+1 480 485 7373
Microsoft Teams
Four routes for putting AI to work on your Dynamics data
Copilot inside the apps
Business Central, Sales and Customer Service ship with Copilot features. The work here is enabling the right ones, setting data movement options, and teaching people when to trust a suggestion.
- ✓Bank reconciliation matching help
- ✓Sales line and item text suggestions
- ✓Case and email summaries
- ✓Chat over records the user can open
Copilot Studio agents
Agents answer questions and carry out steps against your data, in Teams, on a website or inside a Dynamics app. They can call Power Automate flows and connectors.
- ✓Order status agents for customers
- ✓Internal policy and procedure help
- ✓Agents that raise records or tasks
- ✓Handoff to a person when unsure
Azure OpenAI on your data
For needs the packaged tools do not cover: classification, extraction from documents, search across contracts or product manuals, with your own prompts and retrieval.
- ✓Retrieval over approved documents
- ✓Structured extraction to Dataverse
- ✓Logging of prompts and outputs
- ✓Cost limits set per workload
AI steps inside flows
AI Builder and prompt actions in Power Automate add reading, sorting and drafting to an automated process that a person still approves.
- ✓Invoice and receipt reading
- ✓Email triage into queues
- ✓Draft replies for review
- ✓Sentiment and category tagging
Why data readiness decides whether any of this helps
A language model answers from what it is given. If customer records are duplicated, item descriptions are blank, or the knowledge articles are three versions out of date, Copilot will repeat those faults politely and confidently. Much of our effort on these projects goes into the data, not the prompt.
In practice that means checking the tables and documents an agent will draw on, fixing ownership of each source, deciding which SharePoint folders are in scope and which are not, and removing stale content. It also means checking security: an agent that searches a document library can surface anything its identity can read, so permissions that were loose but harmless before now matter.
Taking one use case from pilot to everyday use
- 1
Choose a measurable task
Something frequent and narrow, with a clear before picture: how many emails, how many exceptions, how long each one takes today. Vague goals such as using AI more cannot be judged.
- 2
Prove it on real records
A small group uses the solution against live or copied data. We record where it is right, where it is wrong, and where it should hand over to a person.
- 3
Set the guardrails
Access, data sources, logging and the words the solution must never say are fixed in configuration, not left to training slides.
- 4
Widen carefully
More users are added in waves, with a named owner in the business who reads the feedback and decides on changes.
Governance and responsible use we put in place
- ✓An inventory of every agent and AI flow, with an owner and the data each one touches.
- ✓Environment and data loss prevention policies in the Power Platform admin center so agents cannot use unapproved connectors.
- ✓Human approval on any step that posts, pays, deletes or sends something outside the company.
- ✓Clear labels telling users when content was generated and should be checked.
- ✓Regular sampling of conversations and outputs to catch drift, gaps and wrong answers.
- ✓A written position on what staff may paste into public AI tools, separate from the approved ones.
Configure, build or hold off?
What buyers ask before an AI project
How is this different from your Copilot page?
The Copilot page explains the Copilot features Microsoft includes across Dynamics 365. This page is about the project work: choosing use cases, preparing data, building agents or custom solutions, and rolling them out under clear rules. Many clients start with the built-in features and come to us when they want something shaped to their own process.
Will our data be used to train public models?
Microsoft states that Copilot in Dynamics 365 and Azure OpenAI do not use customer data to train the underlying foundation models. Some Copilot features can involve data moving across geographic regions, and administrators control that setting. We walk through these options with you before anything is switched on.
Do we need Dataverse for Copilot Studio agents?
Not always. Agents can use SharePoint, websites, and many connectors, including Business Central. Dataverse becomes useful when the agent needs to store conversations, create records or work with Dynamics 365 Sales and Customer Service data, which already live there.
Can an agent post transactions in Business Central?
Technically yes, through connectors and APIs. We recommend that anything affecting the ledger, payments or stock goes through a human approval step first. The agent prepares the transaction and a person with the right permissions confirms it.
What if the pilot shows AI is not worth it for that task?
Then that is a useful result and we stop there. Sometimes a plain automated flow or a better report solves the problem more cheaply. We would rather tell you that than keep building something your team will not use.
Talk to us about your project.
Tell us what you run today and what has to change. A senior consultant replies with a written next step.