Automatically capture contacts from email in Dynamics 365 Sales
Yarin Kuper - Research Intern, Dynamics 365 Sales
Email messages are a great source of contact information for salespeople, but having to manually create contacts from messages takes a lot of time. To help you spend more of your time selling, the premium auto capture capability in Dynamics 365 Sales uses advanced AI to mine details from email messages, such as the To and From fields, body text, and signature, to automatically create contacts for you.
Let’s review two examples where premium auto capture uses AI to capture data directly from email text and convert it to contacts.
Automatic email signature detection
Consider the following example. Instead of manually copying and pasting the various elements into a new contact, our machine learning model detects the signature in the email and automatically imports the information. The model was developed from a large data set of examples of email messages, in various shapes and forms.
We represent each line in an example email as a vector, which is a series of numbers. This vector is the fingerprint of a signature line versus non-signature line, and contains numbers that represent uppercase letters, whether or not a closing like “Thanks” or “Regards” appears, what portion of the line contain names, and more. AI learns from the vectors which line is part of a signature. We then run another algorithm to determine the signature block. Eventually, we extract the signature entities: name, organization, phone number, email address, job title, and address. In the example above, the block starts with Regards and ends with firstname.lastname@example.org.
After detecting a signature, the contact will automatically be suggested for you to add, saving you valuable time and costly errors. This is how it looks in the product:
Automatic contact mention detection
In addition to the signature, the email body is also a great source for new contact information, but it requires the challenging task of matching different entities. For example matching a name with an email address. In the example above, we have a mention of Maria, the COO. We identify that this is a name, and then match it with email@example.com. The matching need not be exact; our AI can guess from the email address that it corresponds to a certain name.