Attribution for Shopping
How attribution models and conversion windows decide which Shopping products get credit for a sale, and why changing the model changes which products look like winners.
Attribution is how Google decides which click gets credit when a shopper touches several ads before buying. The model and the conversion window you pick don't just change reporting, they change which products and campaigns look good, and therefore where bidding sends your money. This explains the models, the windows, and why the same account can tell two different stories depending on the settings.
What attribution decides
Most sales aren't one click. A shopper sees a Shopping ad for your product, leaves, sees it again on a different search two days later, then buys. Attribution is the rule that decides which of those clicks (and therefore which product, campaign and keyword) gets credit for the sale.
This matters far beyond reporting tidiness. Smart Bidding optimises toward attributed conversion value. So the attribution model doesn't just describe where credit went; it decides where the next pound of budget goes. Change the model and you change which products the bidder thinks are worth investing in.
The models, briefly
| Model | How it splits credit | When it's useful |
|---|---|---|
| Data-driven (DDA) | Distributes credit across the path based on your account's own conversion patterns | The default; generally the most accurate for a real account |
| Last click | 100% to the final click before the sale | Diagnostic baseline; overweights closing queries |
| First click | 100% to the first click | Rarely used; overweights discovery |
| Position-based | Weights first and last click heavily, middle lightly | A fixed compromise if you distrust DDA |
| Linear / time-decay | Even split, or more credit the closer to conversion | Legacy fixed rules, mostly superseded by DDA |
Google has retired most of the fixed rule-based models in favour of data-driven attribution for new setups, and for good reason: a fixed rule can't know whether your Shopping ads open sales or close them, whereas DDA infers it from the paths your shoppers actually took.
Why the model changes which products look good
Different models reward different roles in the funnel, and Shopping products play different roles.
- Last click flatters closers. Products people search for by name near the moment of purchase (branded, high-intent) hoover up last-click credit. Products that introduce a shopper to your range earlier get nothing under last click: even though the sale wouldn't have happened without them.
- Data-driven spreads credit to assists. Under DDA, those introducer products get partial credit for the sales they helped start. Their measured ROAS rises; the closers' falls slightly as credit redistributes.
So a product that looks mediocre under last click can look valuable under DDA, and vice versa: with no change to the actual ads. Whichever lens bidding uses becomes the reality it optimises toward.
Deep dive Conversion windows, and why the same product can win or lose on a setting
The conversion window is the maximum time between ad click and sale that still counts as a conversion. It's a quieter lever than the model, but it moves the numbers just as hard.
Set the window to 7 days and every sale that took 10 days to close simply vanishes from your reporting: the click looks wasted, the product looks weak, and bidding pulls back from it. Widen the window to 30 days and those same sales reappear, the product's ROAS jumps, and it suddenly deserves budget. Nothing about the product changed; a setting did.
The right window is a fact about your products, not a preference:
- Impulse / low-consideration items (cheap accessories, consumables) convert same-session. A short window loses almost nothing and keeps signal fresh.
- Considered purchases (furniture, high-ticket, anything people compare) routinely take 1–3 weeks. A short window here systematically under-credits them and starves genuinely profitable products.
Two practical cautions:
- Don't compare periods across a window change. If you widen the window mid-quarter, later weeks credit more delayed sales than earlier ones: the "improvement" is partly the setting. Any trend line that straddles a window or model change is contaminated; note the change date.
- Windows interact with lumpy sales. Products with a few large, slow-closing orders per month are especially sensitive: a too-short window can make a profitable product look dead between orders, which whipsaws tROAS learning.
Before trusting any product-level ROAS comparison, confirm both products are being measured under the same model and window. Otherwise you're comparing two different rulers.
How to use attribution in diagnosis
Treat model comparison as a diagnostic, not a setting to fiddle with weekly:
- Run on data-driven for actual bidding: let it optimise on your real paths.
- Compare against last click occasionally to see how much assist credit your Shopping products earn. A big gap means your Shopping ads open more sales than last-click reporting admits: useful when someone claims Shopping "doesn't convert".
- Set the window to your real purchase latency and leave it. Log the date if you change it.
- Remember bidding follows attribution. Any measured ROAS (the number behind every profit decision and product-level target) is an attributed number. Two products with identical true performance can rank differently purely on the model and window. Fix the ruler before you trust the ranking.
Attribution won't make a bad product good. But the wrong model or window will make a good product look bad: and bidding believes what the reporting tells it.
Frequently asked questions
Which attribution model should I use for Shopping?
Data-driven attribution is the default and usually the right call, it distributes credit based on your account's actual conversion paths rather than a fixed rule. The main reason to look at last-click is diagnostic, comparing the two shows you how much assist credit your Shopping products are getting.
What conversion window should I set?
Match it to how long people actually take to buy your products. Impulse buys convert same-day; considered purchases take a week or more. Too short a window drops real conversions; too long a window credits clicks that had little to do with the sale. 30 days is a common default worth revisiting against your data.
Why did my ROAS change when I didn't change my ads?
Check whether the attribution model or conversion window changed. Both silently re-allocate credit across products and campaigns, so the same clicks and sales can produce different ROAS numbers. It's a reporting shift, not a performance shift, but bidding responds to it as if it were real.