“AI-powered ERP” is now a common phrase. For organisations running Dynamics 365 Finance & Operations, the more useful question is narrower: which kinds of automation and AI assistance are available on the platform, what do they need to work well, and where should you start?

Three layers of automation

It helps to separate automation into three layers, because they have very different requirements.

1. Deterministic process automation

This is the most mature layer and the one with the most immediate return:

  • Workflow in D365 F&O for approvals of purchase requisitions, vendor invoices, journals and other documents.
  • Business events that announce when something has happened — an order confirmed, an invoice posted — so other systems can react without polling.
  • Power Automate flows triggered by those events for notifications, approvals in Teams or Outlook, document routing and hand-offs to other applications.
  • Batch jobs and recurring integrations that remove manual imports, exports and scheduled tasks.

None of this involves machine learning. It does involve good design: clear ownership, error handling and monitoring, so that flows do not become invisible dependencies.

2. Predictive and optimisation features

Several D365 F&O capabilities use statistical models or optimisation engines — for example in cash-flow forecasting and customer payment predictions in finance, demand forecasting in supply chain planning, and the planning engine used for master planning. Their quality depends directly on history: consistent transactional data, correct master data and enough of it.

3. Generative AI assistance

Copilot capabilities in Dynamics 365 bring generative AI into the user experience — summarising records and workspaces, drafting communications, answering questions about data and helping users find their way in the application. Microsoft extends this area in each release wave, so the specific features available depend on your version, region and licensing; the release plans are the reliable reference.

Generative assistance is most useful where people spend time reading, summarising or writing around ERP data. It does not replace the need for correct processes and data underneath.

The foundation is the same for all three

Whatever the layer, the same prerequisites decide the outcome:

  • Clean master data — customers, vendors, items and financial dimensions that are complete and consistent.
  • Processes that are actually followed in the system, not in spreadsheets beside it.
  • A deliberate data path for analytics and AI scenarios, rather than ad-hoc queries against the transactional database.
  • Security and governance — which data an assistant or flow can see, and who owns each automation.

Organisations that invest here first get more out of every subsequent automation, AI-based or not.

Where to start

A pragmatic sequence for most D365 F&O environments:

  1. List the manual work around the ERP — approvals chased by email, data re-keyed from documents, recurring exports into spreadsheets.
  2. Automate the deterministic parts first with workflow, business events and Power Automate. These deliver value quickly and expose data-quality issues.
  3. Fix the data issues those automations reveal. They will also limit any predictive or generative feature.
  4. Pilot AI features where the data is ready — and measure whether they save time for the people using them.

AI in ERP is real and useful, but it rewards organisations whose processes and data are already in order. The work that makes ERP reliable is the same work that makes it ready for AI.