Why Attribution Matters 📊
I learned this the hard way. A few years back, I was managing a campaign for an international skincare brand entering China. We were running ads across Douyin, WeChat, and Baidu. The data looked great on Douyin—high conversion rates, low CPA. So I doubled down on Douyin and cut budgets on the other platforms.
Big mistake. 💥
Two weeks later, overall conversions dropped by 40%. What happened? Our attribution model was "last-click"—giving all the credit to the last ad the user clicked. But in reality, users were discovering the brand on Douyin, researching on WeChat, then finally converting via Baidu search. By cutting WeChat and Baidu, we killed the funnel that fed Douyin's conversions.
That's when I realized: attribution isn't just a data problem—it's a strategy problem. The model you choose determines which platforms get credit, which campaigns get budget, and ultimately, whether your strategy actually makes sense.
Your attribution model shapes your reality. If you're measuring wrong, you're optimizing wrong.
The 7 Attribution Models 🔍
Here's the thing: there's no "right" attribution model. Each one tells a different story about your customer journey. The question is: which story helps you make better decisions?
Let me break down the 7 models I use when advising clients, and when each one makes sense.
1. Last-Click Attribution
The logic: 100% credit goes to the last ad the user clicked before converting.
When it works: Direct response campaigns with short customer journeys. If you're selling a low-consideration product (like a $10 snack) and most conversions happen within one session, last-click is simple and actionable.
The problem: It completely ignores the "discovery" phase. In the skincare example above, Douyin got all the credit, but WeChat and Baidu did the heavy lifting of building trust and consideration.
If you're using last-click, you're probably overvaluing "harvesting" ads and undervaluing "planting" ads. Keep that in mind when allocating budget.
2. First-Click Attribution
The logic: 100% credit goes to the first ad the user clicked.
When it works: Brand awareness campaigns. If your goal is to understand which channels are best at introducing new customers to your brand, first-click tells you that.
The problem: It ignores the rest of the journey. A user might click your ad, then ignore you for 3 weeks before converting via a different channel. First-click gives all the credit to that initial click, even if 5 other touchpoints influenced the decision.
3. Linear Attribution
The logic: Credit is split equally across all touchpoints.
When it works: When you want a "fair" view of the whole funnel, and you're not sure which touchpoints matter most. It's a good starting point if you're new to a market and don't have enough data for data-driven models yet.
The problem: It dilutes the impact of critical touchpoints. If your data shows that 80% of conversions involve a Douyin ad in the awareness phase and a Baidu search in the conversion phase, linear attribution treats a random mid-funnel impression the same as those key moments.
4. Time Decay Attribution
The logic: Touchpoints closer to the conversion get more credit. The exact formula varies, but typically it's exponential—a touchpoint 1 day before conversion gets much more credit than one 10 days before.
When it works: Longer consideration cycles (like B2B, real estate, or high-ticket items). It acknowledges that awareness matters, but conversion is still driven by what happened recently.
The problem: The "decay rate" is subjective. Platforms often use different formulas, so comparing attribution across platforms becomes tricky.
5. Position-Based (U-Shaped) Attribution
The logic: 40% credit to first touchpoint, 40% to last, and the remaining 20% split across the middle.
When it works: When you want to balance "discovery" and "conversion". This is my go-to model for most full-funnel campaigns because it acknowledges that both the first impression and the final click matter.
The problem: The 40-20-40 split is arbitrary. For some brands, the first touchpoint might be 60% of the value; for others, it might be 20%. But most platforms don't let you customize the weights.
For most international brands entering China, start with position-based (40-20-40). It's a good balance of recognizing awareness and conversion. Then, as you gather data, move to data-driven attribution.
6. Data-Driven Attribution (DDA)
The logic: An algorithm analyzes your historical data and assigns credit based on each touchpoint's actual contribution to conversion.
When it works: When you have enough conversion data (typically 1000+ conversions per month). DDA can uncover non-obvious patterns—maybe your WeChat ads aren't driving many last-clicks, but they're present in 80% of conversion paths. That's valuable insight.
The problem: It's a black box. You don't know exactly how the algorithm is weighting touchpoints. Also, if your data is messy (duplicate conversions, incorrect tracking), DDA will give you wrong answers.
7. Custom Attribution
The logic: You define the rules. Maybe you want to give 50% credit to the first touchpoint, 30% to the last, and 20% to any touchpoint on Douyin (because you know Douyin drives engagement even if it doesn't get the last click).
When it works: When you have a deep understanding of your customer journey and want to encode that into your attribution model.
The problem: It requires ongoing maintenance. As your strategy changes, your custom rules might become outdated. Also, most self-serve platforms don't offer custom attribution—you'll need to use a third-party attribution tool.
Quick Comparison Table 📋
Here's a side-by-side view to help you choose:
| Model | Best For | Pros | Cons |
|---|---|---|---|
| Last-Click | Short customer journeys, low-consideration products | Simple, actionable | Ignores awareness |
| First-Click | Brand awareness campaigns | Highlights discovery | Ignores rest of journey |
| Linear | New to market, limited data | Fair, full-funnel view | Dilutes key touchpoints |
| Time Decay | Long consideration cycles | Balances awareness & conversion | Subjective decay rate |
| Position-Based | Full-funnel campaigns (RECOMMENDED) | Balanced view | Arbitrary weight split |
| Data-Driven | 1000+ conversions/month | Data-backed, adaptive | Black box, needs clean data |
| Custom | Deep customer journey insights | Tailored to your business | High maintenance, needs 3rd-party tool |
How Platforms Differ 🌐
Here's something that confuses a lot of advertisers: different platforms use different attribution models by default. And if you're comparing CPA across platforms, you're probably comparing apples to oranges.
Let me give you a practical example. On one platform, the default might be "last-click within 7 days". On another, it might be "last-click within 1 day". If you're comparing CPA between these two platforms, the second one will look much more efficient—but that's because it's not counting conversions that happened 2-7 days after the click.
Always check the default attribution window (click-through window and view-through window) before comparing platform performance. A 7-day click window vs. a 1-day click window can make your CPA look 3x different.
Broadly speaking, here's what I see across the major China platforms:
- Douyin (Ocean Engine): Offers multiple attribution options. The default is often "last-click", but you can switch to position-based or data-driven if you have enough data.
- WeChat Ads: Supports click attribution, view attribution, and data-driven attribution. The view-through window is particularly important for WeChat because Moments ads often drive awareness rather than immediate clicks.
- Baidu SEM: Traditionally last-click, but with their Smart Bidding products, they're moving toward data-driven attribution behind the scenes.
- Xiaohongshu (Little Red Book): Their attribution is still evolving. Currently, they tend to use a combination of last-click and time-decay, but it varies by ad product.
The bottom line: don't trust platform-reported CPA at face value. Always triangulate with your own tracking (ideally a third-party attribution tool or at least Google Analytics / your own CRM data).
How to Choose the Right Model 🎯
Okay, enough theory. Here's how I actually choose attribution models for the brands I work with:
Step 1: Understand Your Customer Journey
Before picking a model, you need to know: how do your customers actually buy? For some brands, it's a 1-day journey (see ad → click → buy). For others, it's a 30-day journey (see ad → research → compare → buy).
If you don't know, use a neutral model (like linear) for the first 2-3 months while you gather data.
Step 2: Match the Model to Your Campaign Goal
| Campaign Goal | Recommended Model | Why |
|---|---|---|
| Brand Awareness | First-Click or Position-Based | Recognizes the value of introducing new customers |
| Lead Generation | Last-Click or Position-Based | Focuses on conversion; position-based also values awareness |
| E-commerce (Short Cycle) | Last-Click | Simple, actionable for fast-moving consumer goods |
| E-commerce (Long Cycle) | Position-Based or Time Decay | Acknowledges that awareness and consideration take time |
| Mature Campaign (1000+ conv/month) | Data-Driven | Uses your own data to optimize attribution |
Step 3: Test and Iterate
Attribution isn't "set it and forget it". Every 3-6 months, I revisit the model and ask:
- Are our best-performing campaigns getting the budget they deserve?
- Are we under-investing in awareness because our model doesn't give it credit?
- If we switched to a different model, which platforms would look better/worse?
I also run "attribution sensitivity tests"—basically, I compare campaign performance under different attribution models to see how sensitive my decisions are to the model choice. If a campaign looks great under last-click but terrible under position-based, that's a red flag—it means the campaign is good at "closing" but not at "opening", which might hurt you long-term.
Common Mistakes I See 🪣
Let me save you some pain. Here are the most common attribution mistakes I see international brands make in China:
The Bottom Line 💡
Attribution is messy. There's no perfect model, and anyone who tells you otherwise is selling something.
But here's what I've learned from managing $10M+ in ad spend across China's platforms: the goal isn't to find the "right" model—it's to find a model that helps you make better decisions. And that means understanding the strengths and blind spots of whatever model you're using.
If you're running campaigns across multiple platforms in China, here's my advice:
- Start with position-based (40-20-40). It's a reasonable default that balances awareness and conversion.
- Triangulate with your own data. Don't trust platform-reported numbers. Use a third-party attribution tool or at least your own CRM data.
- Revisit your model every 6 months. As your brand grows and your customer journey changes, your attribution model should evolve too.
- Run sensitivity tests. Compare performance under different models to understand how much your decisions depend on the model choice.
Running paid media across Baidu, Douyin, WeChat, Xiaohongshu, Bilibili, and Bing China for international agencies and brands—we see these attribution challenges every day. If you're confused by conflicting data across platforms, or you suspect your attribution model is hiding the true performance of your campaigns, let's talk. We'll help you cut through the noise and build a measurement framework that actually makes sense for your business.
Next Steps: Not sure which attribution model fits your campaign? Send us your campaign data, and we'll run an attribution audit—free for the first 10 brands that reach out this month. Get in Touch →