Time-Decay Attribution: Credit Recent Touchpoints More

What Time-Decay Attribution Measures

Time-decay attribution answers: "How much did each touchpoint contribute, considering that recent touches probably mattered more?"

It's a compromise between linear (equal credit) and last-touch (all credit to final touch):

The touchpoint on the same day gets the most credit. The touchpoint from 13 days ago gets the least—but it still gets some credit, unlike last-touch.

The Half-Life Concept

Time-decay uses exponential decay with a half-life parameter. At one half-life before conversion, a touchpoint receives 50% of the credit it would get if it happened at conversion time.

Credit decays quickly. A touchpoint from 28 days ago gets only 6% of the weight of a same-day touchpoint.

When Time-Decay Is the Right Choice

1. E-Commerce and Retail

E-commerce purchases often follow a pattern: browse, consider, return, buy. The recent touches—the retargeting ad, the email reminder, the price drop notification—often tip the decision.

2. Short-to-Medium Sales Cycles

For products bought within 2-4 weeks, time-decay captures the intensifying engagement pattern:

Match your half-life to your typical consideration period.

3. Urgency and Seasonality

Products with urgency (events, limited offers, seasonal items) see compressed decision windows. Recent touches matter more:

• Concert tickets: Decision made days before purchase
• Black Friday deals: Hours matter, not days
• Tax services: Last week of tax season drives conversion

Time-decay with a short half-life (3-5 days) captures these dynamics.

4. When You Want More Nuance Than Last-Touch

Time-decay acknowledges that early touches contributed—they just mattered less than recent ones:

Time-decay sits between the extremes, providing a reasonable middle ground.

When Time-Decay Is the Wrong Choice

1. Long B2B Sales Cycles

With 90+ day sales cycles, time-decay aggressively underweights the touchpoints that started the journey:

The LinkedIn ad that sourced the deal gets nearly zero credit—which doesn't reflect its importance.

Fix: Either use a much longer half-life (30+ days) or switch to position-based for B2B.

2. When First-Touch Really Matters

For businesses where the introduction is critical—breaking into a new audience, first-time brand exposure—time-decay undervalues that moment.

3. Content-Heavy Journeys

If your funnel relies on progressive education (SaaS with whitepapers → webinars → demos), each content piece builds on the previous. Time-decay penalizes the foundational content.

How Time-Decay Works

The Math

Time-decay uses exponential decay:

Weight = 2^(-t / half_life)

Where:
- t = time between touchpoint and conversion (in days)
- half_life = your chosen parameter (commonly 7 days)

Then normalize weights so they sum to 1:

Credit = Weight / Sum(all weights)

Example Calculation

Journey: 3 touchpoints at Day 1, Day 7, and Day 14 (conversion)
Half-life: 7 days

Implementation

class TimeDecayAttribution
  def initialize(half_life_days: 7, lookback_days: 30)
    @half_life_days = half_life_days
    @lookback_days = lookback_days
  end

  def attribute(conversion)
    touchpoints = conversion.user.touchpoints
      .where("occurred_at >= ?", conversion.occurred_at - @lookback_days.days)
      .where("occurred_at <= ?", conversion.occurred_at)
      .order(:occurred_at)

    return [] if touchpoints.empty?

    # Calculate raw weights
    weights = touchpoints.map do |tp|
      days_before = (conversion.occurred_at.to_date - tp.occurred_at.to_date).to_i
      calculate_weight(days_before)
    end

    total_weight = weights.sum

    # Normalize and build results
    touchpoints.zip(weights).map do |touchpoint, weight|
      {
        channel: touchpoint.channel,
        source: touchpoint.source,
        medium: touchpoint.medium,
        campaign: touchpoint.campaign,
        credit: weight / total_weight,
        touchpoint_at: touchpoint.occurred_at
      }
    end
  end

  private

  def calculate_weight(days_before)
    2.0 ** (-days_before.to_f / @half_life_days)
  end
end

Choosing Your Half-Life

Validate your choice: Look at your time-to-conversion distribution. Set half-life so touchpoints within your typical buying window still get meaningful credit.

Comparing Time-Decay to Other Models

Time-Decay vs Last-Touch

When to prefer time-decay: When you want last-touch's recency emphasis without completely ignoring the journey.

Time-Decay vs Linear

When to prefer time-decay: E-commerce, short cycles, urgency products.
When to prefer linear: B2B, long cycles, education-heavy funnels.

Time-Decay vs Position-Based

When to prefer time-decay: Short cycles where introduction is less critical.
When to prefer position-based: B2B where both sourcing and closing matter.

Time-Decay in Practice

Visualizing the Decay Curve

Impact on Channel Credit

Here's how a typical e-commerce business sees credit shift between models:

Time-decay rebalances from pure last-touch but still favors conversion channels.

Setting Up Multiple Half-Lives

Some teams run time-decay with different half-lives for different analyses:

Compare results to understand how half-life assumptions affect credit.

Common Time-Decay Mistakes

Mistake 1: Using Default Half-Life Without Analysis

The 7-day default may not match your business. If your sales cycle is 30 days, 7-day half-life crushes early-touch credit.

Fix: Analyze your time-to-conversion distribution. Set half-life to match your business reality.

Mistake 2: Forgetting About Lookback Window

Time-decay still needs a lookback window. Too short = missing early touches. Too long = including irrelevant ancient touches.

Fix: Set lookback to 2-3x your typical sales cycle. A 14-day cycle should use 30-45 day lookback.

Mistake 3: Assuming Recency = Causation

Just because email was recent doesn't mean it caused the purchase. The user might have already decided; email just reminded them.

Fix: Validate time-decay with incrementality tests. Does email's attributed credit match its actual incremental impact?

Mistake 4: Using Same Half-Life for All Segments

New customers vs returning customers have different journey patterns. Applying one half-life to both distorts credit.

Fix: Consider segment-specific half-lives or analyze segments separately.

Time-Decay and GA4

Google Analytics 4 removed time-decay attribution in 2023. Your options:

1. Third-Party Attribution Tools

Use mbuzz or similar tools that support time-decay with configurable half-life.

2. Build in Your Data Warehouse

Export data to BigQuery/Snowflake and implement time-decay:

WITH touchpoints_weighted AS (
  SELECT
    user_id,
    conversion_id,
    channel,
    conversion_value,
    touchpoint_time,
    conversion_time,
    -- Calculate weight using 7-day half-life
    POWER(2, -DATE_DIFF(conversion_time, touchpoint_time, DAY) / 7.0) as weight
  FROM touchpoint_data
),
normalized AS (
  SELECT
    *,
    weight / SUM(weight) OVER (PARTITION BY conversion_id) as credit
  FROM touchpoints_weighted
)
SELECT
  channel,
  SUM(credit * conversion_value) as attributed_revenue
FROM normalized
GROUP BY channel
ORDER BY attributed_revenue DESC;

Implementing Time-Decay in mbuzz

mbuzz uses AML (Attribution Modeling Language) — a small Ruby DSL — to define how credit is distributed. The time_decay helper takes a half_life argument and normalizes credits to sum to 1.0 across all touchpoints in the window.

Basic time-decay

within_window 30.days
  time_decay half_life: 7.days
end

A 30-day lookback with credit halving every 7 days. The most recent touchpoint gets the highest weight; a touchpoint 7 days older gets half; 14 days older, a quarter; and so on.

Tuning the half-life

Change the half-life to match your sales cycle:

# E-commerce: fast decay
within_window 14.days
  time_decay half_life: 3.days
end

# B2B SaaS: slow decay, don't crush early touches
within_window 90.days
  time_decay half_life: 21.days
end

The full AML reference — including segment weights, conditional logic, and channel filtering — is at docs/attribution-models.

Tuning Time-Decay for Your Business

Time-decay has more levers than linear. Here's how to tune them.

Half-Life by Business Type

In AML, both levers are arguments: within_window for the lookback and half_life: on time_decay.

Seasonal adjustments

Buying behavior compresses during high-intent periods. For BFCM week, run a parallel model with a 2-day half-life and a 7-day window — credit lands on the immediate pre-purchase touches. For the post-holiday gift card window, stretch the lookback to 45 days and the half-life to 10–14 days so original holiday campaigns still get credit when redemptions land in January.

Campaign launch adjustments

For a launch attribution view, narrow the lookback and accelerate the decay so credit lands on the immediate pre-conversion touches:

within_window 14.days
  time_decay half_life: 3.days
end

To restrict the model to a single launch campaign, or to vary half-life by conversion value, see the channel-filtering and conditional-logic patterns at docs/attribution-models.

Parameter Tuning Cheatsheet

Summary

Time-decay attribution credits touchpoints based on recency—recent touches get more credit, older touches get less. It's a middle ground between last-touch (all credit to final touch) and linear (equal credit).

Use time-decay when:
- Short to medium sales cycles (< 30 days)
- E-commerce and retail where recency matters
- You want last-touch's emphasis without ignoring the journey
- Products with urgency or seasonality

Don't use time-decay when:
- Long B2B sales cycles (it crushes early-touch credit)
- First-touch genuinely matters for sourcing
- Content-heavy journeys where early education matters

Best practice: Start with a 7-day half-life, then adjust based on your conversion cycle. Validate with incrementality tests to confirm recency actually correlates with influence.

Further Reading

On Attribution Models:
- Linear Attribution — The neutral baseline
- Position-Based Attribution — Emphasize first and last
- How to Choose the Right Attribution Model — Decision framework

On Validation:
- MTA vs MMM — Where attribution fits in the measurement stack
- Triangulating Measurement Methods — Validating with multiple approaches

Holly Henderson

Co-Founder, mbuzz

Holly Henderson is Co-Founder of mbuzz. With 10+ years in marketing including roles at Westpac, Avon, and Forebrite, she's obsessed with making measurement actually useful.

Harvard Extension School Forebrite Westpac Avon