Role mining is the process of analyzing the access people already hold to find shared patterns and turn them into roles. An identity governance (IGA) platform groups identities by attributes such as department, job title, or manager, then finds the entitlements most members of each group hold, so access can be granted by role instead of one request at a time.
See how C1.ai builds roles from real access →
What is role mining?#
Role-based access control (RBAC) only works if the roles exist. Building them by hand means sorting through spreadsheets of entitlements, exception requests, and tribal knowledge about who needs what. Most role projects stall there, so access stays request by request and new hires get set up by copying a peer.
Role mining starts from evidence instead. It looks at the access that is actually granted today and asks which bundles of entitlements show up again and again inside the same group of people. Those bundles become candidate roles that an owner can review, adjust, and publish.
How role mining works#
- Collect. Pull identities and their attributes from the HR system or directory, and pull accounts and entitlements from every connected application.
- Group. Split identities into cohorts that share an attribute, such as everyone in Finance, everyone titled "Account Executive," or everyone who reports to the same manager.
- Measure. For each cohort, calculate how many members hold each entitlement. The share of members who hold it is the entitlement's confidence score.
- Review. An owner checks the candidate role, removes entitlements that look like exceptions, and confirms the role matches the job.
- Publish. The approved role grants access automatically by attribute, and the extras stay requestable.
- Repeat. Teams, titles, and tools change. Re-running the analysis keeps roles current.
Top-down vs bottom-up role mining#
| Approach | Starts from | Strength | Watch out for |
|---|---|---|---|
| Top-down | The org chart and job definitions. Business owners describe what each job should need. | Roles match how the business thinks about work. | Slow, and roles reflect what people say they need, not what they use. |
| Bottom-up | Existing access data. Analysis finds the entitlements people actually share. | Fast, and grounded in real grants. | Can bake in access that should never have been granted. |
| Hybrid | Attribute-based cohorts (department, title, manager) mined from real access, then reviewed by an owner. | Evidence from the data, judgment from the business. | Needs a named owner for each role. |
Most teams land on the hybrid approach. The data proposes, a person decides.
Example: role mining a support team#
Five people share the title Support Specialist. Here is what they hold today:
| Entitlement | Holders | Confidence |
|---|---|---|
| Zendesk: Agent | 5 of 5 | 100% |
| Slack: #support channel | 5 of 5 | 100% |
| Salesforce: Service Cloud read | 4 of 5 | 80% |
| Jira: Reporter | 2 of 5 | 40% |
| Zendesk: Admin | 1 of 5 | 20% |
With an inclusion threshold of 80%, the Support Specialist role gets Zendesk Agent, the #support channel, and Salesforce read access. Jira Reporter stays requestable for the two people who need it. Zendesk Admin is a red flag: one specialist holds admin rights the rest of the team does not, which is a sign of privilege creep. That grant goes to the app owner for review instead of into the role.
The next Support Specialist starts with the three role entitlements on day one and requests only the extras.
Role explosion and how to avoid it#
Role explosion happens when an organization ends up with more roles than anyone can understand or maintain, sometimes more roles than people. It usually comes from creating a new role for every exception, mining at a grain that is too fine, or never retiring roles nobody uses. Reviewers cannot certify roles they do not understand, so they rubber-stamp, and the roles stop meaning anything.
Ways to keep the role count under control:
- Set a threshold. Only include entitlements most of the cohort holds. Everything else stays requestable.
- Keep exceptions out of roles. One person's special access is a request, not a new role.
- Mix birthright and requestable access. A role can grant the core automatically and list the extras for self-service requests. See birthright access.
- Give every role an owner who reviews its contents on a schedule.
- Merge near-duplicates when two roles differ by one or two entitlements.
- Retire roles with no members after each analysis run.
Role mining during an IGA migration#
An IGA migration is the best chance most teams get to clean up their role model. Legacy platforms often carry years of IT roles, business roles, and nested bundles that nobody fully understands. Copying that hierarchy line for line into a new platform moves the clutter with it.
A better path: connect applications to the new platform, mine roles from the access people hold now, and compare the results with the old role catalog. Keep the roles the data supports, rebuild the ones that drifted, and retire the rest. For a broader look at the move, see Identity Modernization: Moving from Legacy to Modern IGA and our guide to SailPoint IdentityIQ migration.
Role mining for AI agents and non-human identities#
Non-human identities (NHIs) such as service accounts, API keys, and AI agents now outnumber human users at nearly half of organizations, according to the C1.ai 2026 Future of Identity Report. They hold entitlements too, but they have no job title or department.
Role mining still applies. Group agents by the team that owns them, the function they perform, or the agent type, then mine the scopes and entitlements they share. The result is a baseline for each kind of agent. An agent that holds access outside its baseline is an outlier to review, the same way Zendesk Admin stood out in the support example. Pair each agent role with a human owner and include agents in user access reviews.
Related regulations#
No regulation requires role mining by name. It supports the controls auditors do ask about:
| Regulation or framework | How role mining helps |
|---|---|
| SOX (ITGCs) | Shows that access to financially significant systems is appropriate to each job. |
| HIPAA Security Rule and Privacy Rule | Supports the minimum necessary standard (45 CFR 164.502(b) and 164.514(d)) by limiting default access to what a role needs. |
| ISO/IEC 27001:2022 | Annex A controls 5.15 and 5.18 cover access control and access rights. |
| NIST SP 800-53 | AC-2 (account management) and AC-6 (least privilege). |
Roles built from real access make least privilege the default instead of a cleanup project.
How C1.ai approaches role mining#
C1.ai Role Mining analyzes access patterns across connected applications and groups people by department, job title, and manager's team. Role Mining Suggestions point admins to the groups worth turning into roles, and each suggested entitlement carries a confidence score that reflects the share of group members who hold it. Suggestions refresh after connector syncs, and admins can run a custom analysis on any cohort. Accepting a suggestion creates an access profile, and joiner-mover-leaver policies grant or remove that access as people change roles. Read more in Introducing AI Role Mining.
FAQ#
What is role mining in IGA?#
Role mining is the IGA practice of analyzing existing access to find entitlements that groups of people share, then turning those patterns into roles that grant access automatically.
What is the difference between top-down and bottom-up role mining?#
Top-down role mining starts from job definitions and asks business owners what each job needs. Bottom-up role mining starts from access data and finds what people actually share. Most teams combine the two.
What is role explosion?#
Role explosion is when an organization creates so many roles that nobody can understand, review, or maintain them. It usually comes from turning exceptions into roles.
Is role mining the same as role engineering?#
Role engineering is the broader discipline of designing and maintaining roles. Role mining is the data analysis step inside role engineering that proposes roles from existing access.
How often should you run role mining?#
Run it after major org changes, new application rollouts, and migrations, and on a regular schedule so roles keep up with how teams work.
Does role mining work for AI agents?#
Yes. Group agents by owner, function, or type and mine the access they share. The result is a baseline that makes over-scoped agents easy to spot.
How does C1.ai handle role mining?#
C1.ai suggests groups worth analyzing based on department, job title, and manager's team, scores each entitlement by how many group members hold it, and turns accepted suggestions into access profiles that grant access automatically.
See how C1.ai builds roles from real access →
Sources: C1.ai 2026 Future of Identity Report · C1.ai Role Mining documentation · 45 CFR 164.502(b), 164.514(d) · ISO/IEC 27001:2022 Annex A · NIST SP 800-53 Rev. 5




