ServiceNow + Amazon Bedrock
Run ServiceNow Now Assist on Amazon Bedrock models in your own AWS account. EFS builds the skills, model routing, guardrails and data readiness.
ServiceNow Now Assist can call Amazon Bedrock foundation models — Anthropic Claude, Amazon Nova, Meta Llama, Mistral — directly in your own AWS account. EFS Networks designs, builds and runs that integration end to end.
EFS Networks holds both the AWS Agentic AI and Generative AI Competencies — fewer than 65 partners worldwide hold both — and is a ServiceNow Technology Partner with a Bedrock-powered app, Datum CMDB, live on the ServiceNow Store. We handle the architecture decision, custom Now Assist skills, model routing, guardrails, and the data readiness work that decides whether any of it gives the right answer.
Two ways to run Now Assist on Amazon Bedrock
The choice is made skill by skill, not once for the whole platform. Most of the work is deciding which skills belong on which path.
Path A · ServiceNow-managed
Claude, enabled from the Now Assist Admin console
- No AWS setup. One-click enablement per skill group or skill.
- Billing: included with your Now Assist entitlement.
- Data: governed by ServiceNow platform controls and data-center regions.
- Coverage: most out-of-the-box Now Assist skills (Yokohama Patch 6 or later).
Path B · Customer-managed — your AWS account
Any Bedrock model, through the Amazon Bedrock Spoke
- Data: prompts and responses stay in your AWS account and region.
- Models: Claude, Nova, Llama, Mistral — or a fine-tuned model of your own.
- Billing: on your AWS bill, against your existing AWS commitments.
- Coverage: custom Now Assist Skill Kit skills and select out-of-the-box skills.
What EFS delivers
01
Path & account design
The per-skill path decision, then Amazon Bedrock Spoke and Generative AI Controller configuration, least-privilege IAM scoped to model invocation, region selection, and invocation logging to CloudWatch or S3.
02
Custom Now Assist skills
Skills built in the Now Assist Skill Kit with defined inputs, outputs and prompts — tested against your own records, with acceptance criteria agreed before build.
03
Model routing & cost control
Small, fast models for classification and routing; large models only where reasoning pays. In Datum CMDB, a Bedrock-based ServiceNow app EFS delivers, routing cut inference cost 60–80%.
04
Guardrails & audit
Amazon Bedrock Guardrails for PII and topic filtering, prompt and response logging, and model-change control your security team can review.
05
Agents across both platforms
ServiceNow AI agents and agents built on Amazon Bedrock AgentCore, connected over MCP and A2A and governed from ServiceNow AI Control Tower.
06
Data readiness first
The model reads your CMDB, knowledge base and service map. We check them before go-live — starting with a 30-minute, read-only Datum CMDB diagnostic.
How the integration works
- Workflow context. An incident, change, case, HR or security record — plus the CIs, knowledge and history attached to it.
- Now Assist skill. Out-of-the-box or custom-built in the Skill Kit. It combines the record data with a governed prompt.
- Generative AI Controller. Sends the request through the Amazon Bedrock Spoke, using an IAM role scoped to model invocation.
- Amazon Bedrock, in your account. The chosen foundation model runs the task in your AWS region.
- Guardrails and logging. Bedrock Guardrails filter PII and off-limits topics; invocation logs land in CloudWatch or S3.
- Result in the workflow. A summary, classification, draft or recommendation, shown to the agent or used by the flow.
Where your data goes
On the customer-managed path, prompts and responses stay in your AWS account. Per AWS, Amazon Bedrock does not use your prompts or responses to train models and does not share them with model providers.
For agents built on AWS, Amazon Bedrock AgentCore connects to ServiceNow over A2A and MCP, and ServiceNow AI Control Tower governs both — that combination has been available in AWS Marketplace since May 2026.
Where it pays first
| Workflow | What the model does | Depends on |
|---|---|---|
| ITSM incidents | Summarizes the ticket and work notes, drafts resolution notes, suggests assignment and related incidents. | Knowledge, CI data |
| Change management | Writes a plain-language risk summary from the change plan and the affected CIs and upstream services. | CMDB relationships |
| Knowledge | Drafts knowledge articles from resolved incidents, so the next agent does not start from zero. | Resolution quality |
| HR service delivery | Classifies HR cases, reads sentiment, and drafts responses inside your policy wording. | HR knowledge, guardrails |
| Security operations | Summarizes vulnerability and incident records and the assets they touch, for faster triage. | CMDB ownership |
| Custom workflows | Any process with text in and a decision out — extraction, classification, routing, drafting. | A clear acceptance test |
Every row depends on the data the model reads. That is why each engagement starts with a data readiness check.
How an engagement runs
STEP 01
Assess
A shortlist of use cases with a business case each, the path decision per skill, and a data readiness check across CMDB, knowledge and CSDM.
STEP 02
Pilot
One to three skills built in a sub-production instance against your Bedrock account, measured against success criteria set up front.
STEP 03
Scale
Production rollout, guardrails, cost and usage reporting, and model governance handed over to your team with documentation.
Frequently asked questions
Do we need our own AWS account?
Only for the customer-managed path. The ServiceNow-managed Claude path needs no AWS setup and is included with your Now Assist entitlement. The customer-managed path runs in your AWS account, which is what keeps prompts and responses there and puts inference on your AWS bill.
Which models can we use?
On the customer-managed path, any model Amazon Bedrock exposes in your region — including Anthropic Claude, Amazon Nova, Meta Llama and Mistral — or a model you have fine-tuned. The ServiceNow-managed path uses Anthropic Claude.
Do out-of-the-box Now Assist skills work with our Bedrock account?
Select out-of-the-box skills do, and custom skills built in the Now Assist Skill Kit do. Coverage changes by release, so we confirm it against your instance version during the assessment.
Is our data used to train models?
Per AWS, Amazon Bedrock does not use your prompts or responses to train models and does not share them with model providers. On the customer-managed path, prompts and responses stay in your AWS account and region.
Why start with the CMDB?
Incident summaries, change risk and security triage all read CI data. If ownership, relationships or duplicates are wrong, the model gives confident wrong answers. A Datum CMDB diagnostic takes about 30 minutes, is read-only, and shows which modules your current data can support.
Let's talk about what you're building.
Our team brings over two decades of experience to every engagement. Tell us about your project and we'll show you what's possible.
Let's talk about what you're building.
Our team brings over two decades of experience to every engagement. Tell us about your project and we'll show you what's possible.