How to Change Your Attribution Lookback Window

What Lookback Windows Do

A lookback window defines how far back in time your attribution model searches for touchpoints to credit.

The window determines which touchpoints are "in scope" for attribution. Touchpoints outside the window get zero credit—they're treated as if they never happened.

Recommended Windows by Business Type

Rule of thumb: Set your window to capture 90% of conversions based on time from first touchpoint.

How to Analyze Your Conversion Timing

Step 1: Calculate Time-to-Conversion Distribution

-- Time from first touchpoint to conversion
SELECT
  NTILE(100) OVER (ORDER BY days_to_convert) AS percentile,
  days_to_convert
FROM (
  SELECT
    conversion_id,
    EXTRACT(DAY FROM conversion_at - first_touchpoint_at) AS days_to_convert
  FROM conversions
  WHERE conversion_at >= CURRENT_DATE - INTERVAL '90 days'
) t;

Example Distribution

A 30-day window captures 90%+ of conversions for this distribution.

Step 2: Compare Credit Distribution at Different Windows

See how channel credit changes with different windows:

-- Compare attribution at 7, 30, 60 day windows
WITH conversions AS (
  SELECT * FROM conversions WHERE conversion_at >= CURRENT_DATE - 90
)
SELECT
  channel,
  SUM(CASE WHEN touchpoint_age_days <= 7 THEN credit ELSE 0 END) AS window_7d,
  SUM(CASE WHEN touchpoint_age_days <= 30 THEN credit ELSE 0 END) AS window_30d,
  SUM(CASE WHEN touchpoint_age_days <= 60 THEN credit ELSE 0 END) AS window_60d
FROM attribution_details
JOIN conversions USING (conversion_id)
GROUP BY channel;

Example Comparison

Short windows over-credit closers (email, branded search). Longer windows give more credit to introducers (paid social, display).

The Trade-off: Short vs Long Windows

Short Windows (7-14 days)

Pros:
- Credit is concentrated on immediately relevant touchpoints
- Less noise from old, potentially irrelevant touches
- Matches platform defaults (easier to reconcile)
- Good for fast-cycle businesses

Cons:
- Misses early-funnel touchpoints
- Over-credits closers, under-credits introducers
- May not reflect true customer journey length
- First-touch data becomes unreliable

Best for: Impulse purchases, short sales cycles, within-channel optimization

Long Windows (60-90+ days)

Pros:
- Captures full customer journey
- Fair credit to early-funnel channels
- Better for B2B with long cycles
- First-touch data is accurate

Cons:
- May credit irrelevant old touchpoints
- Signal gets diluted across many touches
- Harder to act on (touches from 3 months ago)
- More complexity to analyze

Best for: B2B, enterprise sales, high-consideration purchases

Different Windows for Different Purposes

By Funnel Stage

This is the tiered approach from funnel stage attribution.

By Channel

Some tools support channel-specific windows:

Caveat: Channel-specific windows add complexity. Only use if you have clear evidence that channels operate on different timelines.

How to Test Window Changes

A/B Testing Windows

Don't change windows blindly—test first:

# Window A/B test design
def design_window_test
  {
    test_type: :holdout,
    duration: "6-8 weeks",

    treatment: {
      name: "New 45-day window",
      window_days: 45
    },
    control: {
      name: "Current 30-day window",
      window_days: 30
    },

    success_metrics: {
      primary: "Forecast accuracy (predicted vs actual revenue)",
      secondary: [
        "Channel ROAS stability",
        "Budget decision confidence"
      ]
    },

    sample: "Split by conversion_id modulo",
    analysis: "Compare attributed ROAS and forecast accuracy"
  }
end

What to Compare

When testing windows, measure:

-- Compare touchpoint coverage by window
SELECT
  lookback_window,
  COUNT(*) AS total_conversions,
  SUM(CASE WHEN has_touchpoints THEN 1 ELSE 0 END) AS with_touchpoints,
  ROUND(100.0 * SUM(CASE WHEN has_touchpoints THEN 1 ELSE 0 END) / COUNT(*), 1) AS coverage_pct,
  ROUND(AVG(touchpoint_count), 1) AS avg_touchpoints
FROM (
  SELECT
    c.conversion_id,
    lw.days AS lookback_window,
    COUNT(t.id) > 0 AS has_touchpoints,
    COUNT(t.id) AS touchpoint_count
  FROM conversions c
  CROSS JOIN (VALUES (7), (30), (60), (90)) AS lw(days)
  LEFT JOIN touchpoints t
    ON c.visitor_id = t.visitor_id
    AND t.occurred_at BETWEEN c.converted_at - INTERVAL '1 day' * lw.days AND c.converted_at
  GROUP BY c.conversion_id, lw.days
) t
GROUP BY lookback_window
ORDER BY lookback_window;

Example Test Results

Window A/B test results — 8 weeks

Recommendation: extend to a 45-day window — captures more first-touch data, better reflects true journey length, and improves forecast accuracy.

Common Window Mistakes

Mistake 1: Using Platform Defaults Blindly

Google Ads uses 30 days. Meta uses 7 days. Neither matches your actual customer journey—they're designed to favor their platform.

Fix: Analyze your own time-to-conversion data and set windows based on your business.

Mistake 2: Same Window for All Purposes

First-touch analysis needs longer windows than last-touch analysis.

Fix: Use tiered windows: longer for ToFU, shorter for BoFU.

Mistake 3: Never Reviewing Window Settings

Customer journeys change. New products, new channels, market shifts.

Fix: Review window settings annually or when making major business changes.

Mistake 4: Changing Windows Without Comparison

Switching from 30 to 60 days will drastically change your attribution data. If you don't compare side-by-side, you'll misinterpret the shift.

Fix: Run both windows in parallel for 4-6 weeks before switching.

Mistake 5: Chasing Window Optimization

Spending months finding the "perfect" 43-day window is not valuable. Windows are approximate—get in the right range and move on.

Fix: Use your business category default, validate it's reasonable, and focus on more impactful work.

Implementation

Configuring in Common Tools

Code Example

class Attribution::WindowConfiguration
  DEFAULTS = {
    impulse_ecommerce: 7,
    considered_ecommerce: 30,
    saas_mid_market: 60,
    b2b_enterprise: 90
  }.freeze

  def initialize(business_type:, custom_window: nil)
    @window_days = custom_window || DEFAULTS.fetch(business_type)
  end

  def filter_touchpoints(conversion)
    conversion.touchpoints.where(
      "occurred_at >= ?",
      conversion.occurred_at - @window_days.days
    )
  end

  # Test multiple windows on same data
  def compare_windows(conversion, windows: [7, 30, 60, 90])
    windows.map do |days|
      touchpoints = conversion.touchpoints.where(
        "occurred_at >= ?",
        conversion.occurred_at - days.days
      )

      {
        window_days: days,
        touchpoint_count: touchpoints.count,
        channels: touchpoints.pluck(:channel).uniq,
        first_touch_captured: touchpoints.any? { |t|
          t == conversion.touchpoints.order(:occurred_at).first
        }
      }
    end
  end
end

Summary

Attribution lookback windows determine which touchpoints get credit:

Setting your window:
1. Analyze your time-to-conversion distribution
2. Set window to capture 85-95% of conversions
3. Use industry defaults as starting points
4. Test changes with A/B experiments

Key principles:
- Match window to your sales cycle (2-3x median)
- Short windows favor closers; long windows favor introducers
- Consider different windows for different funnel stages
- Review annually or when business changes

Common windows:
- E-commerce: 7-30 days
- SaaS: 30-60 days
- B2B: 60-90+ days

Further Reading

On Attribution Configuration:
- How to Choose the Right Attribution Model — Model selection framework
- How to Use Different Attribution Models by Funnel Stage — Tiered windows approach

On Platform Settings:
- Google: Attribution lookback windows — GA4 configuration
- Meta: About attribution settings — Meta Ads configuration

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