Create and Manage AP Helpdesk Processing Rules

Who this is for. Super Admins configuring an Agent Co-worker. See the Roles and Access.

Processing Rules is a capability group on the AP Helpdesk tab under Governed Autonomy. It does not appear in Collaborative Mode. It decides what the agent does for each combination of intent and sender type. The product heads it Sender-aware processing rules.

Navigate to Processing Rules

1. On the Define capabilities step, select the AP Helpdesk tab.

2. In the left navigation, select Processing Rules. The Sender-aware processing rules page opens.

Figure 1. The Processing Rules group, with the Sender-aware processing rules heading, the rule list and the Show rule metadata toggle.

Starter rules for a new agent

A new agent has no rules. The group reads “No rules configured yet. Load starter rules to get started” and offers Load Starter Rules, which populates a working set you can then edit. Below it, the rules list shows “No rules configured. Add one below.” with + ADD RULE, and a Show rule metadata switch at the top right of the list.

What a rule looks like

Rules are not built from dropdowns. Each one is a plain-English instruction that you type, and the group holds an ordered list of them.

Figure 2. The rule list, with a toggle and an actions menu on each row.

Written rules cover most of what the agent does. Typical examples:

• Always escalate bank detail changes for manual verification.

• Escalate PO discrepancies over $1,000 to the AP manager.

• Do not create email drafts for the Invoice Submission email category.

• When the sender type is System Sender, assign the task to a named analyst and do not create a draft.

• Assign all Inquiry and contact-update tasks to a named analyst.

• For all Update tasks, set the priority to Medium.

• Where an email carries multiple intents, assign it to a named analyst and set priority High.

• Forward all email in the PO Info category to a named address.

A rule can name an intent, a sender type, an email category, a phrase in the subject or body, an amount threshold, a person, or a mailbox, and can combine them in one sentence.

What a rule can direct

For each combination of email category and sender type, a rule can direct the Agent Co-worker to add a label, draft a reply for analyst review, send a reply autonomously, transfer attachments, write back to the system of record, create a task, assign it to someone, or set its priority.

Rules from the Intelligence Layer run alongside yours and apply to every sender regardless of type.

Prepare versus execute

The group description states: “Use the prepare/send and prepare/write distinctions to control whether the agent drafts work for review or executes it autonomously.”

Table 1. The prepare and execute pairings

Actions run in order. The order of actions within a rule is part of its behavior.

Add a rule

1. Open the Processing Rules group.

2. Select + ADD RULE.

3. Write the instruction as a sentence, naming the condition and what the agent should do.

Show rule metadata, at the top right of the rules list, adds an expand control to each rule row. Expanding a row shows the directive compiled from your wording, under Trigger, Action and Scope.

What happens when you add a rule

A candidate rule is evaluated across four dimensions before it takes effect.

Table 2. Rule evaluation dimensions

These evaluations are performed by the model rather than by fixed pattern matching, so they reflect judgment rather than certainty.

What the safety screening blocks

Each category names a way a rule could overreach, from elevating access for lower-trust senders to creating unbounded automation.

Table 3. Safety and compliance violation categories

Adding a restriction is never a violation. A rule that requires approval or narrows permissions passes screening even when it contradicts a more permissive existing rule. That is a conflict, not a violation. Screening targets the removal of controls, not the automation of controlled work.

Rules must be specific to AP operations and AP Helpdesk email processing. Vague or overly broad rules are flagged.

What no rule can authorize

These hold regardless of scope, sender type, or business justification:

• Disclosing data the sender is not authorized to access, including another vendor’s data and banking details in replies.

• Executing a payment or changing a payment amount without an Approve action.

• Treating an Unknown sender as External, Internal, or System. For more information, see Contact Management.

• Forwarding email content or attachments outside the sender’s authorized scope.

• Disabling the safety screening itself.

Handle a conflict warning

A conflict means two rules appear to contradict. That is sometimes intentional: a general rule with a narrower exception is a normal pattern. The system flags it and you choose whether to add the rule anyway.

The conflict description names what contradicts what. Where the contradiction was not intended, revising the wording is better than overriding.

Manage existing rules

1. Use the toggle on a rule row to enable or disable it. Disabled rules stay in the list with muted styling, are excluded from agent processing, and take no part in conflict evaluation for new rules.

2. Use the ⋮ menu at the end of a rule row to Edit or Delete that rule. Edit reopens the rule text on the row, where the check mark saves the change and the cross discards it. Delete appears on your own personal rules; organization and team rules can be edited and turned off from the row, but not deleted there. Use Show rule metadata, at the top right of the list, and then expand a row to read that rule’s directive.

Editing a rule’s text triggers full re-evaluation and regenerates its directive and context. Editing only the context does not. Enabling or disabling does not.

Deleting a rule is permanent and cannot be undone. Disabling is reversible and requires no re-evaluation on re-enabling.

Turn a rule off without deleting it

Every rule carries its own toggle at the left of its row and a ⋮ menu at the right. The toggle turns the rule on and off, and the ⋮ menu holds Edit, with Delete on your own personal rules. Switching a rule off grays its text and takes it out of play while leaving it in the list, so a rule can be parked and brought back without being rewritten. A list read at a glance is misleading for this reason: a rule can sit in the list and be out of play.

Check the toggles before concluding that a behavior is configured. A rule that reads exactly like the behavior you want may be sitting switched off.

How rules are applied at runtime

Before applying any rule, the Agent Co-worker checks its triggering conditions against the sender metadata: type, status, match type, and matched record. A rule with a sender-type condition applies only when the metadata matches exactly. Rules whose conditions are not met are excluded entirely. They do not influence reasoning, task scope, or reply composition.

List position is not a priority order. Where rules overlap, the more specific rule wins: narrower conditions, an explicit sender type, or a named vendor takes precedence over a general rule.

Write rules that behave predictably

• Keep each rule to one trigger and one action. Split a rule with independent clauses and different sender scopes into separate rules.

• State the trigger, the action, and the scope. Leave out preamble and repeated qualification.

• Name the sender type when a rule should apply to only one.

• Write a narrower rule for an exception rather than complicating the general rule.

• Reference email categories by the name shown in the Email Categories section. The agent resolves the stored key first and then the name, so either form matches. For more information, see Standard Email Categories.

Preview an Agent Co-worker reply

This article describes previewing a reply before the agent uses it.

Both models offer a reply preview, in different places. Under Collaborative Mode, the Agent Replies group previews the stored email template. Under Governed Autonomy, open the Email Categories group and select the Preview reply template icon on a category row to open the Reply preview dialog for that category. The Preview reply template icon is present on every category, but it is inert on a category whose toggle is off, rendered and grayed rather than hidden. If the icon does not respond, check the toggle before looking for a permission problem.

Why the two previews differ

Collaborative Mode replies come from stored templates, so the preview shows the template you edited. Governed Autonomy composes each reply from the message and the records, so the Email Categories preview shows the effective template for that category, its greeting, tone, body guidance, closing, and signature, rather than the finished reply. To see the exact wording that will go out, open the drafted reply on the task.

Where to check a reply

Table 4. Where to preview a reply, by autonomy model

Open the reply preview

1. On the Define capabilities step, select the AP Helpdesk tab, then select Email Categories.

2. On the category row, select the Preview reply template icon. The Reply preview modal opens, titled with the category name.

Figure 3. The Reply preview dialog for an email category, showing greeting, body, closing and signature.

What the preview shows

The preview banner names the category and notes that the preview uses your body guidance for narrative phrasing. Below it, the rendered reply shows:

• The greeting.

• The body, composed from your response guidance.

• The closing line offering further questions.

• The sign-off and signature block, including the logo.

Greeting, closing, and signature come from the shared defaults rather than from this category’s guidance, so what you are checking here is the body.

What to look for

• Does it answer the question? A reply that is accurate but does not address the ask will generate a follow-up email, which defeats the purpose.

• Is the tone right for the category? A banking detail request and a routine payment-status answer should not read the same way.

• How does missing data read? Where a value is not on file, the reply says so. Whether that phrasing sits acceptably depends on the field and the surrounding wording.

• Is anything overstated? Guidance asking for more certainty than the records support produces replies that commit you to something you cannot stand behind.

Edit the guidance from the preview

1. Select CUSTOMIZE THIS TEMPLATE to open the response guidance for this category.

2. Select CLOSE to dismiss the preview without changes.

Preview from the response guidance modal

A preview is also available inside the response guidance modal through SHOW PREVIEW, which shows the effect of a wording change before saving. The two differ only in where they are opened from.

The preview reflects structure and wording. Actual record values resolve at send time from the system of record, within the sender’s authorization scope.

Under Collaborative Mode, replies are literal templates rather than composed from guidance, and are previewed differently. For more information, see AP Helpdesk Agent Replies.

What the shipped defaults look like

On an unconfigured mailbox the preview renders the defaults, so you can see what a vendor receives before you change anything.

Table 5. Default reply furniture

“Dear Sir/Madam” is a dated and gendered opening for a vendor-facing mailbox. It is a default rather than a recommendation. Change it in the shared defaults before the agent sends anything.

On a default configuration the preview shows Best regards, twice, once as the closing and again above the signature block. A doubled sign-off in the preview comes from the shipped defaults. Setting your own closing and signature removes it.

When the category ships no sample data

Where a category has no sample records the preview substitutes an italic line in place of rendered values, noting that no sample data is shipped for this email category and that the AI would compose a narrative reply based on the body guidance.

The greeting, closing and signature still render, so the structure is checkable. Only the body is standing in.

Related information

• Email Categories in the Agent Configuration

• Rulebook

• Contact Management Overview

• Review and Activate