Who this is for. Super Admins. AP Helpdesk Managers, Specialists and Analysts cannot open AP Intelligence. Requires AP Intelligence to be enabled for your tenant. See the Roles and Access.
An email category is the agent’s answer to “what is this message about?”. A message can carry more than one, each scored separately, and together they drive everything the agent does next: which reply template the agent reaches for, which rules apply, and whether it can act on its own or has to raise a task. A task keeps at least one category, so the last one cannot be removed.
So this screen matters more than its size suggests. Get a category wrong and the agent will answer confidently with the wrong thing. Leave one switched off and mail of that kind falls through to a person every time.
The catalogue comes from three sources: standard categories shipped with the platform, categories the agent proposes after reading your mail, and categories you write yourself.
Figure 1. The Email Categories screen.
Open Email Categories
Open AP Intelligence.
Select Email Categories in the left rail.
Two sub-tabs sit above the list: Email Categories, which is the live catalogue, and Suggested Email Categories, which is what the agent thinks you are missing. The number on the first sub-tab counts categories added since you last looked.
Read a category row
Every row carries the same six things.
Figure 2. A category row: name, intent type, description, badge and controls.
Table 1. Parts of a row.
Part |
What it tells you |
|---|---|
Name |
What the category is called. This is the name that appears on a message in the Mailbox. |
Intent type |
A chip below the name: Acknowledgement, Request, Update or Inquiry. It groups the category by the shape of the message rather than its subject. |
Description |
One line telling the classifier how this category differs from the others. This is not decoration: it is what the agent reads when deciding between two similar categories. |
Standard badge |
The category shipped with the platform. Categories you add yourself sit in their own section further down the list, and each of those rows carries a menu at the end holding Edit and Delete. A standard category has no menu: it can be turned off, but not renamed or deleted. |
Toggle |
Whether the category is in use. It sits at the right-hand end of the row, before the two icons. See Turn a category on or off, below. |
Eye and envelope icons |
Preview reply template shows the reply this category produces. Open email template takes you to that template to edit it. Both sit at the end of the row, after the toggle. |
Turn a category on or off
The toggle at the right-hand end of each row decides whether the classifier may use that category at all. Most arrive on.
Turning one off does not delete it and does not stop that mail arriving. It removes the category from the classifier’s choices, so messages that would have matched it are sorted into whatever fits next, or into none.
That second outcome is the one to plan for. Switching off a category without a sensible neighbour for its mail to fall into produces unclassified messages, and unclassified messages go to a person.
When the agent proposes a category
The agent adds categories on its own. When it does, a banner appears above the list naming what it added and asking you to review and confirm it.
Figure 3. The banner the agent raises when it adds a category itself.
The category is already live at that point, so the banner is telling you what has happened rather than asking permission. Read the new category, check its description reads as you would want, and dismiss the banner with the cross.
Suggested email categories
The second sub-tab is the agent’s reading of what your mailbox contains that your catalogue does not cover.
Figure 4. Suggested Email Categories.
Select Get Suggestions to run the analysis. It reads up to a thousand business emails that did not match a category and proposes new ones from the patterns it finds.
Each suggestion carries a confidence score, a plain explanation of how many message threads shared the pattern, and a list of example emails so you can judge it yourself. The list is ordered by signal strength, and a timestamp above it records when the analysis last ran.
Read the examples before accepting anything. A suggestion scored on twenty-seven threads that turn out to be one automated sender is not a category you want.
Add a custom category
Where neither the standard set nor the suggestions cover something, write your own.
Select + Add custom category. A panel opens below the list.
Choose a TYPE. This is the intent type the category belongs to.
Enter a CONTEXT, in the lower-case underscored form the field suggests, such as
payment_status.Leave DISPLAY NAME to fill itself. It is generated from the type and the context.
Write a DESCRIPTION. It is required, capped at 1000 characters, and the panel says why: it is what lets the agent tell this category apart from similar ones.
Select Add category. It stays grayed until every required field is filled.
Figure 5. The panel for adding a custom category.
Spend the effort on the description rather than the name. The name is for you; the description is what the classifier actually reads.
Test the classifier
Before relying on a new or changed category, put an email through the classifier and watch what it does.
Select Test classifier.
Enter a FROM address.
Choose a SENDER type: External Sender, Internal Sender, System Sender or Unknown Sender. The same message can classify differently depending on who sent it, so pick the one you are actually worried about.
Enter a SUBJECT and paste the body.
Select Classify email. It stays disabled until the message has enough to work with.
Figure 6. The classifier tester.
The tester reports which category won, how confident the agent was, and why. Close it with Close tester.
The pair of controls to use together is this one and the description field. If a test picks the wrong category, the fix is almost always a sharper description on one of the two categories that competed, not a new category.