Importing Conversations

Import your HelpScout conversations to analyze customer feedback. Here's how importing works.

Starting an import

From the mailbox page

1. Navigate to your mailbox in Jetson
2. Click Import
3. Choose your date range
4. Click Start Import

Choosing a date range

Start small to verify everything works, then import more history.

What happens during import

Fetching

Jetson retrieves conversations from HelpScout:

• Batches of conversations are fetched
• Progress is shown in real-time
• Duplicate conversations are skipped

Processing

Each conversation is analyzed:

1. Thread detection - Split multi-issue conversations
2. Classification - Determine category and priority
3. Embedding - Generate semantic representation
4. Duplicate check - Compare to existing items
5. Work item creation - Create bugs, features, or articles

Completion

When import finishes:

• New work items appear in your dashboard
• Patterns are updated
• Metrics are recalculated

Import progress

Real-time tracking

The import page shows:

• Conversations fetched
• Conversations processed
• Work items created
• Estimated time remaining

Background processing

You can leave the page during import:

• Processing continues in the background
• Return anytime to check progress
• You'll see results when complete

Multi-issue detection

What it is

Some conversations cover multiple topics:

• Customer reports a bug, then asks a feature question
• Long thread spanning weeks with topic changes
• Multiple distinct issues in one email

How it works

Jetson detects:

• Conversations spanning 7+ days
• Clear topic pivots
• Customer explicitly changing subjects

Each issue becomes a separate work item.

When splitting occurs

Splits happen when:

• Topic A is resolved, then Topic B is introduced
• Customer says "different question" or "switching topics"
• Significant time gap with new subject

Splits don't happen when:

• Follow-up questions about same issue
• Multiple aspects of one problem
• Iterative troubleshooting

Handling duplicates

Automatic deduplication

Before creating a work item, Jetson checks:

• Semantic similarity to existing items
• Exact match detection
• Product-specific context

What happens with duplicates

When a duplicate is detected:

• No new work item is created
• Customer count on existing item increases
• Conversation is linked to existing item

Adjusting sensitivity

If you see too many duplicates or too few:

1. Go to Settings > Products
2. Adjust Threshold Settings
3. Re-import affected conversations

Webhook vs. import

Webhooks (real-time)

After connecting HelpScout, webhooks handle new conversations:

• Triggered within seconds of new messages
• Fully automatic
• No manual action needed

Import (historical)

Import handles past conversations:

• Manual trigger
• Date range selection
• One-time operation per range

When to use each

Troubleshooting imports

Import stuck

If progress stops:

1. Wait a few minutes (large batches take time)
2. Refresh the page
3. Check HelpScout connection in Settings
4. Try a smaller date range

Missing conversations

If expected conversations aren't appearing:

1. Verify they're in the selected date range
2. Check the mailbox filter
3. Confirm HelpScout permissions
4. Look in "No Action" category (may be filtered)

Duplicate items appearing

If you see duplicate work items:

1. Check threshold settings (may be too low)
2. Review similar items manually
3. Merge or resolve duplicates

Classification errors

If classifications seem wrong:

1. Manually correct a few examples
2. Jetson learns from corrections
3. Re-import after corrections accumulate

Re-importing

When to re-import

• After changing classification settings
• To pick up missed conversations
• After correcting classifications
• When expanding date range

What happens

• New conversations are added
• Existing conversations are updated
• Duplicates are handled automatically
• Work items are regenerated if needed

Data safety

Re-importing doesn't delete existing data:

• Manual edits are preserved
• Resolved status is maintained
• GitHub links remain intact