Course contentsShow
← Data Science
Lesson 1639 of 2,14539. Business Metrics and KPI DesignFree lesson

Time Windows and Attribution Decay

Defining lookback periods and understanding how attribution strength decays over time after an event or intervention.

Time Windows and Attribution Decay

You'll learn how to define lookback periods and model how attribution strength weakens as time passes after an event or intervention.

The Core Idea

When attributing outcomes to causes, time matters. A customer who sees your ad today and purchases tomorrow is more likely influenced by that ad than one who purchases six months later. Time windows set boundaries—how far back you look for attributable touchpoints. Attribution decay models how influence weakens over time.

Think of it like a stone thrown into a pond: the ripples are strongest near the impact point and fade as they spread outward. Similarly, an intervention's effect is typically strongest immediately after it occurs and weakens as days, weeks, or months pass.

Defining Time Windows

A lookback window specifies how far back in time to consider touchpoints. Common examples:

  • 7-day click window: Only clicks from the past week count toward conversion attribution
  • 30-day view window: Ad impressions within the last month are eligible for credit
  • 90-day engagement window: Customer interactions in the past quarter influence purchase attribution

Choosing the right window depends on your customer journey length and business cycle.

Attribution Decay Models

Within your time window, not all touchpoints carry equal weight. Decay models include:

  • Linear decay: Attribution drops steadily over time (e.g., Day 1 = 100%, Day 7 = 30%)
  • Exponential decay: Sharp initial drop, then gradual decline (common in digital marketing)
  • No decay (flat): Equal weight throughout the window—simpler but less realistic
  • Step function: Full credit within X days, zero credit beyond

Example: A 14-day exponential decay might give 80% credit to touchpoints from the last 3 days, 15% to days 4-7, and only 5% to days 8-14.

Key Takeaway: Setting appropriate time windows and decay functions ensures you attribute outcomes to the most relevant causes while avoiding over-crediting distant, weakly-related events.