Escalations and handing conversations back and forth with Arbyn covers what happens once a conversation escalates. This article is about the rules themselves, the builder where you see the eight rules every store starts with and where you can add as many of your own as you like.
What it does
Every store starts with eight built-in rules, each covering a genuinely serious situation (a legal threat, a safety concern, a data-privacy request, and similar), and each one always wins: Arbyn still drafts a reply the same way it would for any message, but the moment one of these rules matches, that draft is discarded and the conversation goes straight to your team instead. On top of those, you can build unlimited custom rules of your own, using a keyword or phrase, a customer's detected intent, negative sentiment over several messages, an order's value, how old an order is, an attempted return on a final-sale item, or your own pattern, so a conversation escalates whenever your own specific situations come up too.
Where to find it
Settings → Escalation Rules.
Setup (step by step)
Open Settings → Escalation Rules. The eight built-in rules are listed first, each shown as Built-in and already active.
To add your own, click New rule, then give it a name.
Choose a Trigger type, covered in detail below, and fill in the fields that appear for that type.
Choose a Priority tier: Urgent, High, Medium, or Low. When more than one rule could match the same conversation, the highest tier wins and fires first.
Toggle it Active and click Save. It applies to every new message from that point on.
Any custom rule can be edited or deleted later. A built-in rule can be turned off, or edited if you want to tune it (its keyword list, for instance), but it can't be deleted outright.
How it works
The eight built-in rules are the backstop, and they are never optional. They cover a mention of suicide or self-harm, a direct threat of legal action, an allergic reaction or injury from a product, a discrimination or harassment accusation, a mention of a data breach, a chargeback already filed, a PayPal claim already filed, and a GDPR or CCPA data-deletion request. Every one of these fires the instant a message matches and hands the conversation straight to your team. Arbyn still generates a reply behind the scenes the same as it always does, that draft is just thrown away rather than sent, so the customer only ever sees your team's response, never a discarded AI draft. You can turn any of them off individually if you genuinely don't want that behavior, and you can edit their underlying keyword list, but they ship on for every store because each one covers something where a wrong autonomous reply carries real legal, safety, or financial risk. A wider set of softer situations (an angry customer, a lost package, a wholesale inquiry, and similar) isn't handled by a rigid keyword rule at all; Arbyn reasons about those with judgment as part of its normal conversation handling instead of matching them against a fixed phrase list.
Trigger types available for your own rules:
Keyword. A list of words or phrases (one per line, or comma-separated). Choose whether any one of them is enough to match, or whether the conversation needs to contain all of them.
Intent. Matches when Arbyn detects the conversation is genuinely about a specific kind of request (a return, a cancellation, a complaint, and so on) rather than just mentioning a word.
Sentiment. Matches when a customer's tone reads as negative, or very negative, across a minimum number of messages on the same unresolved topic, so a single frustrated sentence doesn't trigger it but a customer who stays upset over several exchanges does.
Order value. Matches when a refund amount or an order's total crosses a dollar threshold you set, using greater-than, greater-than-or-equal, or less-than as the comparison.
Days since order. Matches when an order is older than a number of days you choose, useful for routing anything past your usual return or warranty window straight to a person.
Final-sale attempt. Matches whenever a customer tries to return or exchange an item you've marked final-sale, no extra configuration needed.
Custom pattern. A regular expression for anything the simpler types can't express. Arbyn checks that the pattern is valid before saving it, so a typo never silently saves a rule that matches nothing.
Priority decides which rule wins when more than one could match. Rules are checked from highest priority to lowest, and the first one that matches is the one that fires, so a Low-tier custom rule never overrides one of the eight built-in Urgent-tier rules, and two of your own rules at different tiers resolve predictably rather than by coincidence of list order.
No rule at all matching is the normal, expected outcome for most conversations. Arbyn doesn't need a matching rule to handle a conversation well; the rules exist specifically to catch the situations you've decided always need a person, not to gate every reply.
Good to know
The eight built-in rules can be disabled individually, but never deleted. They're the one layer of escalation logic every store keeps by default, precisely because each one covers a serious risk if the AI got it wrong.
A rule match discards the AI's draft rather than sending it. Arbyn still generates a reply behind the scenes; it isn't a low-confidence pause for your review, it's an immediate handoff, and the draft that was generated is thrown away rather than shown to the customer.
You can add as many custom rules as you like, on any of the seven trigger types, in addition to the eight built-in ones.
Priority tiers resolve conflicts predictably. The highest-tier matching rule always fires first when more than one rule's condition is met on the same conversation.
A custom pattern rule is validated before it saves. An invalid regular expression is rejected outright rather than silently saved as a rule that can never match.
A Health column shows whether a rule is actually catching anything. Once a rule has existed for 14 days, this page shows Healthy (it's matched recently), Fires but never matches (it keeps getting checked but nothing trips it, which usually means a typo in a keyword or a pattern that stopped matching after a copy change), or Has not fired (no data at all). New rules under 14 days old show Healthy by default since there isn't enough data yet to judge them, and this applies to the eight built-in rules too, not only your own.
FAQ
Can I turn off one of the eight built-in escalation rules? Yes, disable it individually from the list. It stays visible so you can re-enable it later, but it can never be deleted outright.
What's the difference between a keyword rule and an intent rule? A keyword rule matches specific words or phrases appearing anywhere in the message. An intent rule matches Arbyn's actual understanding of what the customer is asking for, so it can catch a request phrased many different ways without you having to list every possible wording.
What happens if two of my rules could both match the same conversation? The one with the higher priority tier fires. Rules are always evaluated from Urgent down to Low, and the first match wins.
Does a rule matching mean Arbyn definitely didn't draft anything? No, Arbyn still drafts a reply behind the scenes the same way it does for any message. A rule match is a full escalation, not a low-confidence pause, so that draft is discarded and never shown to the customer, who only ever sees your team's response.