1 / -
↑↓ navigate
SLIDE 01 / 37
Frontier Application Consultancy

AI Business Solutions

Giel Cuijpers & Marvin Heilemann
Enterprise Partner Solution Architects
SLIDE 02 / 37

Agenda

  • Frontier Firms & Agentic CRM/ERP
  • Becoming an AI-first implementation partner
  • Vibe Configuration with Dynamics 365
  • Responsible AI
SLIDE 03 / 37

The Microsoft Why

  • Microsoft internal info. Delete slide when presenting to a partner or customer audience.
  • This is a Partner Zero play. Why invest effort in Frontier Application Consultancy?
  • Increase partner pre-sales and delivery capacity
  • Decrease Dynamics 365 implementation cost and lead-time
  • Develop additional partner skills to enable additional AI revenue streams
  • Combining business process knowledge with AI skills leads to more impactful solutions
  • Drive partner internal consumption of M365 Copilot, Github Copilot, Copilot Studio Credits and ACR.
  • Microsoft internal info. Delete slide when presenting to a partner or customer audience.
SLIDE 04 / 37

What we’re hearing from partners

  • There is a shortage of skilled consultants and therefore a drive to reduce the ”boring work”.
  • Dynamics 365 implementations can be long leading, so costly for customers, with uncertain outcomes.
  • Customers are considering agents as part of a M365 Copilot rollout and want to expand to CRM and ERP.
  • We need to develop a skillset for the AI era.
  • Customer key user education is challenging.
  • Key users have little time to deal with Dynamics 365 specifics.
SLIDE 05 / 37

Your Best People Deserve Better Than Boring Work

On a typical D365 implementation, 25% of expert time disappears into tasks that AI + MCP can handle.

📅 Today. Where your team’s time actually goes
Strategic 35%
Delivery 40%
Boring Work 25%
✅ With AI + MCP. Where your team’s time should go
Strategic 45% ▲ +10%
Delivery 50% ▲ +10%
5%
AI + MCP handles this GitHub Copilot · D365 ERP/Dataverse MCP · ADO MCP Server · Copilot Studio

On a 10-person D365 team, 25% boring work = 2.5 FTEs spending their day on tasks AI can do. With Azure DevOps MCP, D365 MCP & GitHub Copilot, that time flows back to what your consultants love.

~12% of total project effort reclaimed per engagement
SLIDE 06 / 37

The Opportunity to decrease implementation cost and time

Traditional LoB implementations are lengthy, costly, and can be risky.

Customer TCO
Traditional implementation cost and timeline 17 months1(ERP example)
+ Accelerated implementation cost and timeline
  • 20–30% timeline acceleration in migrating / modernizing workloads2
  • 30–40% lower cost of migration and modernization of workloads2
  1. International Research Journal of Engineering and Technology, “The Future of ERP Integrations: A Look at Emerging Technologies” – 2024
  2. McKinsey, “In search of cloud value: Can generative AI transform cloud ROI?” – 2023
SLIDE 07 / 37

How to become an AI-first implementation partner

SLIDE 08 / 37

How do I organize my practice?

The five drivers of AI value, anchored on one shared goal: Frontier Transformation.

01

Business Strategy

The alignment of AI investments with strategic business outcomes and measurable impact.

02

Technology & Data Strategy

The data and infrastructure needed to run AI solutions at scale, securely and reliably.

03

AI Strategy & Experience

The organization’s AI expertise and repeatable processes to create sustainable value.

04

Organization & Culture

The vision, operating model, skills, resources, and culture to drive AI adoption.

05

AI Governance & Security

The controls and processes to ensure secure, compliant and responsible AI.

SLIDE 09 / 37

AI-First Toolbelt Opportunities

AI-FIRST FRAMEWORK

AI-First Consulting as the strategic approach to integrate AI across every process and engagement, from presales to delivery, enabling faster, outcome-based delivery and greater efficiency.

MCP Tools Used across the full engagement lifecycle
 
Pre-Sales
Delivery
Dataverse MCP Tool
Rapid configuration setup.
Entity & field configuration assistance directly from the IDE.
ERP MCP Tools
Rapid configuration setup.
Accelerated entity management and understanding.
Development MCP Tool
Rapid prototyping and demo preparation.
Accelerated development.
Agents Pre-Sales Exclusive Delivery Exclusive
  • Pre-SalesArchitect CockpitPre-sales solutioning platform helping architects to visualize solution design, architecture reviews, and proposal scoping.
  • Pre-SalesPursuit AgentDeal insights and support. Identifies competitive positioning, staffing assistance.
  • Pre-SalesSoW Generation / Review AgentRapid solution design, architecture reviews, and proposal scoping.
  • BothKnowledge Management AgentIntellectual property management and reuse across projects.
  • DeliverySDLC AutomationGitHub Copilot & Azure DevOps AI Assistant. Bottleneck identification, backlog creation at scale.
Enabled by M365 Copilot Copilot Studio Visual Studio Code
SLIDE 10 / 37

Evolution of business applications

Three layers of every business application are being rewritten at the same time.

Surface
Apps as UI
Copilot as UI
Logic
Workflows
Agents
Substrate
Data
Knowledge
SLIDE 11 / 37

Frontier Application Consultancy · Approach

Taking an AI-first approach to Application Consultancy revolves around a data-driven approach using Clients, Tools and Knowledge.

Pre-Sales Design
  • RFP agentDraft RFP/RFI responses.
  • Personalized customer demoBased on customer data & public info.
  • Solution BlueprintDraft, review, evaluate.
  • Fit gap analysisMatch customer processes with the Process Repository.
Delivery · Success By Design Build
  • Implementation workshopsCopilot assisted.
  • Customer kickoffIncludes Copilot training.
  • Task initiationPopulate DevOps from Blueprint.
  • AI assisted codingGenerate FDD, TDD, IDD using workshops & ADO as input.
  • Configuration documentationGenerated from the application.
  • AI assisted configurationUsing knowledge and MCP.
  • AI assisted testingUsing MCP.
  • Test planWith evals defined.
  • Ops measurementAgent that measures batch actions and transactions.
KnowledgeWork IQ · Fabric IQ · Foundry IQ
ToolsMCP servers
ClientsM365 Copilot · GHCP · Copilot Studio

Scenarios are potential use cases, not productized capabilities.

SLIDE 12 / 37

Bring work context into your apps and agents with Work IQ

Work IQ stitches productivity data and external systems into a single work-context layer that powers every Copilot surface.

Sources
Microsoft 365 Productivity data
Copilot Connectors External systems
Work IQ Work context Reasoning over your work, ready to be queried by any Copilot client.
Microsoft 365 Copilot
VS Code & GitHub Copilot
Custom GenAI Solution
SLIDE 13 / 37

MCP standardizes how AI agents discover and use enterprise tools and data

  • MCP standardizes how AI agents discover and use enterprise tools and data
  • Browser automation
  • Data and File Stores
  • Model Context Protocol (MCP)
  • Code Editors
  • Developer Tools GitHub, Playwright, Azure
  • Applications
  • Productivity Tools
  • M365, Teams
SLIDE 14 / 37

Azure DevOps MCP

Can I use the Azure DevOps MCP within Copilot Studio agents?

Yes, but with a caveat. Copilot Studio requires an HTTP/SSE-based remote MCP server, not a local stdio one. Microsoft recently released the Azure DevOps Remote MCP Server in public preview. This is exactly what you connect to Copilot Studio.

The local npx @azure-devops/mcp version used in VS Code cannot be directly connected to Copilot Studio. Use the remote HTTP/SSE endpoint for all agent platforms outside the IDE.

  1. 1Onboard to the Azure DevOps Remote MCP Server → learn.microsoft.com/azure/devops/mcp-server/remote-mcp-server
  2. 2In Copilot Studio, navigate to Tools → Add a tool → Model Context Protocol
  3. 3Point the agent at the remote ADO MCP endpoint and sign in with your Azure DevOps organization credentials
  4. 4Scope tool access with domains (work-items, pipelines…) to keep the agent focused and avoid tool overload
Can I do mass / bulk upload of user stories & requirements (e.g. based on the Business Process Catalog)?
mcp_ado_wit_add_child_work_items

Creates multiple child items in one call under a parent work item; e.g. create 10 User Stories under Feature #123, each with title, description, story points & iteration path.

mcp_ado_wit_update_work_items_batch

Updates multiple existing items in batch. Reassign owners, change priority, move to sprint, update state. All in a single operation.

mcp_ado_wit_create_work_item

Creates any work item type (Epic, Feature, Story, Task, Bug) with full field support including Markdown description via the format parameter.

SLIDE 15 / 37

Empower Project Teams to Reshape DevOps Delivery

The Azure DevOps MCP Server enables a two-step process: data retrieval and AI analysis.

Data Retrieval Options

Azure DevOps MCP Server gives your AI assistant secure, real-time access to client project data in Azure DevOps.

  • Projects & Teams Organization structure, team membership, areas & iterations. Instantly understand any client environment.
  • Work Items Assigned tasks, bugs, user stories, epics. With states, priorities, blockers & sprint assignments.
  • Pull Requests Code review status, pending reviewers, linked work items & change scope (+/− lines).
  • Build & Pipelines CI/CD run history, test results, deployment stages & failure diagnostics.
  • Test Plans Test cases, execution results & coverage data. Surface quality risks before client demos.
AI Analysis Options

Your AI assistant processes the retrieved project data to deliver client-ready insights your project team previously spent hours compiling.

  • Sprint & Project Health On Track At Risk Blocked
    Instant sprint health summary with velocity, burndown & bottleneck analysis. Ready for client status calls in seconds.
  • Risk Identification Surfaces stale PRs, flaky pipelines, blocked work items & incomplete test coverage before they escalate. Proactive risk reporting for every engagement.
  • Context-aware Summaries Generates standup notes, sprint reviews & client handover reports that understand your team’s terminology, process & backlog history. Cuts context switching.
  • Actionable Recommendations “Follow up with Integration Team on Feature #4423 (blocked 5 days)” Natural language answers that recommend concrete next steps. No more navigating through ADO portals.
SLIDE 16 / 37

What is Dynamics 365 Agentic CRM?

A layered architecture where CRM solutions, headless business services and a CRM agent foundation come together behind one Copilot-driven experience.

01
Copilot adaptive experience A single conversational surface that reshapes itself for every role and task.
02
Extensibility · Industry verticals Partner-built extensions and vertical accelerators that ride on the same fabric.
03
Headless business services Actions and data tools, powered by the Model Context Protocol (MCP).
04
Dynamics 365 CRM solutions Sales, Customer Service, Field Service, Marketing and Customer Insights working as one composable suite.
05
CRM agent foundation Shared models, governance and runtime that every CRM agent inherits.
SLIDE 17 / 37

Dataverse MCP Server Tools

  • DYNAMICS 365  •  DATAVERSE
  • SCHEMA TOOLS
  • Find entities, get entity metadata, fields, and relationships
  • Create / delete entities, fields (all types), option sets, and relationships
  • Create alternate keys and many-to-many relationships
  • DATA TOOLS
  • Execute SQL SELECT queries against
  • Dataverse TDS endpoint
  • Run INSERT / UPDATE / DELETE SQL
  • Commands and manage option set value  mappings
SLIDE 18 / 37
DYNAMICS 365 · DATAVERSE

CRM MCP Server Tools

Six tool categories that let an agent operate Dynamics 365 customer engagement & Dataverse directly. From schema design to plugins, navigation and process flows.

SCHEMA TOOLS
  • Find entities. Get entity metadata, fields and relationships
  • Create or delete entities, fields, option sets and relationships
  • Create alternate keys and many-to-many relationships
DATA TOOLS
  • Execute SQL SELECT against the Dataverse TDS endpoint
  • Run INSERT, UPDATE and DELETE in SQL
  • Manage option set value mappings
  • Bulk import and export records by FetchXML
FORM & UI TOOLS
  • Get and create forms. Add or remove tabs and sections
  • Add or remove fields, subgrids and web resource controls
  • Validate web pages rendered inside model-driven apps
APP & NAVI TOOLS
  • List, create and configure App Modules with sitemap
  • Add entities and web resources to app navigation
  • Create, list and delete ribbon buttons with JS handlers
  • Publish web resources (JS, HTML, CSS, SVG, RESX, images)
PLUGIN TOOLS
  • Compile, deploy and manage plugin assemblies and packages
  • Register or unregister plugin types and pipeline steps
  • Add pre-image and post-image to plugin steps
PROCESS & API TOOLS
  • Create, update and activate Business Process Flows
  • Register Custom APIs with request and response parameters
  • Unregister Custom APIs. Execute Dataverse HTTP actions
SLIDE 19 / 37

Custom CRM MCP Server – Build Effort Estimate

DYNAMICS 365 · DATAVERSE · .NET
Scope Approx. tools Complexity Notes
Schema tools ~6 tools Low Simple Dataverse metadata API calls
Data tools ~5 tools Low SQL passthrough + Web API proxy calls
Form & UI tools ~8 tools Low Read/write system form, web resource tables
App & Navi tools ~6 tools Medium Sitemap XML manipulation + publishing
Plugin tools ~6 tools Medium Assembly upload via base64, step registration
Process & API tools ~5 tools Medium BPF workflow activation flow + Custom API reg.
Security tools ~4 tools Low Role assignment via standard SDK calls
Total ~40 tools across 7 categories · 2–4 days with .NET + Dataverse SDK
SLIDE 20 / 37

CRM Vibe Configuration Tactics

  • Use a copilot-instructions.md file. An example can be found here: Reference workspace files.
  • When adding files, use folders to make referencing groups of documents easier.
  • Build specialized skills (see below) to do the work. For instance a Dynamics Services Configuration skill.
  • Ensure proper security setup to limit the options available in CRM. This decreases the tool options the LLM has to reason over.
  • It is possible to use multiple MCP connections at the same time. Great for comparing configurations between legal entitities or environments. Do create separate Workspaces to avoid connection mix ups.
  • About agent skills – GitHub Docs
SLIDE 21 / 37

What is Dynamics 365 Agentic ERP?

A layered architecture where ERP solutions, headless business services and a ERP agent foundation come together behind one Copilot-driven experience.

01
Copilot adaptive experience A single conversational surface that reshapes itself for every role and task.
02
Extensibility · Industry verticals Partner-built extensions and vertical accelerators that ride on the same fabric.
03
Headless business services Actions and data tools, powered by the Model Context Protocol (MCP).
04
Dynamics 365 ERP solutions Finance, Supply Chain, Project Operations, Commerce and Human Resources working as one composable suite.
05
ERP agent foundation Shared models, governance and runtime that every ERP agent inherits.
SLIDE 22 / 37

Multiple Dynamics 365 ERP MCP servers

  • ERP Client MCP
  • Dataverse MCP
  • ERP Analytics MCP
  • Work directly with backend data to update tables and validate records at the source
  • Act through the UI & Data Entities to trigger business logic and follow standardized processes
  • Tap into pre-structured analytical data to uncover insights and measure performance trends
    • More tools, smarter agents, better outcomes
SLIDE 23 / 37
ERP MCP SERVER · SCALE
500,000+
ERP AI actions

An agent can do almost anything a person can do in ERP.

Every function a Dynamics 365 user can perform is now exposed to an agent through Microsoft’s ERP MCP server. This is the unlock that turns Agentic ERP from a roadmap slide into something that ships.

SLIDE 24 / 37

ERP MCP Server Tools

  • Data tools
  • Name format data_*
  • Find data entities and get entity meta data. Also custom data entities.
  • Perform CRUD operations
  • Form tools
  • Name format form_*
  • Find menu items. Also custom menu items.
  • Manipulate forms (open form, sort, filter, perform lookups, select rows, set control values, save, close)
  • Action tools
  • Name format api_*
  • Find and invoke actions
  • Can be extended: Create AI plugins for copilots with finance and operations business logic (preview) - Finance & Operations | Dynamics 365 | Microsoft Learn
SLIDE 25 / 37

CE AI Assisted Configuration

CE AI Assisted Configuration
▶ Video placeholder

CE AI-assisted configuration demo. The PPTX shows a screenshot. Replace with a hosted video link before public sync.

SLIDE 26 / 37

CE AI Assisted Configuration Advised Setup and Cost

  • Setup
  • Use VS Code and Github Copilot Pro (or better).
  • D365 Premium license, or M365 Copilot license, or Copilot Studio Credits setup (see Running Costs below)
  • Install Python library (required when working with Excel files).
  • Select a model with good MCP capability. Currently advised are Anthropic models Sonnet & Opus.
  • Dataverse MCP Server configuration: Connect to the Dataverse MCP server by using Visual Studio Code
  • Allow GitHub Copilot MCP client in your Power Platform environment
  • MS Learn MCP Server configuration: Microsoft Learn MCP Server overview
  • Running Costs
  • For a user working with VS Code + local stdio MCP server + a Dynamics 365 or Power Apps per-user license: The only incremental cost is the GitHub Copilot subscription. The Dataverse API calls are well within the 40k/day entitlement for a normal user.
  • Official documentation
  • MCP Server Licensing, Copilot Studio licensing, Copilot Studio Billing rates and management
SLIDE 27 / 37

Slide 29

Slide 29
▶ Video placeholder

CE AI-assisted configuration demo (~174 MB embedded video). Replace with a hosted video link before public sync.

SLIDE 28 / 37

F&O AI Assisted Configuration

  • Feature Configuration
  • Feature Use (Demo)
  • Running Cost
SLIDE 29 / 37

F&O AI Assisted Configuration Advised Setup and Cost

F&O AI Assisted Configuration Advised Setup and Cost
▶ Video placeholder

F&O AI-assisted configuration demo. The PPTX shows a screenshot. Replace with a hosted video link before public sync.

SLIDE 30 / 37
DEMO SCENARIO · F&O WAREHOUSE

Zava DIY Warehouse Implementation

A simplified warehouse layout used in the F&O AI-assisted configuration demo. Goods flow from inbound on the left, through storage and picking in the centre, to outbound on the right.

→
Reception
↕Production access
Z2.1
Z2.2
Z2.3
Zone group 2
Z1.1
Z1.2
Z1.3
Zone group 1
Bulk & PickingRack · Level · Bin
Picking (fixed locations)
Packing
Staging bay
→
Bin location coordinates
F1Aisle
14Row
2Level
3Bin
SLIDE 31 / 37

Demo video

Demo video
▶ Video placeholder

Finance and Operations demo (~153 MB embedded video). Replace with a hosted video link before public sync.

SLIDE 32 / 37

F&O Vibe Configuration Tactics

  • Use a copilot-instructions.md file. An example can be found here: Reference workspace files.
  • When adding files, use folders to make referencing groups of documents easier.
  • Build specialized skills (see below) to do the work. For instance an Accounts Payable Configuration skill.
  • Ensure proper security setup to limit the options available in ERP. This decreases the tool options the LLM has to reason over.
  • It is possible to use multiple MCP connections at the same time. Great for comparing configurations between legal entitities or environments. Do create separate Workspaces to avoid connection mix ups.
  • About agent skills – GitHub Docs
SLIDE 33 / 37

Responsible AI

SLIDE 34 / 37

Design responsible user experiences

  • Be transparent about AI’s roles and limitations, e.g. highlight potential mistakes and disclose AI’s role
  • Ensure humans stay in the loop, e.g. encourage human intervention and restrict automatic posts to social media
  • Mitigate misuse and overreliance, e.g. provide data citations and prepare pre-determined responses
  • Provide documentation, user guidelines, and best practices
SLIDE 35 / 37

Evaluation is an ongoing, iterative process

  • Provide inputs (prompts)
  • Generate outputs (completions)
  • Learnings
  • Evaluate outputs
  • Metrics
SLIDE 36 / 37

Next steps

SLIDE 37 / 37

Closing