AI in D365 Finance & Operations

Artificial Intelligence in Dynamics 365 Finance and Operations is no longer just a roadmap promise; it is actively embedded into the application. For consultants, understanding AI requires separating the marketing buzz from the actual capabilities deployed in the system. The AI ecosystem in F&O falls into three primary categories.

• Generative AI (Copilot): The conversational interface powered by Large Language Models (LLMs). This includes the Copilot sidecar where users can ask natural language questions about their data, generate email drafts for collections, and summarize large datasets.
• Embedded Machine Learning: Traditional predictive AI models built directly into the business processes. This is branded heavily as Finance Insights and includes models that predict when a customer will pay an invoice or generate baseline cash flow forecasts based on historical trends.
• Extensibility: The ability to build custom AI solutions using the Power Platform. This includes using AI Builder to extract data from vendor invoices via OCR (Optical Character Recognition) or using Copilot Studio to build custom chatbots that query F&O data via OData endpoints.

Copilot in F&O provides users with an AI assistant that lives directly in the side panel of the application. It grounds its responses securely in the organization's ERP data.

Prerequisites for Enabling Copilot

Consultants must ensure the environment meets strict technical requirements before Copilot features can be activated.

• The environment must be running Dynamics 365 Finance version 10.0.38 or later.
• Power Platform Integration must be enabled within Microsoft Dynamics Lifecycle Services (LCS). Copilot relies on the Dataverse connection to process requests.

Configuration Steps

1. Assign Security Roles: In the Power Platform Admin Center, navigate to the connected Dataverse environment and assign the Finance and Operations AI security role to the users who need Copilot access. Without this role, the Copilot panel will not load.
2. Enable Features: Log into D365 F&O and navigate to System administration > Workspaces > Feature management. Search for "Copilot" and enable the core capabilities (like user experiences for Copilot and specific workspace integrations).

Consultant Actionable Use Cases

Once enabled, train your clients on high value use cases. For example, in the Credit and Collections workspace, users can click a button to have Copilot draft a personalized email to a customer summarizing their past due invoices. In procurement, users can ask Copilot to summarize changes made to a purchase order.

Finance Insights brings predictive Machine Learning models to the finance department. Instead of answering "what happened", these models attempt to answer "what will happen" based on historical data.

Prerequisites and Data Requirements

Unlike Copilot, Machine Learning models require significant historical data to train accurately. You must have a Tier 2 sandbox environment or higher. The minimum data requirements are strict:

• Customer Payment Predictions: Requires at least one year of historical customer invoice and payment data.
• Cash Flow and Budget Proposals: Microsoft strongly recommends at least three years of historical data for accurate trend analysis.

Additionally, the tenant must have AI Builder credits allocated (typically 20,000 credits per month are included with the required licensing).

Configuration Steps

1. Install the Add-in: Open Lifecycle Services (LCS), navigate to the environment details, and install the Finance Insights add-in. You may also need to install the Export to Data Lake add-in depending on your architecture.
2. Enable Feature Management: In F&O, enable the Finance insights feature in the Feature management workspace.
3. Check Provisioning: Go to System administration > Setup > Process automation and verify that the "Insights provisioning status check" background process is executing successfully.

Once provisioned, the models must be trained. Navigate to the specific workspaces (e.g., Credit and collections) to initialize the models. The models will score open invoices, predicting whether they will be paid on time, late, or very late, allowing collections agents to prioritize their daily calls.

Supply Chain Management (SCM) benefits heavily from AI, focusing on forecasting accuracy and inventory visibility.

Demand Planning and Azure ML

The native Demand Planning app utilizes Azure Machine Learning to generate baseline forecasts. As a consultant, you configure data profiles that feed historical sales data into the ML models. The models automatically select the best forecasting algorithm (like ARIMA or ETS) to predict future demand. You can also integrate external signals like weather data or macroeconomic indicators to refine the predictions.

Inventory Visibility with Copilot

For organizations with complex, multi-warehouse setups, the Inventory Visibility Add-in allows external systems to query stock levels in real time. With Copilot enabled, users can type natural language queries directly into the app, such as "Do we have any Surface Pro laptops available in the Seattle warehouse?" Copilot translates this into the necessary API calls and returns a human readable inventory summary.

When out-of-the-box AI features do not meet a specific business requirement, consultants must look to the broader Power Platform for extensibility.

AI Builder for Process Automation

AI Builder provides pre-built and custom models that integrate seamlessly with Power Automate. A classic consulting scenario is Accounts Payable automation. You can configure a Power Automate flow that triggers when a vendor emails an invoice. The flow uses the AI Builder Invoice Processing model to extract the vendor name, invoice number, and line items. It then pushes this structured data directly into a D365 F&O Pending Vendor Invoice journal via standard Data Entities, eliminating manual data entry.

Copilot Studio Custom Agents

If a client wants an internal chatbot that can answer proprietary questions, you build it using Copilot Studio. You can create a custom agent and configure its "Topics". For example, you can build an HR agent that queries F&O OData endpoints to answer employee questions about their remaining leave balances or expense report statuses.

During an implementation, the Chief Information Security Officer (CISO) will inevitably ask how Microsoft uses their financial data. A consultant must be prepared to confidently address governance, privacy, and security.

The Golden Rule of Microsoft AI

Your data is your data. Microsoft explicitly states that tenant data (including customer information, financial records, and prompts) is never used to train the foundational OpenAI models. The LLMs are stateless; they do not remember the data passed to them after the response is generated.

Role Based Access Control (RBAC)

Copilot adheres strictly to the existing D365 security model. If a user does not have permission to view payroll data in the standard F&O user interface, they cannot bypass that security by asking Copilot to summarize the payroll data. Copilot runs in the context of the logged-in user.

Data Residency

Depending on the geographic region of the F&O environment, the Azure OpenAI endpoint processing the requests might be located in a different data boundary. Consultants must check the Power Platform admin center to see if cross-region data movement consent is required and ensure the client formally approves it.

AI is heavily marketed, leading to inflated client expectations. The role of the consultant is to bridge the gap between hype and reality.

Scoping AI Realistically

Never treat AI as magic. Machine learning models require clean, consistent data to function. If a client is migrating from an old legacy system with terrible data hygiene, Finance Insights will generate terrible predictions. Consultants must advocate for aggressive data cleansing as a prerequisite for any AI initiative.

The Licensing Conversation

Be extremely clear about licensing early in the project. While basic Copilot features are included in the standard D365 F&O license, intensive operations like running custom AI Builder models consume capacity credits. If a client plans to process ten thousand invoices a month through AI Builder, you must calculate the required add-on capacity and present the true cost of ownership before building the solution.