Find where AI can make a measurable difference.
Explore 250 healthcare AI use cases, shortlist what fits, and build a business case with transparent assumptions. Start with a few workflows you can measure — not an enterprise-wide promise.
Seven of these run on a Microsoft Copilot seat alone; the three marked Cowork also need Copilot Cowork turned on (billed on usage). No line-of-business data and no IT project — pick one, copy the prompt, and try it today.
Refine results
Consumption None Yes TBD what costs extra to run
None — nothing extra to pay. A licensed employee invokes the agent under their own identity. Microsoft's published rate card lists every billable event — generative answers, agent actions, tenant-graph grounding and the connectors the agent calls — as no charge for employee-facing use, within fair-use limits. Helpers that touch no organizational data are covered free by Copilot Chat, with no seat at all.
Yes — it consumes Copilot Credits. Something puts the work outside the seat: it runs unattended (no signed-in user for Microsoft to attribute the run to), the user has no Copilot seat (frontline F1/F3, so every prompt bills), or it is Copilot Cowork (multi-step orchestration billed on usage).
Hover any row to see its own arithmetic against Microsoft's published rate card, or open a build card for the full explanation.
My Selection
Your selected agents, with modeled value and the costs HAVEN can price. Work through the steps below — every number updates as you go.
Set your organization
Start here — every number in your Impact business case is calculated from these inputs. values you override are outlined in blue
Value assumptions — two dials that scale every estimate below
🔒 Copilot Credits are priced at Microsoft's standard $0.01 per credit — fixed, not editable. See the estimate breakdown below.
Tune each use case
Override users, frequency and economics per agent wherever you know better than the default.
🪪 Licensing model — what's included vs. what consumes Copilot Credits
Licensed users — no added credit cost: when a licensed employee invokes an agent under their own identity, Microsoft’s published rate card lists every billable event — generative answers, agent actions, tenant-graph grounding and the connectors the agent calls — as no charge for employee-facing (B2E) use. Subject to fair-use limits. Two carve-outs worth knowing: agent flows are only included on the When an agent calls the flow trigger, and computer-using agents are excluded.
Consumes Copilot Credits:
• Unlicensed users — frontline staff on F1/F3 with no Copilot seat. Nothing is covered for them, so every prompt against an agent bills to the tenant.
• Unattended agents — they run with no signed-in user, so there is no license for Microsoft to attribute the run to. Billable for everyone. HAVEN prices them two ways. An agent that handles one item per run (a new request, form or email) is priced on the Standard harness at about 17 credits per item (10 tenant graph grounding + 2 generative answer + 5 agent action). An agent that runs on a schedule and builds a file is priced on the GitHub Copilot harness, using Microsoft’s published runtime bands: 100–300 credits for a light run, 300–500 medium and 500+ heavy.
• Copilot Cowork — multi-step orchestration, billed on usage. It needs an active Microsoft Copilot license plus Cowork, usage-based billing and Anthropic models switched on in the tenant. Microsoft publishes planning bands by task weight: light 100–300 credits, medium 300–700, heavy 700+.
• GitHub Copilot harness — agents built on this harness bill in Copilot Credits both while you author them (a one-time 1–60+ credits per agent, immaterial) and every time they run. It is the harness Copilot Studio offers for complex, multi-step work that produces files, so HAVEN assumes it only for the scheduled agents that build files. For licensed users in Microsoft channels, agents on the Standard and Copilot Chat harnesses are included with the Copilot license under fair use.
Every rate here traces to Microsoft’s published Copilot Credit pricing at $0.01/credit — the per-event rate card for metered prompts (classic answer 1 · generative answer 2 · agent action 5 · tenant-graph grounding 10 · agent flow actions 13/100), and the published scenario bands for Cowork tasks and GitHub Copilot harness runs. Each agent shows its own arithmetic.
Buying credits: pay-as-you-go at $0.01/credit through an Azure subscription, billed monthly in arrears — no commitment, no expiry, and it decrements MACC. Credits pool at the tenant level, so a single meter covers every agent here; spend caps and alerts are set in the Microsoft 365 admin center. Discounts and other commercial terms remain deal-specific — confirm them on the customer agreement.
Microsoft Scout is not yet priced — excluded from the cost total.
Verified 6 Aug 2026 against Microsoft Learn (Copilot Studio billing rates, harness guidance, credit purchasing, Cowork prerequisites) and re-checked 11 Sep 2026 against the Microsoft Copilot Credits Guide, September 2026 — which is also where the Microsoft 365 Copilot → Microsoft Copilot rename (August 2026) used throughout this tool comes from.
⚠️ How the Copilot Credit estimates work (rates and assumptions)
⚠️ Every rate traces to Microsoft’s published Copilot Studio rate card, at $0.01 per Copilot Credit. The rates are Microsoft’s; how many of each event an agent fires is our estimate.Agent prompt, unlicensed user — 10 tenant graph grounding + 2 generative answer = ~$0.12, plus $0.05 if the agent also calls a line-of-business system.
Unattended agent, one item per run — about 17 credits per item (~$0.17): 10 tenant graph grounding + 2 generative answer + 5 agent action, on the Standard harness.
Scheduled agent that builds a file — 200–600 credits (~$2–6) per run, depending on what it produces, priced on the GitHub Copilot harness using Microsoft’s published runtime bands (light 100–300 · medium 300–500 · heavy 500+).
Cowork — 375–700 credits (~$3.75–7) for a task’s first request, inside Microsoft’s published Cowork task bands (light 100–300 · medium 300–700 · heavy 700+); each follow-up counts as a lighter refinement (~20% of the first), so a typical task totals about 675–1,100 credits (~$7–11), and the heaviest about 1,700.
Interactive use scales with users × prompts/month; unattended agents are priced per item or per scheduled run, not per user. Agents invoked by a licensed user add no credits; the Copilot seat itself is not costed here.
Set Users or ⏱ frequency per agent to override. Validate with the Copilot Studio agent usage estimator (link in the exported brief).
📊 Assumptions & sources (steady-state)
📊 Labor rates are per department — each carries its own loaded rate (BLS OES healthcare wage families + ~40% benefits/overhead), from ~$21/hr for food service to ~$145/hr for the C-suite. A single blended figure would overstate frontline value and understate executive value. Override the rate on any individual use case in step 2. Usage is declared per use case — who uses it, how often and how many items it handles — from a review of every record, rather than inferred from keywords; adjust any of them per agent in My Selection. Realization 70% and AI-attribution 50% are deliberate conservative haircuts against published Microsoft Copilot productivity findings (Microsoft WorkLab Work Trend Index; Forrester Total Economic Impact). Hard-dollar cash defaults to $0 until you enter your own base. All figures are directional steady-state estimates for a mature deployment — not current adoption.🔄 What changed in this update (the library review)
🔄 Every use case was re-read and scored against one rubric: a named role does it every week; it uses the right Microsoft tool; it works today with standard connectors; it sits in the right department with the right audience; its minutes, runs and credits are realistic; and its value type is honest — only true hard-dollar savings count as cash.Only 15 of the 988 use cases carry a validated cash lever, and that cash counts only once you enter your own base; none of the 15 is in Pharma. Another 65 name the P&L line they move but put no figure on it. The other 908 give time back or reduce risk — real value, but not cash.
147 overlapping use cases were merged, 31 were retired and 62 were added, including a new Physicians & Advanced Practice department. The review left 987 use cases across Provider, Payor and Pharma.
Added since the review: the Tiered digital huddle board and escalation agent in Quality & Patient Safety, from the Microsoft Healthcare AI Business Process Solutions team — its build card shows the pilot measures and a 2-minute demo. The library now holds 988 use cases.
Usage is now declared for each use case — who uses it, how often and how many items it handles — instead of inferred from keywords, so modeled totals are lower than in earlier versions by design. If you built a business case in an earlier version, re-run it here before you use its figures again — the two sets of numbers are not comparable.
Boards and share links from earlier versions still open: merged agents open as the agent that replaced them, and retired agents are named when they are removed.
Earlier ideas
Have an agent in mind that isn't in the library? Add it here — it drops into your selection and the library so you can include it alongside the rest.
Your Impact
Impact, cost and net return — recalculated live from the setup & assumptions you set in My Selection.
Security & IT enablement
Suggested controls for your selection: connector access, data protection, human review and Agent 365 planning. These are deployment-specific checks for IT, not a security approval or a promise of existing license coverage.
Before your first pilot
A connector list is not an approval. Agree these launch gates with your business, IT, privacy and security owners.
← Back to the Use Case Library
🛠 How to build these agents
A plain-English guide for anyone building these — whether that's your own department or IT. Every agent in this library is built with one of four Microsoft tools — here's what each is and when to use it.
Microsoft Copilot
CopilotIn-app help inside Word, Excel, PowerPoint, Outlook, and Teams. No build required — just a well-written, reusable prompt. Best for drafting, summarizing, and analyzing content already in M365.
No added credit cost for a licensed user — it is the seat you already pay for. Frontline staff without a seat get Copilot Chat, which cannot reach your organization's data.
Copilot Studio agent
StudioA custom agent you configure with knowledge and tools. Build these as a Microsoft Copilot agent (the “Copilot chat” harness): policy/Q&A agents connect an authoritative SharePoint library, and a few read a line-of-business system via a connector. Best for repeatable Q&A and workflows.
No added credit cost when a licensed user invokes it themselves, connectors included. Consumes credits when it runs unattended, or when the user has no Copilot seat.
Pick the right build path. The main Agent card on the Copilot Studio home builds on the GitHub Copilot harness, which bills Copilot Credits and is not covered by a Copilot seat. For the agents in this catalogue, choose Microsoft Copilot agent instead.
Copilot Cowork
CoworkMulti-step, orchestrated work that spans people, documents, and apps to produce a deliverable. Best for coordinated processes like a board pack or month-end close.
Consumes credits — Cowork is billed on usage rather than covered by the seat, and needs an active Microsoft Copilot license plus Cowork, usage-based billing and Anthropic models enabled in the tenant.
Microsoft Scout
ScoutAn autonomous assistant that works across your apps (Outlook, Teams, SharePoint, the web) to gather, analyze, and produce. Best for cross-app research, monitoring, and reconciliation.
Not yet priced — Microsoft Scout is pre-GA, so it is excluded from the cost total rather than guessed at.
✅ Build a Copilot Studio agent in 8 steps
The most common build in this library. Every agent's card gives you the exact name, Instructions, knowledge source, and starter prompts; this is the general flow in the current Copilot Studio experience.
- Go to copilotstudio.microsoft.com, sign in, and pick your environment if prompted.
- Create a Microsoft Copilot agent — the type that extends Microsoft Copilot with your own knowledge. In the new Home it's under Other ways to build; in the classic Home it's on the Create page. Don't use the main Agent card — that builds on the GitHub Copilot harness, which bills Copilot Credits from the moment you start building.
- Set the agent name and paste the Instructions from the agent's card.
- Connect the source the card names, and stop it falling back on the open web or its own training:
New experience: remove the “Search all websites” chip (✕). Classic: leave Web browsing off and turn on Only use specified sources.
- SharePoint sites, libraries, and lists — the default for governed policies, procedures, trackers, and team content.
- OneDrive or attached files — use for bounded pilots and one-time analysis; move durable shared knowledge to a governed team location.
- Teams and Outlook context — add only when answers depend on meetings, chats, channels, or mail the signed-in user can access.
- Public websites — allow only named, authoritative domains; keep general web fallback off for controlled Q&A.
- Dataverse or structured business data — use governed tables/views with least-privilege access when the scenario needs current records.
- Line-of-business systems — add an IT-approved connector or tool, prefer read-only operations, and test its separate licensing and billing behavior.
- Add the card's starter prompts so people know what to ask.Classic: the Suggested prompts section on the Configure page. New experience: ⋯ → Settings → Greeting & prompts.
- Test it with 3–5 de-identified or approved test examples and refine the Instructions until the output is reliable.Classic: the Test pane. New experience: the Preview tab.
- Confirm it respects permissions and sensitivity labels and keeps a human in the loop where the card notes it.
- Publish, then share it — a Microsoft Copilot agent is not deployed to Copilot and Teams just by publishing. On Availability options, use Show to my teammates for a pilot or Show to everyone in my org to send it for admin approval.New experience: Channels → Add channel → Teams + Microsoft 365.
📋 Before you build — a quick checklist
- Is the data source reachable today (already in Microsoft 365), or does it need an authoritative SharePoint library, a one-time file export, or an IT-approved connector first?
- Does the requesting department have permission to the underlying data? The agent inherits the user's access.
- Is any of the data sensitive or PHI? These agents handle the administrative work around care, not clinical decisions — where one touches PHI, use only the minimum necessary and follow your privacy review.
- Who owns the agent after launch, and how will people find and use it?