← All playbooks·Digital attribution

Playbooks · Digital attribution

Set an attribution window for post-exposure conversions

This job sets the click-through and view-through lookback windows an ad server or DSP will use to credit a conversion to an exposure. The output is a specific day count per channel type that gets written into the measurement plan and configured in the platform before the campaign credits a single conversion.

What you need first

  • Day-by-day conversion lag distribution for the last 3-6 comparable campaigns, pulled from the analytics platform (GA4, MMP, or ad server conversion path report)
  • Category purchase cycle length (average days from first touch to purchase), from client CRM or category research
  • Platform lookback options and current defaults for click-through and view-through, from the DSP/ad server UI
  • Media plan flight dates and channel mix, from the current media plan
  • Attribution window used on this account's prior campaigns, from the last campaign report, for comparability

The procedure

  1. Pull the day-by-day conversion count after exposure for 3-6 comparable campaigns from the analytics platform, producing a raw lag distribution table
  2. Convert the lag counts into cumulative percentages at day 1, 7, 14, 30 and 45 to locate where the curve flattens
  3. Compute marginal conversions per day within each interval (1-14, 14-30, 30-45) to find the point where extra days stop paying for themselves
  4. Cross-check the candidate cutoff against the category's average purchase cycle from CRM data as a plausibility check on the curve, not the primary driver
  5. Set separate click-through and view-through windows in the ad server/DSP, since view-through conversions decay faster than click-through
  6. Document the chosen window and rationale in the measurement plan and flag the change to anyone consuming the attributed-conversion feed downstream, including dashboards and MMM inputs

Worked through with numbers

Pull the last 4 comparable home-goods campaigns from GA4: 12,400 tracked conversions within 45 days of exposure. Cumulative capture: day 1 = 40% (4,960), day 7 = 68% (8,432), day 14 = 82% (10,168), day 30 = 94% (11,656), day 45 = 100% (12,400). Marginal rate by interval: day 1-14 adds 5,208 conversions over 13 days = 400/day; day 14-30 adds 1,488 over 16 days = 93/day; day 30-45 adds 744 over 15 days = 50/day. The client's CRM shows an 18-day average purchase cycle, close to the day-14 mark where the curve bends. Set the click-through window to 14 days, which captures 82% of conversions at a 400/day marginal rate before the curve drops to 93/day. Set the view-through window to 1 day, matching this account's own decay data in the DSP. Read it as: everything past day 14 is mostly picking up activity better explained by other channels or organic return, not the exposure itself.

Where it goes wrong

  • Set view-through and click-through windows separately: view-through decays within a day or two, click-through can run 14-30 days, and one shared window miscredits one of them
  • Treat the purchase cycle length as a sanity check, not the primary input: the lag curve from actual campaign data should set the cutoff, the CRM number only confirms it's plausible
  • Recompute the curve for this account instead of keeping the platform default (often 30-day click, 1-day view): defaults are tuned to the median advertiser and can overstate this account's real decay pattern
  • Reset the window when the channel mix changes materially, for example adding TV or offline, since those carry longer and differently shaped lag curves than the campaigns the original window was based on

How to know it is right

The window holds if the marginal conversion rate in the interval just past the cutoff stays below half the rate in the interval before it, and this pattern repeats in a second, independent reporting period.

Terms used

← All playbooks Calculated in ValueMetrics →