Time Decay Attribution

Who this is for

• Marketing Analysts who need to credit multiple touchpoints fairly across a customer journey.
• Performance marketers optimizing budgets across channels and campaigns.
• Analysts building dashboards or reports that compare channel ROI.

Prerequisites

• Know your conversion event (e.g., purchase, lead) and lookback window.
• Have touchpoint data with user/session IDs, timestamps, and channels.
• Comfort with basic math and spreadsheets or SQL.

Why this matters

Real tasks you will face:

• Allocating revenue/lead credit to channels without overvaluing last click.
• Comparing campaign effectiveness when journeys have many touches.
• Explaining why recent touches often matter more and how you weighted them.
• Running budget shift scenarios based on time-weighted performance.

Concept explained simply

Time Decay Attribution gives more credit to touches that happen closer to conversion. Earlier touches still matter, just less. This mirrors real buyer behavior: memory fades and intent strengthens as people get closer to purchase.

Mental model

• Think of attention like a melting ice cube: the closer to conversion, the more solid the impact.
• We control how fast it melts with a “half-life” (h). After h days, weight halves.

Weight formula (using days as the unit): w = 0.5^(age_in_days / half_life). Normalize weights so they sum to 1 per conversion, then allocate credit.

How to implement (step-by-step)

1. Define scope. Choose the conversion and lookback window (e.g., purchases within 30 days of first touch).
2. Pick a half-life (h). Start practical: 7 days for fast cycles (retail), 14–30 days for longer cycles (B2B). Plan to backtest.
3. Compute touch weights. For each touch: age = conversion_time − touch_time. Weight w = 0.5^(age/h). Keep time units consistent.
4. Normalize per journey. Sum all w for that conversion; each touch credit = w_i / sum_w.
5. Aggregate. Sum credits by channel, campaign, ad group, etc.
6. QA. Randomly sample journeys to ensure credits sum to 1, timestamps make sense, and single-touch paths get 100% credit.

Worked examples

Example 1: B2C, half-life 7 days

Path: Facebook (14d), Paid Search (7d), Email (2d), Direct (0d). h=7.

Example 2: B2B, half-life 30 days

Path: LinkedIn (60d), Webinar (20d), SDR Call (5d), Direct (0d). h=30.

Example 3: Repeated channel touches

Path: Display (9d), Email (5d), Email (1d), Direct (0d). h=7.

Exercises

Do these now. Answers are available below each exercise in the Exercises panel and in the Solutions.

Exercise 1 — Single journey calculation

Given: Facebook (14d), Paid Search (7d), Email (2d), Direct (0d). h=7. Compute normalized credit by channel.

• Formula: w = 0.5^(age/7)
• Normalize per journey
• Round to one decimal place

Exercise 2 — Aggregate across multiple journeys

Five conversions; h=7. For each journey, compute per-touch credits, normalize, then sum by channel.

• Output needed: Channel share % across all 5 conversions.

Common mistakes and self-checks

• Not normalizing per journey. Self-check: Pick 10 random conversions; confirm channel credits sum to 100% each.
• Wrong time unit. Self-check: Ensure age and half-life are in the same unit (both days or both hours).
• Over-crediting Direct. Self-check: Compare results with and without last-non-direct rule; large swings imply Direct is absorbing brand recall.
• Changing half-life per channel without evidence. Self-check: Keep one h for fairness; only vary after documented backtests.
• Including post-conversion touches. Self-check: Filter touches strictly before conversion timestamp.
• Timezone/timestamp errors. Self-check: Convert all times to a single timezone before calculating ages.

Practical projects

• Project A: Build a spreadsheet model that takes touch timestamps and outputs time-decay channel credits with a configurable half-life. Deliverable: a template with inputs, weights, and pivoted channel totals.
• Project B: Backtest half-lives (3, 7, 14, 30 days) on last month’s data and pick the one that best correlates with this month’s actual conversions or CAC. Deliverable: a one-page summary with charts and your recommendation.
• Project C: Create a dashboard visualizing channel credit under last-click vs time-decay (h=7). Deliverable: side-by-side comparison and a 3-bullet stakeholder takeaway.

Quick Test — how it works

Everyone can take the test below; only logged-in users get saved progress.

Mini challenge

Your median time-to-conversion is 14 days. You want a time-decay model that still values mid-funnel touches but clearly favors recency. What half-life would you start with, and why?

Learning path

• Before: UTM taxonomy and channel mapping, conversion tracking, session stitching.
• This lesson: Time Decay Attribution (setup, math, QA, reporting).
• After: Position-based models, data-driven/algorithmic attribution (e.g., Markov, Shapley), and Marketing Mix Modeling for high-level budget planning.

Next steps

• Implement time-decay in your reporting with a documented half-life and lookback window.
• Share a short note explaining the model and how it differs from last-click.
• Run a 2-week experiment shifting 10–15% budget toward channels that gain credit under time-decay; monitor CAC/ROAS and conversion quality.