In the Basic level we saw that attribution shares each conversion among the Google Ads clicks that preceded it, and that data-driven is the default. This level answers the useful question: which decisions can I make with it and which ones can't I?
How data-driven attribution (DDA) works
Google compares conversion paths with and without each touchpoint (campaign, keyword, ad, device) using all the data in the account, and estimates how much each click increased the probability of a conversion. Credit is shared in proportion: a generic first click that opened the path might take 40%, the final brand click 60%. It is specific to your account and it is recalculated continuously. Advantages: it allocates in favour of what starts things off; Smart Bidding bids with that allocation and gives generic terms room to breathe.
What changes between last click and DDA… and what doesn't
| Changes | Doesn't change |
|---|---|
| The allocation between Ads campaigns and keywords | The account's conversion total |
| Brand loses, generic terms and Display gain credit | What happens outside Google Ads (email, organic, social) |
| Smart Bidding's bids (they follow the allocation) | The date (always the click's) |
| Conversions with decimals in the reports | Offline conversions are shared out too |
Rule: DDA changes how it is shared, not how much there is. If the total does not go up, DDA has not "improved" anything: it just counts better.
Conversion paths and windows
The attribution reports (Goals → Attribution) show paths (sequences of clicks before converting), days to conversion and clicks to conversion. Use them to:
- Set windows with data: if 90% convert within 12 days, a 30-day window is fine; 7 days is too short.
- See which campaigns assist (they appear at the start of paths) even if they don't close: they are the ones last click undervalued.
- Spot self-cannibalisation: generic → brand paths within the same session mean brand is claiming what the generic term brought in.
The limits of any attribution in Google Ads
- It only sees Google Ads clicks (and some video impressions). A real path of "Instagram → Google brand → purchase" shows up as "Google brand → purchase".
- It does not measure incrementality: attributing 100% of a purchase to a brand click does not tell you whether the purchase would have happened without it (it almost always would).
- It depends on consent and modelling: part of the paths are estimated.
- It changes over time: DDA is recalculated; today's report does not match last month's for the same period.
The rule for deciding budgets
- Attribution to allocate: decide which campaigns, groups and keywords deserve more or less inside Google Ads, with DDA and the business-objective judge (module 2).
- Incrementality to confirm: before raising or cutting a channel significantly (brand, remarketing, Display, video), run an experiment (holdout, geo, before/after on the total) that answers "what happens without this?" (module 7).
- The CRM to validate: the relationship between conversions and real customers, lead by lead.
What not to do: argue for months about which model "is right". No model is right; they have uses.
Practical cases
- Brand with a ROAS of 15 and generic with a ROAS of 2 under last click: with DDA, brand drops to 8 and generic rises to 3.5. The money has not changed; the reading has. A brand holdout tells you whether brand is incremental (it usually is, partly, when competitors bid on your name).
- Display "doesn't convert" under last click: DDA gives it assist credit. Raise the budget? Only if a geo experiment shows a total lift.
- Conversions that "move" from week to week: delay + DDA recalculation. Evaluate closed periods.
💡 Ninja trick: attribution inside Google Ads is a layer on top of conversions that, in lead generation, cannot tell a good lead from a bad one. That is why the order of priorities is: first lead quality (lesson 5) —so that Google receives conversions with their real value—, then attribution. A perfect allocation of worthless conversions is a perfect allocation of nothing.
What you should remember
- DDA shares credit by how much each click increased the probability of a conversion; it changes the allocation, not the total.
- Paths and windows: set windows with data; spot assists and cannibalisation.
- Limits: only Ads clicks, no incrementality, modelling involved, and it changes over time.
- Attribution to allocate, incrementality to confirm, the CRM to validate.