Email Organization Strategy
Learn how to classify and organize emails intelligently using patterns, confidence levels, and learned preferences.
Core Classification Approach
Pattern Recognition
Email classification starts with identifying domain and keyword patterns:
Receipts - Look for:
- Transaction indicators: "order", "receipt", "invoice", "confirmation", "purchase"
- Sender domains: Amazon, eBay, Stripe, PayPal, Square, etc.
- Content clues: order numbers, tracking info, totals
- Confidence: Very High - These are straightforward to identify
Example receipt email:
From: order-confirmation@amazon.com
Subject: Order Confirmation - Your purchase of [item]
Body: Thank you for your order #123-456-789
Newsletters - Look for:
- Subscription language: "newsletter", "digest", "weekly update", "monthly summary"
- Sender characteristics: bulk mailer templates, unsubscribe links
- Content: curated content, regular schedule indicators
- Quality assessment: depth of content, personalization, author credibility
- Confidence: Medium - Need user input to assess value
Example newsletter email:
From: editor@techsite.com
Subject: Weekly Tech Digest #47
Body: This week's top stories: [curated links]
[Unsubscribe]
Promotions - Look for:
- Marketing language: "limited time", "exclusive", "sale", "discount", "offer"
- Time urgency: "ends today", "24 hours only", "flash sale"
- Visual/formatting clues: CAPS LOCK, multiple exclamation marks
- Sender types: retail brands, product companies, flash sale services
- Confidence: Medium-High - Usually obvious but can be deceptive
Example promotional email:
From: marketing@store.com
Subject: LIMITED TIME: 50% OFF Everything! ⏰
Body: Hurry! Sale ends tonight!
Learning Loop: Confidence → Feedback → Preferences
Step 1: Initial Classification
- Analyze sender domain and email content
- Assign category (receipts, newsletters, promotions)
- Calculate confidence score (0-100%)
- Suggest action (archive, review, keep)
Step 2: User Review
Present classified emails with:
- Category and confidence level
- Suggested action
- Sender and subject preview
- Pattern reasoning (why this category?)
User decides:
- ✅ Approve action
- ❌ Reject (wrong category)
- 📌 Add to preferences (always archive this sender, always keep, etc.)
Step 3: Preference Updates
Store user decisions:
- keep_newsletters: Senders to preserve in inbox
- archive_newsletters: Senders to auto-archive
- keep_senders: High-trust senders that bypass classification
- blocked_domains: Spam/unwanted domains
Step 4: Continuous Improvement
Next time the classifier encounters:
- Learned senders → Apply stored preference immediately
- New senders in blocked domains → Skip inbox
- Trusted senders → Always keep (even if looks promotional)
Quality Tier Assessment for Newsletters
Not all newsletters are equal. Assess quality based on:
High Quality (Keep in inbox):
- Professional content, well-researched
- Personalization or customization
- Clear author/publication credibility
- Infrequent, valuable-per-email (weekly or less)
- User has engaged recently
Medium Quality (Review periodically):
- Decent content, some value
- Standard template, some personalization
- Established publication
- Moderate frequency (2-3x per week)
- User has engaged but may not prioritize
Low Quality (Archive):
- Generic content, minimal effort
- No personalization, template-heavy
- Frequent sends (daily or multiple daily)
- User hasn't engaged or marked as read without opening
- Looks/feels like spam but not quite
Building Effective Filters
Domain-Based Rules
Most receipts come from known e-commerce domains:
amazon.com, ebay.com, stripe.com, paypal.com,
etsy.com, shopify.com, square.com, checkout.com
Store these in blocked_domains for known spam:
discount-spam.com, promotional-blast.net, unsolicited-offers.org
Sender Pattern Rules
- Newsletters: Often from
newsletter@, digest@, noreply@ addresses
- Receipts: Often from
order@, confirmation@, billing@ addresses
- Promotions: Often from
marketing@, sales@, deals@ addresses
Keyword Rules
Build patterns incrementally as you review emails:
- Add new receipt keywords: "shipment", "tracking", "delivery"
- Add new promotion keywords: "hurry", "while supplies last", "exclusive access"
- Refine newsletter patterns: "curated for you", "top picks", "trending"
Safety-First Confirm-Before-Archive Strategy
Always start conservative:
-
Classify with confidence scoring
- High confidence (>90%): Can auto-archive with user opt-in
- Medium confidence (70-90%): Require review before archiving
- Low confidence (<70%): Highlight for user decision
-
Queue all actions for review
- Show user summary: "Found 12 receipts, 8 newsletters, 5 promotions"
- Per-email confirmation for first few of each type
- After learning, user can opt into auto-archive for high-confidence
-
User override always available
- "Keep in inbox" - adds sender to keep_senders
- "This isn't a receipt" - removes from suggested action
- "Always archive from this sender" - adds to archive list
Real-World Example: Newsletter Triage
Scenario: User has 47 unread newsletters
-
Classify all 47
- 15 high-quality tech newsletters → tier: high
- 20 medium-quality product updates → tier: medium
- 12 low-quality promotional newsletters → tier: low
-
Show user results
Found 47 newsletters:
- 15 High Quality (suggest: keep in inbox)
- 20 Medium Quality (suggest: archive, review 1x/week)
- 12 Low Quality (suggest: archive)
-
User makes decisions
- Approves archiving low-quality
- Keeps 3 high-quality tech newsletters as important
- Rejects 2 ("Those aren't newsletters!")
- Adds "techblog@news.com" to keep_newsletters
-
System learns
- Archives 12 low-quality immediately
- Marks 3 high-quality as priority
- Removes 2 rejected from consideration
- Remembers "techblog@news.com" for future
-
Next run
- Finds "techblog@news.com" in keep_newsletters → automatically keeps
- Classifies new newsletters with updated patterns
- Shows fewer options (because many are already learned)
Tips & Best Practices
✅ Do:
- Start with high-confidence categories (receipts)
- Review user's learned preferences regularly
- Use sender reputation (established senders get trust)
- Explain your reasoning to the user
- Let user incrementally tune filters
❌ Don't:
- Archive important emails without confirmation
- Ignore user feedback (it's the learning signal!)
- Over-generalize from single examples
- Use only subject lines (check body content too)
- Forget about sender domain reputation
Expanding Beyond Three Categories
In future phases, add:
- Bills/Utilities: Automatic archive, low-confidence review
- Account Notifications: Keep in inbox (security-sensitive)
- Social Media: Archive or quality-tier like newsletters
- Work: Complex rules based on project/sender
- Personal: Keep inbox, but can organize into labels
Each new category follows the same pattern:
- Define clear patterns and confidence rules
- Show to user for review and feedback
- Learn preferences over time
- Adjust confidence thresholds based on accuracy