Case study · Healthcare
Giving care coordinators their afternoons back
A care provider's coordinators were spending more of the day writing up contacts than making them. We implemented Dynamics 365 CRM with Copilot so the notes draft themselves — with a person reviewing every one before it reaches the record.
Engagement · Implementation & AI rollout
CRM and Copilot for a clinical coordination team.
- Industry
- Healthcare
- Solution
- Care coordination CRM
- Platform
- Dynamics 365 CRM
- AI
- Copilot, human-reviewed
- Engagement model
- Dedicated product team
- Scope
- Implementation, AI integration
- Users
- Care coordinators, clinicians
- Constraint
- Patient data governance
Outcomes
Where the coordinator day actually went
The bottleneck in care coordination is almost never the calls. It is what has to be written down afterwards.
-
60%
Less time writing up contacts
Copilot drafts the summary from the call notes and the record; the coordinator corrects and approves it rather than composing from a blank box.
Measured across coordinator time logs before and after
-
100%
Of AI drafts reviewed before saving
No Copilot output reaches a patient record without a named person approving it. In a clinical context an unreviewed summary is a safety risk, not a convenience.
By design: approval required, recorded against the user
-
3x
Faster shift handover
A structured handover summary is generated from the day's contacts rather than assembled by the outgoing coordinator from memory and notes.
Measured at shift change
Context
The situation before the implementation
Why coordination work quietly turns into administration.
The organisation
A community care provider coordinating visits, assessments and follow-ups for patients across a region, with coordinators as the point of contact.
The starting point
Contacts were logged in a shared system with free-text notes. Quality varied by who wrote them, and the write-up took longer than the call.
The trigger
Coordinators were making fewer contacts because each one cost twenty minutes of typing. The administrative load was reducing the care delivered.
What they wanted
A CRM shaped around a patient rather than a customer, with Copilot drafting the routine writing so coordinators spend their time on people.
Constraints
Patient data cannot leave the tenant · AI output must be reviewed by a person before it enters a clinical record · access has to be attributable · coordinators are not technical and the system must not need training nobody has time for.
System
What it runs at today
The system as it runs today.
-
1
Record per patient
Contacts, assessments and history in one place
-
2
Copilot uses
Contact summaries and shift handover drafts
-
100%
Access recorded
Every view of a patient record is attributable
-
0
AI text saved unreviewed
Approval is required, not encouraged
The engineering problem
Four problems with AI in a clinical record
Copilot is genuinely useful here. The care is in deciding exactly where it is allowed to act.
-
An unreviewed AI summary is a clinical risk
A generated summary that subtly misstates what a patient said becomes part of their record and informs later decisions. The convenience is not worth the exposure.
What we did
Copilot drafts; a named coordinator reviews, edits and approves. Nothing reaches the record without that step, and the approval is recorded against the person.
-
Patient data cannot leave the tenant
A care provider cannot send patient information to a general-purpose model, whatever the productivity argument.
What we did
Copilot operating inside the Microsoft tenant against the organisation's own data, with no patient information leaving the environment.
-
Coordinators will not fight the software
Anyone whose job is talking to patients will work around a system that slows them down, usually by keeping notes somewhere else.
What we did
The drafting step sits where the writing already happened, so approving a summary is faster than typing one. The path of least resistance became the correct one.
-
Access has to be provable
A care record carries the same obligations as any clinical record: a provider may be asked who accessed a patient's file and when.
What we did
Dynamics 365's auditing configured from the start, so access is recorded at the data layer rather than inferred from the interface.
Architecture
How it fits together
Simplified — the shape of the system rather than every service in it.
-
CRM
- Patient records
- Contacts & assessments
- Care plans
Dynamics 365 CRM modelled around a patient rather than a sales contact, which is mostly an entity-design decision.
-
Copilot
- Contact summaries
- Handover drafts
- In-tenant only
Copilot operating within the tenant, drafting from the organisation's own records.
-
Review layer
- Draft state
- Named approval
- Edit history
AI output held as a draft until a person approves it, with the edit recorded.
-
Governance
- Role-based access
- Auditing
- Retention
Access control and auditing configured at the platform level, not bolted onto the interface.
The architectural decision that matters is the review layer. Copilot writes into a draft state, never directly into the record — which is what makes the rest defensible.
Solutions
What we implemented
A CRM shaped for care, with AI in exactly one place and a person in front of it.
-
Patient-centred CRM
Dynamics 365 entities modelled around a patient, contacts and care plans.
-
Copilot summaries
Contact write-ups drafted from notes and record context.
-
Review workflow
Draft state with named approval before anything reaches the record.
-
Shift handover
Structured handover generated from the day's contacts.
-
Access governance
Role-based access with reads and writes audited.
-
Coordinator dashboard
The day's contacts, overdue follow-ups and escalations in one view.
Key capabilities
What it does day to day
Six capabilities across the coordination day.
| Capability | Runs | Refresh | What it does |
|---|---|---|---|
| Patient record | Coordinator | Continuous | Contacts, assessments, care plan and history in one view |
| Contact logging | Coordinator | Per contact | Notes captured at the point of the call |
| AI summary draft | Copilot | On request | A structured summary drafted from the notes and the record |
| Review & approve | Coordinator | Per draft | Edit and approve; nothing saves without a named approval |
| Shift handover | Copilot | Per shift | A structured handover generated from the day's contacts |
| Access audit | Automatic | Continuous | Every view of a patient record recorded and attributable |
Integrations
How the moving parts plug in
A call becomes a record through a person, never around one.
The contact
- Call or visitNotes taken live
- Patient contextHistory surfaced in view
- Care planWhat was due
Copilot draft
- Summary generatedFrom notes and record
- Held as draftNot yet part of the record
- Coordinator editsCorrections captured
The record
- Approved summary
- Attributed to a person
- Audit entry
Because the draft is a distinct state, the organisation can say precisely which text was machine-drafted and who approved it — which is the question a regulator asks about clinical AI.
Security & data
What protects patient information
Adding AI to a clinical system raises the governance bar rather than lowering it.
-
Data stays in tenant
Copilot operates against the organisation's own data inside its Microsoft tenant. No patient information goes to an external model.
-
Attributable approval
Every summary that enters a record carries the person who approved it and what they changed.
-
Role-based access
Coordinators, clinicians and administrators see what their role requires, enforced by the platform.
-
Access auditing
Reads as well as writes are recorded, so the provider can answer who saw a patient record and when.
The brief
AI in a clinical record is a governance design, not a feature toggle
Turning Copilot on is trivial. Deciding where it is permitted to act, what happens to its output before anyone relies on it, and how the organisation proves that afterwards — that is the actual work.
We built the review layer first and the productivity gain second, because the reverse order produces something nobody can defend.
- Copilot drafts, a person approves
- Patient data never leaves the tenant
- Approval attributed and recorded
- Access audited at the data layer
What the implementation had to respect
- 01Patient data governance obligations
- 02Coordinators who are not technical and have no training time
- 03A clinical record that must remain defensible
- 04Existing referral and scheduling systems that stay in place
Process
We settled the governance before switching Copilot on
The order matters. A pilot that puts AI text into records first creates a problem nobody can unpick afterwards.
-
Modelling the care record
Dynamics entities shaped around patients, contacts and care plans rather than leads and opportunities.
-
Governance design
Deciding where AI may act, what review means, and how approval is recorded — agreed with the clinical lead before build.
-
Review workflow first
The draft-and-approve mechanism built and tested before Copilot was connected to it.
-
Copilot integration
Summary and handover drafting, in-tenant, writing only into the draft state.
-
Coordinator rollout
A small group first, with the time saved measured rather than assumed, before wider release.
Technology
Dynamics 365 and Copilot, inside one tenant
Everything runs within the organisation's Microsoft environment, which is what makes the data position tenable.
CRM
- Dynamics 365 CRM
- Custom entities
- Care plan model
AI
- Copilot
- In-tenant grounding
- Draft-only output
Governance
- Role-based access
- Audit configuration
- Retention policy
Experience
- Coordinator dashboard
- Contact logging
- Handover view
Business impact
What changed for the service
Three things the coordination team noticed.
-
More contacts per coordinator
Time moved from typing to patients, which is the only productivity gain that matters here.
-
Handovers that hold
A structured summary at shift change beats a verbal handover from someone at the end of a long day.
-
A defensible AI position
The provider can explain exactly what the AI does, what it never does, and who approved every word of it.
The result
Coordinators spending the day on patients rather than paperwork
Contact summaries and shift handovers are drafted by Copilot inside the tenant and approved by a named coordinator before they reach the patient record.
The provider can state precisely what the AI does, what it is never permitted to do, and who signed off every word that entered a clinical record.
- Copilot drafting with mandatory human approval
- Patient data never leaving the Microsoft tenant
- Approval and access both attributable
- Coordinator time measurably moved onto care
What we hold to for AI in regulated records
- 01Build the review state before connecting the model
- 02Never let generated text write directly into the record
- 03Record who approved it and what they changed
- 04Measure the time saved rather than assuming it
Verified reviews
What clients say about our Dynamics work
Verified reviews from clients of ours on similar work, published on Clutch. They are not from this project.
-
Custom Software Development
Web-Based Coupon App Development
The work produced by the team was of exceptional quality.
Jon Thies
CTO, Model Rocket Feb 2022 – Mar 2023
-
Custom Software Development
Mobile & Web Platform Development & Design
Silver Scintilla was responsive, respectful, and enjoyable to work with, quickly addressing any questions or requests.
DeShawn Brown
CEO & Founder, Lithios Jun 2021 – Nov 2022
-
Custom Software Development
Web-Based HRM System Development
Silver Scintilla provided the best solution with scalable functionality, personalized features, and robust security.
Suheb Khan
Director, Collaborative Insight Technologies Jul 2020 – Jun 2022
Free discovery call
Have an idea? Let's turn it into AI-powered software.
Book a free discovery call with our experts. Share your idea and we will help you shape the scope, timeline and budget, under NDA.
- Free consultation
- NDA before we talk
- Transparent estimate