Impact Calculation for Top Contributors

How a change reason contributor influences a target variable

What is an Impact Score?

Impact Scores in Tellius spotlight the dimensions that genuinely move your metric—scaled for size, corrected for noise, and tuned to the exact comparison you’re running—so you can act on what truly matters.

When Tellius highlights a “Top Contributor”, it is answering:

“If this contributor hadn’t changed, by how much would the overall metric still have moved?”

Impact scales each contributor’s movement against the overall movement, so small but volatile segments don’t eclipse large steady movers.

% = percent of overall change (used for numeric metrics) pp = percentage-points (used for rates or share comparisons)

How the calculation adapts to your analysis?

Tellius looks at two things before it computes the score:

• Insight type – either Trend Drivers (comparing two time periods) or Cohort Drivers (comparing two groups).

• Target metric – either a numeric total (revenue, units), a rate or percentage (win‑rate, approval‑rate), or market share.

The logic is then:

• Numeric totals (Trend or Cohort)

  1. Work out the contributor’s change (or difference) – let's call this ΔC.

  2. Work out the overall change (or difference) – let's call this ΔO.

  3. Impact percent = (ΔC ÷ ΔO) × 100. A value of 50 % means the contributor explains half of the total movement.

• Rates and percentages (Trend or Cohort)

  1. In each period or group, take the contributor’s rate minus the rate for everything else – that gives you a gap G₁ and G₂.

  2. Impact is the gap change: G₂ minus G₁, expressed in percentage‑points. A positive value means the contributor pushed the overall rate upward.

• Market share

  1. Compute the contributor’s share in each period or group.

  2. Impact is simply the change in share, again in percentage‑points. This works whether you’re looking at overall share or a share inside a filter such as “Shoes in California.”

This same logic covers special cases like 4‑week vs 5‑week retail calendars because the calculation looks only at the periods or cohorts you’ve selected.

Why the score is statistically trustworthy

• Population weighting – Every contributor’s impact is multiplied by how much data it represents, so two approvals in Iowa never outweigh thousands in California.

• Significance test – Tellius treats the raw impact values as a distribution and only surfaces contributors that lie beyond a default 90 % confidence threshold.

• False‑discovery control – When you examine dozens of dimensions at once, a Benjamini–Hochberg procedure keeps the false‑positive rate below 5 %.

Reading the score

• Around +100 % – this contributor alone accounts for the full rise; celebrate or double‑down.

• Around 0 % – it moved in line with the pack; nothing special to act upon.

• Around ‑100 % – it fell while the total rose (or vice‑versa); investigate the drag.

Example 1: Trend Driver on a numeric total

Metric: total revenue. Periods compared: Week 5 vs Week 6.

• What happened overall? Revenue rose from $100 M to $130 M, a lift of $30 M.

• What happened for the contributor “Office Supplies”? It rose from $25 M to $40 M, a lift of $15 M.

• Impact calculation: $15 M (contributor change) divided by $30 M (overall change) × 100 = 50 %.

• Read‑out: Office Supplies explains half of the entire revenue growth in that period.

Example 2: Trend Driver on a rate or percentage

Metric: approval‑rate for Home Loans. Periods compared: Week 5 vs Week 6.

• Home‑Loan approval improved from 6.5 % to 7.4 %.

• Approval for all other loans improved from 45.4 % to 49.6 %.

• Gap in Week 5 = 6.5 % – 45.4 % = ‑38.9 pp (percentage‑points).

• Gap in Week 6 = 7.4 % – 49.6 % = ‑42.2 pp.

• Impact = change in the gap = (‑42.2 pp) – (‑38.9 pp) = ‑3.3 pp.

• Read‑out: the approval gap versus other loans widened by 3.3 points, pulling the overall rate down slightly.

Example 3: Trend Driver on market share

Metric: share of Office Supplies in overall sales. Periods compared: Week 5 vs Week 6.

• Week 5 share = $25 M ÷ $100 M = 25 %.

• Week 6 share = $50 M ÷ $150 M = 33 %.

• Impact = 33 % – 25 % = +8 percentage‑points.

• Read‑out: Office Supplies grabbed eight points of extra share week‑over‑week.

Example 4: Cohort Driver on a numeric total

Metric: total Office‑Supplies sales. Cohorts compared: Chicago vs NYC in the same week.

• Chicago sells $25 M of Office Supplies; NYC sells $50 M.

• Overall sales difference between cities is $100 M (Chicago) vs $150 M (NYC) → $50 M gap.

• Contributor difference is $25 M.

• Impact = $25 M ÷ $50 M × 100 = 50 %.

• Read‑out: half of the sales gap between the two cities comes from Office Supplies.

Example 5: Cohort Driver on a rate or percentage

Metric: approval‑rate for Home Loans. Cohorts compared: Chicago vs NYC.

• Home‑Loan approval: 6.5 % in Chicago, 7.4 % in NYC.

• All‑other loans: 45.4 % in Chicago, 49.6 % in NYC.

• Gap in Chicago = 6.5 % – 45.4 % = ‑38.9 pp.

• Gap in NYC = 7.4 % – 49.6 % = ‑42.2 pp.

• Impact = –42.2 pp – (–38.9 pp) = ‑3.3 pp.

• Read‑out: Home‑Loan approval lags other loans slightly more in NYC than in Chicago.

Example 6: Cohort Driver on market share

Metric: market share of Nike within California. Sub‑brands compared: Nike vs Reebok.

• Nike sells $25 M of shoes; Reebok sells $50 M. Total shoe sales in CA = $100 M.

• Nike share = 25 %; Reebok share = 50 %.

• Impact for Nike = 25 pp below Reebok.

• Read‑out: Nike trails Reebok by twenty‑five share points in California.

In every case, a contributor’s change is measured in relation to the overall change—and when you're working with rates, the calculation also adjusts for how the rest of the data behaves. This scaling is what makes Impact a reliable signal, not just a raw difference.