Cross-border marketing attribution: what to actually measure
Short version: in cross-border China marketing, the clean last-click funnel you're used to at home does not exist, and no dashboard can reconstruct it. Each platform — Tmall, Douyin, WeChat, Red — is a walled garden that won't let a third party follow a shopper from impression to purchase, so stop trying to prove one channel "drove" a sale. Measure three things instead: whether spend is causing incremental lift (test it, don't attribute it), whether branded pull is growing (localized brand search, saves, repeat), and unit economics per garden. Everything else on the dashboard is decoration.
The single most common request I get from a headquarters finance team goes like this: "Show us the China funnel — impression to click to add-to-cart to purchase, by channel, so we can shift budget to what's working." It's a completely reasonable question in most markets. In China it's a fantasy, and chasing it wastes months. The sooner you accept that, the sooner you start measuring things that are actually true.
Why the tidy funnel doesn't exist here
China's internet is a set of walled gardens that were, until recently, openly hostile to each other. Alibaba's Taobao and Tmall long barred shoppers from paying with Tencent's WeChat Pay; Douyin still won't let you cleanly forward a video into a WeChat chat. The platforms have loosened a little under regulatory pressure, but the data walls never came down. A third party cannot follow a user from a Douyin ad into a Tmall purchase, because Tmall does not expose sales-conversion data to anyone outside its own ecosystem. There is no shared cookie, no cross-garden user ID, no Google-and-Meta duopoly stitching it together for you.
So the discovery almost always happens in one garden and the purchase closes in another. A shopper sees a Red review on Monday, watches a Douyin livestream on Wednesday, searches your name on Taobao on Sunday, and buys. Every platform's dashboard will claim full credit for that sale, and every platform is lying by omission. Add them up and you've "attributed" 300% of your revenue. This isn't a tooling gap you can buy your way out of — it's structural.
Stop attributing. Start testing.
The Western measurement world has quietly moved past last-click for the same reason (privacy walls, iOS changes, signal loss), and the 2026 consensus is a triangulation stack: marketing mix modeling for the portfolio view, incrementality testing for causal ground truth, and platform attribution as a directional, in-flight signal only. That stack happens to be exactly right for China, where the walls are higher than anywhere else. The order of trust matters:
| Method | Question it answers | Trust it in China for |
|---|---|---|
| Incrementality test (geo holdout, on/off) | Did this spend cause sales that wouldn't have happened? | The real decisions: is a channel worth the money at all |
| Marketing mix modeling (MMM) | How does total spend across gardens map to total sales over time? | Budget splits at the portfolio level, once you have ~12 months of data |
| Platform attribution (Tmall, Douyin dashboards) | What did this platform's own algorithm optimize toward? | Directional, in-flight tuning inside one garden — never cross-garden credit |
| Brand-pull signals | Is demand for you growing independent of paid? | The leading indicator that the whole thing is working |
The unlock is incrementality. You don't need cross-platform tracking to run a holdout: turn a channel off in three comparable cities for four weeks, leave it on in three others, and watch what happens to total sales. If nothing moves, that channel wasn't driving the sales its dashboard claimed. This is the one measurement that survives the walled gardens, because it measures the outcome, not the path.
The brand-pull signals worth watching weekly
Attribution tells you where a sale came from. Pull tells you whether you're building an asset. In a market where you start as a stranger, pull is the number that predicts next quarter. Track a short list, and track the trend, not the absolute:
- Localized branded search. Searches for your Chinese name on Taobao, JD, and within Red. Rising branded search with flat paid spend is the cleanest signal that seeding is working.
- Saves and shares, not likes. On Red and Douyin, a save (收藏) is a purchase-intent bookmark; a like is a reflex. Watch the save-to-like ratio on your KOC content.
- Repeat and private-traffic retention. Reorder rate and how many buyers you pull into your owned WeChat channels, where you can actually reach them again without paying for it.
- Unaided mention. Are people naming you in comments and reviews you didn't seed? Slow to move, hard to fake, worth more than any impression count.
Measure economics per garden, not one blended CAC
Because the gardens don't share data, a single blended cost-per-acquisition across all of China hides more than it reveals — it averages a cheap, high-intent Taobao search buyer with an expensive Douyin impulse buyer and tells you nothing actionable. Hold each garden to its own honest math: contribution margin per channel, payback period, and repeat behavior of the buyers it brings. Some channels exist to convert existing demand; some exist to create it. Judge them on the job they actually do, not on a shared ROAS column that pretends they're interchangeable.
The vanity metrics to delete from the deck
If a number goes up whenever you spend more and tells you nothing about whether that spend paid off, it's decoration. The usual suspects: gross impressions, total followers, GMV during a subsidized festival window (618 and Singles' Day flatter everyone), and any platform's self-reported "attributed conversions" presented as cross-channel truth. Keep them for context if you must, but never let them drive a budget decision. A phone number reconciled across your CRM and one incrementality test will out-inform a forty-widget dashboard every time.
Bottom line
Cross-border attribution in China isn't hard because you lack the right tool; it's structurally impossible in the clean form headquarters wants, because the platforms are built not to share. So change the question. Prove causation with incrementality tests, model the portfolio with MMM once you have the history, watch branded pull as your leading indicator, and hold each garden to its own economics. Do that and you'll make better budget calls than the competitor with the prettier funnel diagram — the one that's quietly 300% over-attributed.
If you're trying to build a China measurement framework the finance team will actually trust, that's the kind of problem I help untangle — reach out. For the channel that makes repeat measurable in the first place, read how WeChat private traffic actually works, or step back to the full cross-border marketing playbook.