Attribution (Conversion Attribution)

Attribution is the logic that distributes conversion credit across ad interactions. A single purchase might be preceded by a search ad click, a social video view, and an email open — attribution decides which gets the credit (and how much).

Why it matters: Your ROAS and CPA numbers are only as reliable as the attribution model behind them. Two advertisers running the same campaign but using different models will see different “winners” and make different budget decisions.

Common attribution models

• Last-click: 100% credit to the last click before conversion. Simple, but ignores all earlier awareness work.
• First-click: 100% credit to the first interaction. Useful for understanding top-of-funnel effectiveness.
• Linear: Equal credit to every touchpoint. Fair but noisy.
• Time-decay: More credit to touchpoints closer to conversion. A middle ground.
• Data-driven (algorithmic): Uses historical data to assign fractional credit based on each touchpoint’s incremental impact. Available in Google Ads and Meta for accounts with sufficient conversion volume.

So what: Choose a model that matches your decision horizon. If you optimize for same-day purchases, last-click may be fine. If you run multi-week nurture sequences, you need a model that credits early touchpoints.

There is no single formula for attribution — the calculation depends on the model. But the general structure is:

Conversion credit per touchpoint = Weight(touchpoint) × 1 conversion

Where Weight is defined by the model:

• Last-click: Weight = 1 for the final touch, 0 for all others.
• Linear: Weight = 1 / (number of touchpoints in the path).
• Time-decay: Weight = e^(−λ × time gap) normalized across touchpoints.
• Data-driven: Weight = predicted incremental lift of that touchpoint from a machine-learning model.

Important caveats

1. Lookback window — Conversions outside the window get zero credit. A 30-day click window vs. a 7-day click window will produce very different attribution splits.
2. Conversion action — Attribution is per conversion action (purchase, sign-up, add-to-cart). Mixing actions in one report can distort results.
3. Cross-device — Most platforms cannot stitch a user across devices unless they are logged in. Reported attribution is often device-specific.
4. Privacy changes — With signal loss (cookie deprecation, iOS App Tracking Transparency), platforms rely more on modeled conversions. Treat reported ROAS with uncertainty; always check the “modeled vs. observed” breakdown if available.

Attribution is not a single number — it is a setting that changes every other metric. When you look at a dashboard, the first thing to check is which attribution model is applied. If the label says “Last-click” or “Data-driven,” you know the ROAS and CPA numbers are calculated under that rule.

What to look for:

• Model label — Usually shown in the report header or a dropdown. If missing, assume last-click (the platform default in most tools).
• Lookback window — 30-day click / 1-day view? 7-day click? This can change CPA by 20-40% for consideration-stage products.
• Conversion action — Are you looking at “All conversions” or a specific action? Mixing actions inflates ROAS.

What to pair it with:

• Path length — Average number of touchpoints before conversion. If it’s >3, last-click will undercredit your upper-funnel channels.
• Assisted conversions — A report that shows which channels appeared in the path but did not get the last click. High assisted conversion share = the channel is doing awareness work.
• Model comparison — Run a side-by-side report (e.g., last-click vs. data-driven) to see which channels gain or lose credit. The difference is your “attribution gap.”

Attribution is a setting, not a lever you pull daily. But you can influence how credit flows by changing campaign structure, bidding strategy, and measurement setup.

Levers that shift attribution outcomes

• Conversion window — Shortening the lookback window concentrates credit on last-click channels. Lengthening it spreads credit to earlier touchpoints.
• Bid strategy — Target CPA and Target ROAS bid strategies optimize toward the attribution model you have selected. Switching from last-click to data-driven can cause bids to shift spend toward upper-funnel placements.
• Campaign structure — Separating prospecting (awareness) campaigns from retargeting (conversion) campaigns makes it easier to see each channel’s role. Blended campaigns make attribution noisy.
• Offline conversion import — If you import phone calls or in-store purchases, attribution becomes more complete. Without it, digital touchpoints that drive offline sales get zero credit.

Tradeoffs

• Last-click is simple but blind — It overcredits the final touch and starves upper-funnel channels. You may see CPA drop in the short term while total addressable audience shrinks.
• Data-driven is smarter but data-hungry — It requires ~300+ conversions per campaign per 30 days to stabilize. Below that, the model may flip credit randomly week to week.
• Multi-touch models (linear, time-decay) are fair but imprecise — They assume all touchpoints matter equally (linear) or that recency is the only factor (time-decay). They do not measure true incrementality.

Easy mistake: A home-services advertiser switched from last-click to linear and saw ROAS drop 30%. They panicked and switched back. The drop was expected — linear simply spread credit across more touchpoints. The actual revenue had not changed. Always run a model comparison report before changing your optimization model.

Attribution is the lens through which you see campaign performance — change the lens, change the picture. Key takeaways:

• Always state the attribution model, lookback window, and conversion action when reporting ROAS or CPA.
• Last-click is simple but hides upper-funnel value; data-driven is better but requires volume.
• Privacy changes increase modeled conversions — validate with incrementality tests or holdout experiments.