Who this is for. Super Admins configuring an Agent Co-worker. See the Roles and Access.
Email Categories is a capability group on the AP Helpdesk tab. It holds the categories the agent uses to classify inbound mail. Classification drives your AP Helpdesk processing rules, so a category that is wrong or missing shows up later as a rule that never fires.
Navigate to Email Categories
This section appears when the autonomy level is Governed Autonomy.
On the Define capabilities step, select the AP Helpdesk tab.
In the left navigation, select Email Categories. The section opens on the Email Categories tab. A second tab, Suggested Email Categories, holds patterns the Agent Co-worker has proposed.
Figure 1. Email Categories.
How the group is arranged
The group is split into two tabs.
| Tab | Holds |
|---|---|
| Email Categories | The categories the agent classifies against today. |
| Suggested Email Categories | Categories the agent has proposed from traffic it has seen but which are not yet part of the configuration. |
The Email Categories tab holds two lists.
| List | Holds |
|---|---|
| Standard Email Categories | The categories shipped with the platform. You can turn one off, but you cannot rename or delete it. |
| Custom Email Categories | Categories you have added yourself, or agent suggestions you have accepted. |
A count sits beside the Email Categories tab when the agent has auto-added new categories from a recent mailbox triage.
How to read a category row
Each row carries several pieces of information.
Table 1. Elements of a category row
Element |
What it tells you |
|---|---|
Display name |
The name shown to analysts on tasks and email badges, for example Payment Status Inquiry. |
Type badge |
The kind of message it is. One of four: Request, Inquiry, Update, or Acknowledgement. Used for grouping in reporting. |
Key |
The identifier naming the business pattern, shown in a fixed-width font, for example |
Description |
What the category covers, and where relevant what it excludes. |
Standard or Custom badge |
Whether the category ships with the product or was added for your mailbox. |
Usage count |
How many times this category has been detected in your mailbox. A count of zero means the pattern has not arrived. |
Toggle |
Whether the category is enabled. |
Preview reply template |
Opens the composed reply for the category. |
Open email template |
Opens the response guidance for the category. |
The key, not the display name, is what a processing rule matches on. When a rule does not fire as expected, compare the rule against the key here rather than the label.
Decide which categories to enable
Sort your attention by usage count. High counts are the categories worth configuring replies for; they represent the traffic your team actually handles.
Disable categories your mailbox does not receive. Fewer active categories means fewer ways for a message to be classified, which makes behavior easier to predict.
Pay particular attention to Banking Detail Update Request. It carries high fraud risk, so confirm you have a rule that escalates these emails for manual verification. For more information, see Create and Manage AP Helpdesk Processing Rules.
Use the toggle on each row to enable or disable.
What happens when no category matches
Detection is not a filter. A message that matches no specific category is still processed: the Agent Co-worker applies a broader category and reasons from there.
Table 2. Broader categories used when no specific category matches
Category |
Applied when |
|---|---|
Request |
The sender wants something done, but not a pattern with its own category. |
Inquiry |
The sender wants information, but not a pattern with its own category. |
Update |
The message is informational, with no action requested. |
Acknowledgement |
The message confirms or thanks, with no action required. |
Unintelligible |
The message cannot be interpreted. |
Undetermined |
No determination could be reached. |
Detection scope and multiple categories
Detection runs for emails from all sender types, including Unknown senders. What the Agent Co-worker may then do with the message still depends on the sender’s authorization scope.
A single message can carry more than one category. An email that submits an invoice and queries a deduction is classified with both, and the resulting task carries work for each.
Detected categories are surfaced on the email and on any task created from it, can be edited on an individual email, and can be used to filter both email and task lists.
Changes to category configuration are recorded in an audit log.
Set what each reply says
Select the Preview reply template icon on a category row to read the reply as a vendor would receive it. See Reply preview below.
Select the Open email template icon to change how the Agent Co-worker composes the body. For more information, see Email Templates.
Icons on a category row
A standard category row carries two icons; a custom or agent-created row carries a third. The icons are unlabeled, so point at one to name it. The names differ from the actions they perform.
Figure 2. Two category rows, showing the type badge, context code, description, origin badge, usage count, toggle and the two row icons.
Table 3. Row icons
Tooltip |
Opens |
|---|---|
Preview reply template |
The Reply preview modal, the composed reply, read only. |
Open email template |
The Response guidance modal, where you edit the body guidance and format. |
Neither tooltip says “response guidance”, which is what the second one actually opens.
Reply preview
The eye icon on a category row opens Reply preview, headed with the category name. It renders the reply as the vendor would receive it, read only. A strip above the body names the category and says how the body is produced, for example that it uses your body guidance for narrative phrasing. Where the category ships no sample data, the preview says so in place of the record values rather than inventing any.
Two controls sit at the foot: Customize this template, which closes the preview and opens Response guidance for the same category, and Close.
Figure 3. Reply preview, with the sample-data note and the two controls.
Response guidance
The envelope icon opens Response guidance, headed with the category name and a pill reading Default or Customized according to whether anyone has edited it. This modal controls how the agent composes the body for that one category. Greeting, signature, closing and tone are shared across every category on the source and are set in the Defaults section instead.
Table 4. Response guidance fields
Field |
What it does |
|---|---|
Response guidance for the AI |
Free text telling the agent what to emphasize: what data to lead with, what to mention or avoid, and how long the reply should be. Up to 2,000 characters, counted live. The counter carries a reminder that data accuracy from the record context overrides any conflict with the wording you ask for. |
Body format |
Table or Plain. Table suits replies carrying several records and lets you define the columns the agent renders. Plain suits single-record or conversational replies. |
Per-record template |
Shown for Plain only. Applied once per record in the reply. Use |
Edit the guidance
Select the envelope icon on the category row.
Edit Response guidance for the AI.
Select Table or Plain under Body format.
For Plain, fill in the Per-record template or leave it empty.
Select Show preview to render the result, and Hide preview to collapse it again.
Select Save, or Cancel to discard.
Figure 4. Response guidance, with Body format set to Plain and a per-record template.
Figure 5. Response guidance with the preview expanded. The control now reads Hide preview.
Once a category has been edited its pill reads Customized and a Reset to default control appears at the foot of the modal. It does not appear while the pill reads Default.
The standard categories
Each row carries a chip showing how the category is applied: Required categories are part of the core AP Helpdesk flow, and Detected categories are matched from mail content. The count beside Active above the list is how many the agent currently holds, which is not the same as the number the platform ships: a tenant enables the subset its mailbox actually receives.
For the full catalogue, with each category’s intent type and intent context, see Standard Email Categories.
Custom Email Categories
Below the standard list, the Custom Email Categories area lists categories added for your mailbox. Each carries a Custom badge, a usage count, a toggle, and a menu. The area also offers Test classifier and + Add custom category.
For more information, see Email Categories.
Test classifier
Beside + Add custom category sits Test classifier. Its tooltip states the point of it: Try the classifier against a sample email without adding a category.
Figure 6. The Custom Email Categories header, with Test classifier and + Add custom category on the right.
For more information, see Email Categories.
Row menu on custom categories
Custom rows carry a kebab menu that standard rows do not. It holds two actions, Edit and Delete. Edit opens the category for changes to its display name, intent context and description, not the name alone. Standard categories can be disabled but not renamed or deleted, so they have no menu.
Auto-added categories banner
When a mailbox triage detects categories and adds them for you, an information banner appears above the list. It names the count and the categories, and asks you to review and confirm. For example, 3 new email categories were detected from your latest mailbox triage and auto-added to Email Categories: …. Review and confirm.
The banner has a close control. Dismissal lasts for the browser session only, so it returns on your next sign-in. It also reappears immediately if a later triage adds more categories, even in the same session.
The banner ends with Review and confirm, but there is no Confirm control. The category is already committed when it is added, so there is nothing to confirm. Read the sentence as “review this”, not as an instruction to find a button.
A category stops counting as new in one of three ways: you edit its response guidance, seven days pass since it was created, or, for the banner only, you dismiss it. Dismissing hides the banner; it does not clear the New chips on the rows themselves.
Treat it as a work item rather than a notice. Categories added this way are already in the list and already classifying mail; the banner is telling you that classification changed without your involvement.
New chip
A blue New chip sits beside the display name of any category the agent created that you have not yet tuned. It is driven by the same rule as the banner: the category came from the agent rather than from a person, its response guidance is still the default, and it was created within the last seven days. A category you typed yourself never carries the chip.
What you see when a list is empty
An empty list means different things depending on which list it is. One state is normal on a new tenant, another points at a data problem.
Table 5. Empty and loading states
Situation |
What the section shows |
|---|---|
The page is still loading |
A spinner in place of the whole list. |
No standard categories |
No standard email categories available. Unexpected. The platform ships a standard set, so this points at a data or connection problem rather than at configuration. |
No custom categories |
No custom email categories yet. Click + Add custom category above to create one. Normal on a new tenant. |
Suggestions cannot load |
Suggestions unavailable — tenant context missing. The tab cannot resolve which tenant it is reading for. Reopen the agent from the Agent Co-workers list rather than from a bookmarked URL. |
Review Suggested Email Categories
Suggested Email Categories is the second of the two tabs in the Email Categories capability group. The first tab, Email Categories, holds the Standard and Custom lists. It holds categories the agent has proposed from traffic it has seen, but which are not yet part of the configuration.
Suggested against standard
The difference is the source. One list is proposed by the agent from mail it has processed, the other ships with the platform.
| List | Holds |
|---|---|
| Suggested Email Categories | Categories the agent proposes, drawn from mail it has processed. |
| Standard Email Categories | The twenty-one categories shipped with the platform. |
A count sits beside the Email Categories heading. On a tenant where the agent has proposed nothing, the suggested list is empty.
What a suggestion carries
A suggestion carries a display name, a type (one of acknowledgement, inquiry, request, or update), the number of supporting emails from the mailbox analysis, a description, and sample subjects, alongside a Review & add button. It does not carry a key or a usage count; those appear once you add the category.
A suggestion is evidence that mail is arriving which the standard categories do not describe well. Reviewing the list periodically is how you notice a new kind of vendor request before it shows up as a run of mis-routed tasks.
Open the Suggested Email Categories tab
On the Define capabilities step, select the AP Helpdesk tab, then select Email Categories.
Select the Suggested Email Categories tab.
Figure 7. The Suggested Email Categories tab.
Where suggestions come from
The Agent Co-worker assigns provisional categories internally as it processes email, using them to detect recurring patterns. These are not acted on and are not visible while provisional.
A pattern surfaces for your review only after it crosses a volume threshold confirming it recurs. That threshold matters: it means a single unusual message never generates a proposal, and anything you are shown has happened enough times to be worth naming.
Suggestions arrive as a batch, with example emails and match counts so you can judge whether the pattern is worth naming.
Review a suggestion
Read the proposed category and the example emails behind it.
Check the match count to judge how much traffic the pattern represents.
Select Review & add to promote the suggestion to a custom category, or reject the suggestion to dismiss the pattern.
How to judge a suggestion
Three questions settle most cases:
Is it genuinely distinct? If the pattern is already covered by a standard category, accepting it creates two ways to classify the same message and makes rule behavior harder to predict.
Would you handle it differently? A named category is only useful if a rule, a reply, or a routing decision depends on it. If nothing would change, the broader category is fine.
Is the volume worth it? The match count tells you. A pattern arriving weekly justifies a category; one arriving twice a year does not.
What happens on acceptance
An accepted suggestion becomes a custom category. Its intent type and intent context are editable at that point, and the display name is generated and editable. The same validation rules apply as for a category you create yourself. Accepting without editing applies snake_case conversion and uses the generated display name.
Because intent type locks on save, this is the moment to correct it if the proposal has it wrong.
For more information, see Email Categories.
What happens on rejection
Rejection is remembered. A rejected pattern is not proposed again, so you will not be shown the same suggestion repeatedly.
Suggested categories are not active until promoted. Processing rules cannot reference a suggestion; rules bind only to configured categories.
The patterns a mailbox receives change as the vendor base and processes change, so the tab continues to produce new suggestions over time.