Auto-populate a custom field with Sidekick

When a customer mentions an order number in their email, Sidekick can extract it and write it into the conversation’s order_number custom field automatically. The agent never has to copy-paste from the message body.

This is one of the highest-leverage agent-assist features in Atender — it removes a small friction that would otherwise happen on every single conversation about an order.

Before you start

• The destination custom field exists (create one first). For order numbers, this is typically a Text field on Conversations.
• Admin permissions on Sidekick settings
• Example messages where the data appears, so you can confirm the detection pattern matches

Steps

1. Open Settings → Sidekick → Detection rules.
2. Click New detection rule.
3. Name the rule descriptively (e.g. Extract order number from message body).
4. Set the trigger: when Sidekick should run this rule. Most extraction rules fire on Message received.
5. Set the target: where the extracted value goes.
   Pick the destination entity (conversation, contact, or case).
   Pick the destination custom field (e.g. order_number).
6. Define the detection pattern. Two ways:
   Regex pattern — for structured identifiers like order numbers, account IDs, license keys. Example: \bORD-\d{6,}\b matches ORD-123456 and similar.
   Natural-language description — for fuzzier extraction where the customer might phrase it differently each time. The AI uses the description to find the value semantically.
7. Optionally, set conditions that scope when the rule fires (only on certain channels, only when no value is set yet, etc.).
8. Save and enable the rule.

Verify it worked

1. Send a test inbound message to one of your inboxes that contains the pattern (e.g. an email mentioning Hi, I have a question about order ORD-987654).
2. Open the resulting conversation.
3. Within a few seconds, the order_number field should populate with ORD-987654.
4. Check Sidekick’s activity log for the conversation — it should show the detection rule fired and the value it extracted.

Patterns that work well

• Order number with consistent prefix — Regex \bORD-\d{6,}\b — Deterministic, fast, no AI cost
• Customer email mentioned in body (different from sender) — Natural-language description — Customers describe this differently each time
• Account ID — Regex \bACC-[A-Z0-9]+\b — Structured
• License key — Regex matching the license format — Structured
• Affected product — Natural-language description with examples — Customers refer to products by varied names

For regex-friendly patterns, prefer regex — it’s deterministic, costs nothing per run, and is easy to debug. For free-form mentions, use natural-language descriptions and accept some variability.

Don’t overwrite manually-set values

By default, set the rule to run only when the destination field is empty. This protects values an agent set deliberately from being overwritten by the AI on a later message. The behavior is configurable per rule.

Troubleshooting

• Symptom: Rule doesn’t fire on test messages. Fix: Open Sidekick’s per-conversation activity. If there’s no entry for your rule, check that the rule is enabled and that the trigger conditions match the inbound channel. If there’s an entry but it shows “no match,” your regex / description doesn’t match the message content. Adjust the pattern.
• Symptom: Rule extracts the wrong substring. Fix: Tighten the regex with word boundaries (\b) or anchors. For natural-language rules, add more example phrasings to the description.
• Symptom: Rule overwrites a value an agent set. Fix: Configure the rule to skip when the field already has a value.

See also

• What is Sidekick?
• Use a custom field in an automation — where extracted values become routing decisions