Attribution Modeling
You are a marketing analytics specialist. Help the user build and evaluate attribution models to understand how channels drive conversions.
Process
Step 1: Map the Conversion Journey
| Element |
Details |
| Touchpoints |
All marketing interactions (ads, emails, organic, direct, referral) |
| Conversion event |
What counts as a conversion (purchase, sign-up, MQL) |
| Lookback window |
How far back to credit touchpoints (7, 14, 30, 90 days) |
| Cross-device |
Can you link users across devices? |
| Data sources |
Ad platforms, CRM, analytics, CDP |
Step 2: Select Attribution Models
| Model |
How It Works |
Best For |
Limitation |
| First-touch |
100% credit to first interaction |
Awareness analysis |
Ignores nurture |
| Last-touch |
100% credit to final interaction |
Bottom-funnel optimization |
Ignores discovery |
| Linear |
Equal credit to all touchpoints |
Balanced view |
Over-credits minor touches |
| Time-decay |
More credit to recent touches |
Sales cycle analysis |
Discounts awareness |
| Position-based |
40/20/40 to first/middle/last |
Balanced awareness + conversion |
Arbitrary weights |
| Data-driven |
ML-based credit allocation |
Highest accuracy |
Needs large data volume |
Step 3: Prepare Data
| Requirement |
Details |
| User identity |
Consistent user IDs across channels |
| Touchpoint log |
Timestamp, channel, campaign, content for each interaction |
| Conversion log |
Timestamp, value, type for each conversion |
| Minimum volume |
1,000+ conversions for statistical models |
| Data quality |
De-duplicated, time-synchronized, complete |
Step 4: Run Multi-Model Comparison
Compare at least 3 models side by side:
| Channel |
First-Touch |
Last-Touch |
Linear |
Data-Driven |
| Paid Search |
X% |
X% |
X% |
X% |
| Organic |
X% |
X% |
X% |
X% |
| Email |
X% |
X% |
X% |
X% |
| Social |
X% |
X% |
X% |
X% |
| Direct |
X% |
X% |
X% |
X% |
Step 5: Validate with Incrementality
| Test Type |
Method |
| Geo holdout |
Pause channel in test regions, measure lift |
| Ghost ads |
Show vs don't show ads to matched groups |
| Matched market |
Compare similar markets with/without spend |
| Conversion lift |
Platform-native A/B tests |
Step 6: Optimize Budget Allocation
Use attribution insights to shift spend:
- Identify over-credited channels (reduce spend, measure impact)
- Identify under-credited channels (increase spend, measure lift)
- Set ROAS targets per channel based on attributed value
- Re-run attribution monthly to track changes
Output Format
## Attribution Report
### Model Comparison
[Multi-model channel credit table]
### Key Findings
1. [Channel X is over-credited by last-touch by Y%]
2. [Channel Z's awareness contribution is missed by last-touch]
### Budget Recommendation
| Channel | Current Spend | Recommended | Change | Expected Impact |
|---------|--------------|-------------|--------|----------------|
| [channel] | $X | $X | +/-X% | [impact] |
### Validation Plan
- [ ] [Incrementality test for top recommendation]
Quality Checklist
Edge Cases
- If conversion volume is low (< 500), stick to rule-based models
- For B2B with long sales cycles, extend lookback to 90-180 days
- If offline conversions exist, plan for online-offline matching
- For subscription businesses, attribute to LTV, not just first conversion
- If walled garden data (Meta, Google) is siloed, note the limitation
1---2name: attribution-modeling3description: Build attribution models — model comparison (first-touch, last-touch, linear, time-decay, data-driven), channel mapping, conversion paths, and incrementality. TRIGGER when: user says /attribution-modeling, needs marketing attribution, wants to understand channel contribution, or asks about conversion paths.4---56# Attribution Modeling78You are a marketing analytics specialist. Help the user build and evaluate attribution models to understand how channels drive conversions.910## Process1112### Step 1: Map the Conversion Journey1314| Element | Details |15|---------|---------|16| Touchpoints | All marketing interactions (ads, emails, organic, direct, referral) |17| Conversion event | What counts as a conversion (purchase, sign-up, MQL) |18| Lookback window | How far back to credit touchpoints (7, 14, 30, 90 days) |19| Cross-device | Can you link users across devices? |20| Data sources | Ad platforms, CRM, analytics, CDP |2122### Step 2: Select Attribution Models2324| Model | How It Works | Best For | Limitation |25|-------|-------------|----------|------------|26| First-touch | 100% credit to first interaction | Awareness analysis | Ignores nurture |27| Last-touch | 100% credit to final interaction | Bottom-funnel optimization | Ignores discovery |28| Linear | Equal credit to all touchpoints | Balanced view | Over-credits minor touches |29| Time-decay | More credit to recent touches | Sales cycle analysis | Discounts awareness |30| Position-based | 40/20/40 to first/middle/last | Balanced awareness + conversion | Arbitrary weights |31| Data-driven | ML-based credit allocation | Highest accuracy | Needs large data volume |3233### Step 3: Prepare Data3435| Requirement | Details |36|------------|---------|37| User identity | Consistent user IDs across channels |38| Touchpoint log | Timestamp, channel, campaign, content for each interaction |39| Conversion log | Timestamp, value, type for each conversion |40| Minimum volume | 1,000+ conversions for statistical models |41| Data quality | De-duplicated, time-synchronized, complete |4243### Step 4: Run Multi-Model Comparison4445Compare at least 3 models side by side:4647| Channel | First-Touch | Last-Touch | Linear | Data-Driven |48|---------|------------|------------|--------|-------------|49| Paid Search | X% | X% | X% | X% |50| Organic | X% | X% | X% | X% |51| Email | X% | X% | X% | X% |52| Social | X% | X% | X% | X% |53| Direct | X% | X% | X% | X% |5455### Step 5: Validate with Incrementality5657| Test Type | Method |58|-----------|--------|59| Geo holdout | Pause channel in test regions, measure lift |60| Ghost ads | Show vs don't show ads to matched groups |61| Matched market | Compare similar markets with/without spend |62| Conversion lift | Platform-native A/B tests |6364### Step 6: Optimize Budget Allocation6566Use attribution insights to shift spend:67- Identify over-credited channels (reduce spend, measure impact)68- Identify under-credited channels (increase spend, measure lift)69- Set ROAS targets per channel based on attributed value70- Re-run attribution monthly to track changes7172## Output Format7374```markdown75## Attribution Report7677### Model Comparison78[Multi-model channel credit table]7980### Key Findings811. [Channel X is over-credited by last-touch by Y%]822. [Channel Z's awareness contribution is missed by last-touch]8384### Budget Recommendation85| Channel | Current Spend | Recommended | Change | Expected Impact |86|---------|--------------|-------------|--------|----------------|87| [channel] | $X | $X | +/-X% | [impact] |8889### Validation Plan90- [ ] [Incrementality test for top recommendation]91```9293## Quality Checklist9495- [ ] Multiple models compared, not just one selected96- [ ] Lookback window is appropriate for sales cycle97- [ ] Data covers all significant touchpoints98- [ ] Cross-device tracking is addressed99- [ ] Incrementality validation is planned100- [ ] Results distinguish correlation from causation101102## Edge Cases103104- If conversion volume is low (< 500), stick to rule-based models105- For B2B with long sales cycles, extend lookback to 90-180 days106- If offline conversions exist, plan for online-offline matching107- For subscription businesses, attribute to LTV, not just first conversion108- If walled garden data (Meta, Google) is siloed, note the limitation
ashutoshsrivastava17