployed Engineer can do it alongside them.
Do we need to fix our data before starting?
›
Often some of it, and it is better to know early. How we check whether a process and its data are ready is part of AI Use Case Discovery.