DIRECT TRAFFIC IS WHERE LAST-TOUCH GIVES UP
A typical e-commerce account sees 25–45% of last-touch conversions land in the "Direct" bucket — no UTM, no referrer, no platform claim. Last-touch then credits these conversions to nothing.
Run the join: of those Direct conversions, ~70% had at least one identified marketing touch within the prior 24 hours, and ~85% had one within the prior 7 days. The marketing worked. The user just typed the URL or clicked a bookmark instead of clicking the ad again.
A simple recovery: re-attribute Direct conversions to the most recent prior identified touch within a 7-day window. This preserves the "closer" semantics of last-touch while stopping the bleeding into the unattributed Direct bucket.
## Comparing Last-Touch to Other Models
### Last-Touch vs First-Touch
| Aspect | Last-Touch | First-Touch |
|--------|-----------|-------------|
| **Credits** | Final interaction | Initial interaction |
| **Favors** | Closers (email, retargeting) | Introducers (paid social, content) |
| **Best for** | Conversion optimization | Demand gen measurement |
| **Blind spot** | Ignores awareness | Ignores conversion |
| **B2B fit** | Less useful (long cycles) | More useful (sources pipeline) |
**Use both together** to see the full picture of channel roles.
### Last-Touch vs Linear
| Aspect | Last-Touch | Linear |
|--------|-----------|--------|
| **Credit distribution** | 100% to one | Equal to all |
| **Complexity** | Very simple | Still simple |
| **Bias** | Strong closer bias | No bias (but naive) |
| **Best for** | CRO, simple funnels | Baseline, long journeys |
Linear is often a better default than last-touch because it doesn't systematically distort channel value.
### Last-Touch vs Data-Driven
| Aspect | Last-Touch | Data-Driven |
|--------|-----------|-------------|
| **Credit distribution** | Fixed (100% last) | Algorithmic |
| **Transparency** | Fully transparent | Black box |
| **Data needs** | Minimal | Requires volume |
| **Best for** | Simplicity | Scale with ML |
Data-driven can improve on last-touch, but many implementations (like Google's) are opaque and may have platform bias.
## Moving Beyond Last-Touch
### Step 1: Add First-Touch Comparison
Start by running first-touch alongside last-touch:
```sql
-- Compare first vs last touch by channel
SELECT
channel,
SUM(first_touch_credit) as first_touch_revenue,
SUM(last_touch_credit) as last_touch_revenue,
SUM(first_touch_credit) / SUM(last_touch_credit) as ratio
FROM attribution_results
GROUP BY channel
ORDER BY ratio DESC;
```
Channels with ratio > 1 are under-valued by last-touch.
Channels with ratio < 1 are over-valued by last-touch.
### Step 2: Implement Multi-Touch
Graduate to linear attribution for a neutral baseline, then explore:
- Time-decay (credits recency without ignoring history)
- Position-based (emphasizes first and last)
- Data-driven (if you have volume)
### Step 3: Validate with Incrementality
Any attribution model—including multi-touch—is still correlational. Use holdout tests to validate:
1. Pick a channel you suspect is over/under-credited
2. Run a geo-holdout or randomized pause test
3. Measure incremental lift vs attributed lift
4. Calibrate your attribution model to match
A WORKED EXAMPLE
A subscription brand uses last-touch as its only model. The dashboard says Email drives 40% of revenue — the team reads this as "Email is our hero channel" and increases the email program budget two quarters in a row.
Then they pause Paid Social in three matched geos for a month. Revenue in the holdout regions falls 22% below control. Recomputing with Linear attribution, Email's share drops to 18% and Paid Social's share rises from 12% to 31%. The Email flow was closing customers Paid Social had already brought to the site — not driving them. They cut the email program back, reallocate to Paid Social, and blended ROAS climbs because the budget is now upstream of the conversion, not piled on top of it.
## Summary
Last-touch attribution is simple, universal, and systematically misleading for budget decisions.
**Use last-touch when:**
- Optimizing conversion rate and final-step creative
- Sales cycles are single-session
- Comparing against platform reporting
- You need a simple baseline
**Don't use last-touch alone because:**
- It over-credits closers, under-credits awareness
- It leads to the attribution death spiral
- It conflates "last" with "causal"
- It ignores 90% of most customer journeys
**Best practice:** Run last-touch alongside first-touch and linear attribution. Use the gaps between models to understand channel roles. Graduate to multi-touch for budget decisions. Validate with incrementality tests.
## Further Reading
**On Attribution Model Selection:**
- [First-Touch Attribution: When to Use It](/articles/first-touch-attribution) — The counterpart to last-touch
- [How to Choose the Right Attribution Model](/articles/how-to-choose-attribution-model) — Framework for selection
- [What is Multi-Touch Attribution?](/articles/what-is-multi-touch-attribution) — The alternative to single-touch
**On the Attribution Death Spiral:**
- [Rand Fishkin on Attribution](https://sparktoro.com/) — SparkToro's critique of last-click
- [Kevin Hillstrom on Incrementality](https://blog.minethatdata.com/) — Why attribution alone isn't enough
## Key takeaways
- Last-touch credits only the final interaction—everything before gets zero
- Over-credits closers (email, brand search), under-credits introducers (paid social, content)
- Still useful for conversion rate optimization and bottom-funnel analysis
- Using last-touch alone for budget decisions leads to the 'attribution death spiral'
## FAQ
**Why is last-touch the default in most analytics tools?**
Simplicity and defensibility. Last-touch is easy to implement (just look at the referrer on conversion), easy to explain (this is what converted them), and hard to argue with (it was literally the last thing). It's also what advertisers demanded when digital measurement started.
**Is last-click the same as last-touch?**
Essentially yes. Last-click specifically refers to the last clicked ad or link, while last-touch can include non-click interactions like email opens or ad views. In practice, most implementations treat them the same—the last trackable interaction before conversion.
**Did GA4 remove last-touch attribution?**
No—last-click is one of only two models GA4 kept (along with data-driven). Google removed first-touch, linear, time-decay, and position-based. Last-click remains the default for most GA4 reports.
**Should I use last-touch for my paid ads?**
For optimizing ad creative and landing pages, yes—last-touch shows what directly converted. For budget allocation between channels, no—it will tell you to cut the awareness channels filling your funnel. Use multi-touch for budget decisions.
**What's the 'attribution death spiral'?**
A failure pattern from last-touch optimization: cut 'underperforming' awareness channels → funnel dries up → increase spend on closers → diminishing returns → CAC rises → revenue plateaus. You've starved the funnel by optimizing toward a misleading metric.