OpenAI and ServiceNow: Lessons for Integrating AI Into Real Workflows

What changed

On 28 October 2026, OpenAI announced an expanded partnership with ServiceNow. OpenAI says ServiceNow will bring more OpenAI models and agents into its products and connect ServiceNow workflows with ChatGPT and Codex. The announcement also says ServiceNow will expand ChatGPT Work and Codex to employees.

OpenAI describes an integration that brings ServiceNow business context and workflow actions into ChatGPT. Its example involves an IT employee finding similar incidents, assessing change risk and updating a ServiceNow record. Customers connect their own instance and choose what to expose, while OpenAI says ServiceNow Action Fabric applies existing business rules and permissions.

Some experiences remain previews

OpenAI says ServiceNow is previewing interactive ChatGPT experiences, including an on-call workflow involving alerts, recent changes and affected services. A preview should not be described as generally available or proven across organisations.

The announcement does not establish pricing, deployment timing for every customer or realised incident-resolution savings. MindFish has not implemented or tested this integration. The lessons below are our interpretation for smaller organisations.

Lesson one: actions belong in the business process

Many AI experiments stop at answers. A more useful system connects the answer to the work: creating a task, updating a status, requesting approval or recording an outcome. However, the more an AI interface can change, the more important it is that the existing system enforces who may do what.

For an SMB, the system might be a CRM, job management platform or accounting workflow rather than enterprise service management. The principle is similar: avoid giving an AI tool a separate, loosely controlled route around normal business rules.

Lesson two: separate finding, recommending and doing

Retrieving similar records is useful context. Assessing risk is a judgement that requires evidence. Updating a record changes information. Executing a remediation can have operational consequences. These stages should not be bundled into one generic permission.

A practical design might allow an AI assistant to retrieve and summarise, propose a next step and then require a person to approve record changes. Over time, low-risk actions might be automated after testing. The decision depends on consequences, reversibility and evidence.

Lesson three: integration needs ownership

Connected workflows depend on accurate records, maintained permissions and clear procedures. If a field changes or a rule becomes outdated, the AI layer can repeat that problem quickly. Someone should own the process, approve changes and monitor failures.

Auditability matters too. Teams should be able to see what information was used, what was proposed and who approved the action. Logs should support investigation without retaining unnecessary sensitive data.

For example, a small service business could retrieve an existing job record, draft a status update and ask its owner to approve the change. If the customer match is uncertain or the job has already closed, the workflow should pause. A repeated request should not create a duplicate task. This hypothetical example shows how the enterprise lesson translates into ordinary business work without assuming ServiceNow is the appropriate product for every SME.

Before adding another action, identify its consequences and how it can be reversed. A mistake in an internal draft is not the same as an incorrect customer message or an operational change. Approval requirements should increase with that difference.

What this means for Australian businesses

Australian SMBs can apply the lesson without enterprise platforms. Start by identifying the system that should remain authoritative for customers, jobs, stock or finance.

Use existing permissions where possible and add explicit approval for actions with customer, financial or operational consequences. Check provider terms and privacy implications before connecting systems.

Practical next steps

1. Name the system of record for the workflow.
2. Classify actions as retrieve, recommend, update or execute.
3. Map each action to existing permissions and approval needs.
4. Test incorrect records, rule changes and connection failures.
5. Keep an audit trail and assign maintenance responsibility.

Explore CRM and business-system integration