A case manager looks after people who need help, like someone claiming support or recovering from an illness. They decide what help to offer and when to check in again.
Case managers work with individuals over time, in social care, health, housing, insurance or employment support. Their decisions shape outcomes for a real person, so the stakes are high even when the volume is modest. AI can save time on notes and preparation. The judgements about the person should stay with the case manager.
No single O*NET code fits; the role spans social work, healthcare and claims settings. Decisions (eligibility, plan, contact cadence, escalation, closure) are medium frequency, high stakes and often partly irreversible. Use AI for summarisation, drafting and evidence gathering with permission-trimmed grounding; keep eligibility and risk judgements human with recorded rationale.
The role
"Case manager" means different things in different sectors. In social care it may be a social worker or support coordinator; in health, a nurse coordinating care; in insurance, someone managing a claim; in employment support, an adviser. There isn't one ONET occupation that covers them all, so this card describes the common shape of the work. A real card for a specific team should start from the closest ONET or ESCO occupation and be checked against what the team actually does.
The common shape is this: take on a person's case, assess their situation, agree a plan, coordinate others, keep records, review progress, and close the case when it's done.
The decisions
| Decision | Options | Frequency | Stakes | Reversible? |
|---|---|---|---|---|
| Whether a referral meets the criteria | Accept, decline, or ask for more | Medium | High | Partly; a wrong decline can be appealed, but harm may occur meanwhile |
| What goes in the plan | A set of services or actions | Medium | High | Yes, at the next review |
| How often to make contact | Weekly, monthly, or as needed | Medium | Medium | Yes |
| Whether to escalate a concern | Escalate now, monitor, or no action | Low to medium | Very high | Often no |
| When to close the case | Close, extend, or transfer | Low | Medium to high | Partly |
Frequency here is per case manager per month. Stakes are high across the board because each decision affects a specific person, often someone in a vulnerable position.
Where AI helps
Most case managers spend a lot of time on records: reading back through notes before a visit, writing up after one, and preparing for reviews. That's where AI tools fit best.
Microsoft 365 Copilot could summarise a case history from documents and emails the case manager already has access to, or draft a visit note from bullet points. A Copilot Notebook could hold the key documents for one case, so questions are answered only from those sources, as described in Notebooks: one place for the sources behind a decision. An agent grounded in the eligibility policy could help check a referral against the written criteria and show which criteria are met.
Where it shouldn't decide
Eligibility, escalation and closure are judgements about a person. They depend on things that may not be written down: how someone seemed on a visit, what they didn't say, what a family member mentioned. An AI tool can't see those, and its summary of the written record could make the record seem more complete than it is.
For these decisions I'd keep AI to gathering and checking: listing the criteria, pulling relevant notes, flagging missing information. The case manager decides and records why. For escalation, I'd add a rule that an AI summary must never be the only thing read before deciding not to escalate.
A worked example
These numbers are illustrative. A case manager holds 30 cases and spends about 45 minutes before each review reading notes. With a Copilot summary to start from, that drops to around 15 minutes, with the time spent checking the summary against the notes that matter most.
| Before | After | |
|---|---|---|
| Reviews a month | 20 | 20 |
| Preparation per review | 45 minutes | 15 minutes |
| Preparation time a month | 15 hours | 5 hours |
The ten hours saved are only worth having if the summary is reliable. The risk is a missing fact that changes a judgement. So I'd ask case managers to note, for a month, every time a summary missed something important. That count tells you whether the time saving is safe.
What I'm still checking
Case records often contain sensitive personal data about health, family or finances. Before using any AI tool on them, I'd want to confirm the oversharing position on the sites where records live, as in Purview and Copilot: oversharing checks before rollout, and check the organisation's data protection assessment covers the use.
Sources
This role card describes a generic role and uses no external facts or figures. It refers to Microsoft 365 Copilot and Copilot Notebooks, whose capabilities are sourced in the linked Capabilities articles. All times are illustrative.