AB-100 Dynamics 365 integration
AB-100 expects you to know which Dynamics 365 app offers which AI capability, and when Dataverse is (and isn't) the right place for data.
1. D365 AI Capabilities by App
Each Dynamics 365 application has distinct AI/Copilot capabilities. The exam tests whether you know WHICH capability belongs to WHICH app.
| D365 App | AI Capabilities |
|---|---|
| D365 Finance | AI-powered forecasting, automated journal entries, cash flow predictions, budget proposals |
| D365 Supply Chain | Demand forecasting, inventory optimization, predictive maintenance, supply chain disruption alerts |
| D365 Sales | Copilot for Sales: email drafts, meeting summaries, opportunity insights, lead scoring |
| D365 Customer Service | Copilot for Service: case summaries, knowledge article suggestions, draft responses |
| D365 Contact Center | Agent integration with contact center channels (voice, chat, email routing) |
Memory Aid
- Finance = money predictions (cash flow, budgets, journals)
- Supply Chain = stuff predictions (demand, inventory, maintenance)
- Sales = relationship help (emails, meetings, leads)
- Customer Service = resolution help (cases, knowledge, drafts)
- Contact Center = channel routing (voice, chat, email)
2. Copilot Feature Prerequisites
The exam frequently tests what must be configured BEFORE a Copilot feature works. Getting the prerequisite wrong is a common trap.
| D365 App | Copilot Feature | What It Does | Prerequisite |
|---|---|---|---|
| Finance (Accounts Payable) | Prebuilt AI invoice processing | Automates invoice data extraction and matching | Assign Finance & Operations AI security role |
| Contact Center (Live handoff) | Agent transfer to live rep | Escalates from bot to human with full context | Copilot Studio + Customer engagement hub |
| Supply Chain (Orders) | AI-driven insights from orders | Surfaces patterns and anomalies in order data | Enable AI Summaries with Copilot |
| Supply Chain (Planning) | Anticipate future needs | Generates demand forecasts with AI | Enable Generative insights for Demand planning |
Key Traps
- Finance AI invoice processing requires a security role, not a license or feature flag
- Contact Center live handoff requires two components (Copilot Studio AND engagement hub)
- Supply Chain has two different Copilot features with two different prerequisites
3. D365 Customer Service Escalation Flow
This is a high-frequency exam pattern. Know the exact sequence and what transfers between tiers.
Customer contacts support
|
v
Copilot Studio Agent (Tier 0 - Self-Service)
|-- Resolves autonomously? --> Done. Case logged in Dataverse.
|
|-- Cannot resolve?
v
Transfer to Omnichannel for Customer Service
|-- Full conversation context transfers automatically
|-- No "please repeat your issue"
v
Human Agent (Tier 1+)
|-- Sees full conversation history
|-- Receives AI-generated summary of the interaction
|-- Case record updated in Dataverse
v
ResolutionCritical Details
- Full conversation context transfers -- the customer does NOT repeat themselves
- AI-generated summary is provided to the human agent at handoff
- Case record lives in Dataverse throughout the entire flow
- The agent and the human both read/write to the same Dataverse tables
4. Dataverse as the Integration Backbone
4.1 When to Use Dataverse
Dataverse is the default data store for anything touching D365 or Power Platform. The exam tests whether you can identify when Dataverse is the RIGHT choice and when it is NOT.
| Scenario | Dataverse | SharePoint | Custom SQL | External API |
|---|---|---|---|---|
| Central glossary/taxonomy for agents | YES | |||
| Agent governance metadata | YES | |||
| D365 integration (Sales, Service) | YES | |||
| Row-level / field-level security | YES | Partial | ||
| Unstructured document storage | YES | |||
| High-volume transactional data (millions/sec) | YES | |||
| Custom ML model inference | YES |
4.2 When Dataverse Is NOT the Answer
This is just as important as knowing when it IS the answer. Five scenarios where you should NOT use Dataverse:
| Scenario | Use Instead | Why |
|---|---|---|
| High-throughput streaming data | Azure Event Hubs | Dataverse is not designed for millions of events/sec |
| Unstructured blob storage | Azure Blob Storage / SharePoint | Dataverse stores structured rows, not files |
| Custom ML model hosting | Azure AI Foundry | Dataverse has no compute for model inference |
| Non-Microsoft SaaS as system of record | Custom connector to source system | Don't replicate external SaaS data into Dataverse |
| Data that never touches D365/Power Platform | No Dataverse needed | No benefit to adding a layer |
4.3 Decision Tree: Where Does This Data Go?
Use this tree in order. The FIRST match wins.
Step 1: Is it D365 transactional data?
YES --> It's already in Dataverse. Don't move it. STOP.
Step 2: Is it governance/config metadata?
YES --> Dataverse custom table. STOP.
Step 3: Is it high-volume streaming data?
YES --> Event Hubs / Azure Data Explorer. STOP.
Step 4: Is it unstructured documents?
YES --> SharePoint or Azure Blob Storage. STOP.
Step 5: Is it external SaaS data?
YES --> Custom connector to the source system. Don't copy it. STOP.
Step 6: Is it ML model data?
YES --> Azure AI Foundry. STOP.5. D365 Sales + Agent Integration
Core Principle: No Data Duplication
When a Copilot Studio agent works with D365 Sales data, it reads and writes the same Dataverse tables that the D365 Sales UI uses. There is no separate data store, no sync job, no ETL.
Three Integration Rules for D365 Sales
| Rule | What It Means |
|---|---|
| Use existing Dataverse tables | Don't create parallel tables. The agent uses Account, Contact, Opportunity, Lead -- the same tables D365 Sales uses. |
| Respect the user's security context | The agent operates under the same Dataverse security role as the user. If the user can't see a record in D365 Sales, the agent can't see it either. |
| Solution-aware packaging | Agent customizations must be packaged in Dataverse solutions for proper ALM (dev/test/prod promotion). |
6. Cross-D365 Integration Rules
These five rules apply to ALL D365 integration scenarios, not just Sales. Memorize them.
| # | Rule | Exam Trap if Violated |
|---|---|---|
| 1 | Never duplicate Dataverse data | Any answer suggesting "copy D365 data to a custom table" is wrong |
| 2 | Security context matters | Agents must respect the user's Dataverse security role |
| 3 | Solution-aware packaging | Unmanaged customizations outside solutions = wrong answer |
| 4 | Environment strategy | Dev/test/prod separation is required, not optional |
| 5 | Native Dataverse integration first | Within D365/Power Platform, connectors are secondary to native integration |
Rule 5 Clarification
- Within D365/Power Platform: Use native Dataverse access (tables, views, security roles). Do NOT use a "Dataverse connector" to access data that's already in the same environment.
- Outside D365/Power Platform: Connectors (standard or custom) are the correct pattern to bring external data in.
7. Security Pattern Decision Tree
When the exam describes a security requirement, map it to the correct Dataverse security mechanism.
| Security Requirement | Correct Mechanism | Key Detail |
|---|---|---|
| Regional data isolation (e.g., EU users only see EU data) | Business Unit scoping | BUs create data boundaries by org structure |
| Manager sees subordinate records | Hierarchy security | Follows the manager hierarchy in Dataverse |
| Hide sensitive columns (SSN, salary) but not entire rows | Field-level security | Applied per field, per security profile |
| Ad-hoc access to specific records | Access teams | Dynamic membership, record-level sharing |
| Use Entra ID groups for Dataverse access | Entra Group teams | Maps Entra groups to Dataverse teams automatically |
| Agent acts on behalf of a signed-in user | Delegated access | Agent inherits the user's permissions |
| Agent runs as a background service | Application user + dedicated role | Service principal with its own security role |
Two Agent Security Modes
This distinction appears frequently on the exam:
| Mode | When to Use | How It Works |
|---|---|---|
| Delegated access | Agent is helping a specific user in real-time | Agent can only see/do what THAT user can see/do |
| Application user | Agent runs unattended (batch, scheduled, background) | Agent has its own identity and dedicated security role |
8. Knowledge Sources for D365 Agents
When an agent needs to answer questions using D365-related knowledge, the exam tests which knowledge source architecture to use.
| Knowledge Source | Best For | NOT For |
|---|---|---|
| File upload (to Copilot Studio) | Adding knowledge WITHOUT creating new topics; quick, simple docs | Large-scale or frequently changing content |
| Azure AI Search | Unifying heterogeneous data (SQL + files + APIs) into one searchable index | Simple, static documents |
| Cosmos DB | NoSQL document storage at scale | Relational data; not a knowledge search tool |
| Data Lake | Raw data storage and analytics | Direct agent knowledge retrieval |
| Dataverse | Structured business data already in D365 | Unstructured documents, blob storage |
Key Distinction
- File upload = simplest path. "Add knowledge" means file upload, not new topics.
- Azure AI Search = when you need to search ACROSS multiple data sources (heterogeneous).
- The exam will try to trick you into picking Cosmos DB or Data Lake for agent knowledge. These are storage, not knowledge retrieval mechanisms.
9. D365 Business Terms Configuration
What It Is
D365 allows you to configure business terms -- industry-specific terminology definitions so that AI features interpret domain language correctly.
Why It Matters for the Exam
- If a question asks how to ensure Copilot/agents understand industry jargon (e.g., "SKU" in retail, "NDC" in pharma), the answer is business terms configuration.
- This is NOT the same as training a custom model. Business terms are a configuration setting, not a model customization.
- Business terms are defined at the D365 environment level and apply across AI features in that environment.
Summary: The Three Big Ideas
Dataverse is the center of gravity. If data touches D365 or Power Platform, it lives in Dataverse. Don't duplicate it, don't move it, don't create parallel stores.
Know each app's AI features AND prerequisites. The exam doesn't just ask "what can D365 Finance do with AI?" -- it asks "what must be configured first?" The prerequisite is often the differentiator between correct and incorrect answers.
Security is per-pattern, not one-size-fits-all. The correct security mechanism depends on the specific requirement (rows vs. columns vs. hierarchy vs. delegation). Map the requirement to the mechanism, don't guess.