Processing vendor emails automatically in D365FO purchase orders with the help of AI

The combination of using AI and D365FO. It is a very logical step when mentioning Dynamics 365FO as 'AI ERP'. But what if you want a solution where no standard Copilot functionality is already available? That was a test scenario I worked on with a colleague to find out how this can work. In this article I describe our first experiences, which made me even more enthusiastic about what AI can do, also in combination with D365FO.

Our test scenario was the following: we see a lot of organizations sending D365FO purchase orders confirmations via e-mail to their vendors. Their vendors responses to it via e-mail and the response is processed manually in D365FO to update the purchase order information based on the vendor response. What if we use AI to see if we can update the purchase order in D365FO automatically without manual work?

In this article I show you the current experiences, in my opinion already impressive. For this test we used the Microsoft Contoso Environment demo data.

Scenario 1

At first we made a purchase order where we purchased three items in different quantity and prices:

Based on the purchase order confirmation we sent an email to the vendor where we confirm the purchase. I know this is not the standard confirmation template so far, but we wanted to start with analyzing e-mails so via a Power Automate flow we sent this confirmation out to the vendor via e-mail:

Playing the vendor, I responded to that e-mail. In the e-mail I confirm some purchase information as requested, but also I describe some changes from my vendor side in relation to the original purchase order confirmation received:

So the changes needs to be:

• Item 1000 price change to USD 900,-.
• Item A0001 quantity change from 1
• Item A0002 quantity change to 12 and a price change to USD 18,-

When the e-mail is received, Power Automate automatically picks up this received response from the vendor. But this email is unstructured, so this can't be imported in D365FO already. And that's where we use AI to structure this e-mail into a JSON that can be imported. In our AI Parameters we have added AIResources OpenAI and Google Gemini:

In the AIParameters we choose Google Gemini as the proposed AI model. But to be able to switch to OpenAI in the Power Automate flow that picks up the response emails we have a step where both AI models can be used to structure the e-mail from the vendor into a JSON that can be used to import in D365FO:

So based on the received e-mail, the started Power Automate flow will request AI to read the vendor e-mail and convert it into a JSON we can import in D365FO. When the Power Automate flow is finished, we go back to the purchase order in D365FO to check the result. Here we see the following purchase order line information now:

On more time the changes described in the e-mail:

• Item 1000 price change to USD 900,-.
• Item A0001 quantity change from 1
• Item A0002 quantity change to 12 and a price change to USD 18,-

All of the changes described in the mail are converted correctly by AI into the JSON which is imported in D365FO to update the purchase order. So without any manual intervention, the purchase order in D365FO is automatically correct updated with the help of Power Automate using an AI model to convert the mail into the needed JSON.

Scenario 2:

In the second scenario we use the same Power Automate flow, but now the response of the vendor is in a different language. Where sent the purchase order confirmation in English, the response mail of the vendor is in Dutch:

Using the same Power Automate flow we are interested what happens now. Will the AI model understand the change and convert the e-mail to the JSON correctly? To get the answer, first in English the changes described in the Dutch email above:

• Item 1000 quantity will go to 6
• Item A0001 will be delivered as requested (quantity 2 for USD 12,- each)
• Item A0002 quantity will go to 10, price will go to USD 20,- each

When the Power Automate flow is finished, the purchase order is updated as following:

So also with the e-mail from the vendor written in Dutch language as response to the English purchase order confirmation e-mail, the AI model converts the e-mail correctly into the right JSON that we need to update the purchase order in D365FO. Again the purchase order was updated automatically via the Power Automate flow without the need of manual actions.

Conclusion

At first, I know that we didn't test all scenarios already. And I also know that organizations maybe want to have a check before the proposed changes in the vendor e-mail will be used to update the purchase order in D365FO. But it is a start with a very positive outcome in relation to test the AI possibilities in optimizations. We will build further on these experiences. My conclusion is that the AI models can interpret the needed information very well to give Power Automate the needed structured JSON information to update D365FO.

Also one of my conclusions is that when no standard Copilot functionalities are (already) available for a specific process in relation to D365FO, with the help of Power Automate and AI models you can automate a lot already to optimize processes. Especially when there is need for conversion from unstructured to structured data, AI can really be a game changer.