Choose a statistical comparison that answers a marketing decision. Work from account counts to conversion rates, segment comparisons, sample checks and contribution using a reconciled campaign example.
Statistical analysis in marketing starts by comparing outcomes for consistently defined groups, with the numerator, denominator and observation window visible. Choose the method around the budget or campaign decision, then check whether audience mix explains the apparent difference. Descriptive results show what happened; evidence of incremental effect requires a suitable study design.
Write the question in business terms: should the team expand a campaign, change its audience or investigate a weak segment? Then choose the smallest comparison that can answer it.
| Decision | Useful starting method | What must accompany the result |
|---|---|---|
| Describe campaign outcomes | Counts, rates and dollar totals | Eligible population and outcome definition |
| Compare audiences | Rates within consistent segments | Segment counts and the total audience mix |
| Compare launch periods | Cohort comparison at equal age | Start date, cutoff and follow-up window |
| Estimate incremental effect | A planned experimental comparison | Assignment method, uncertainty and cost |
Use marketing funnel analysis for stage movement and marketing cohort analysis for equal-age customer groups. Those answer different questions from a broad campaign average.
An experiment needs decisions made before results arrive. NIST's explanation of completely randomized designs describes assigning factor levels randomly to experimental units. In marketing, that means a deliberate assignment process; labeling two existing campaigns A and B does not make their comparison randomized.
For this example, a US business supplies campaign assignments, customer history, orders and campaign expenses. The comparison unit is an eligible account. Count an account as converted if it places a qualifying order within the agreed follow-up window, even if it places several orders.
Retain account ID, campaign ID, new-or-returning status at assignment, assignment date, outcome cutoff, conversion flag, net sales and variable delivery cost. Keep campaign expenses in a separate register until they have been allocated once. Otherwise, joining an expense row to multiple orders can multiply the cost.
Resolve accounts assigned to both campaigns before producing the comparison. Keep nonbuyers in the denominator, and distinguish missing order information from confirmed nonconversion. Finance should confirm which costs belong in contribution; marketing operations should confirm the assignment population and cutoff.
The following table summarizes that prepared file. Campaign costs are allocated to the displayed groups, and all sales are net of recorded returns. This is an illustrative company example, not a performance benchmark.
| Campaign | Account segment | Eligible accounts | Converted accounts | Net sales (USD) | Variable cost (USD) | Campaign cost (USD) |
|---|---|---|---|---|---|---|
| A | New | 900 | 27 | 5,400 | 3,240 | 900 |
| A | Returning | 100 | 15 | 2,400 | 1,200 | 100 |
| B | New | 100 | 3 | 600 | 360 | 100 |
| B | Returning | 900 | 135 | 21,600 | 10,800 | 900 |
| Total | 2,000 | 180 | 30,000 | 15,600 | 2,000 |
With the table's columns in A:G and its data in rows 2:5, put conversion rate in H2 and contribution after campaign cost in I2. Fill both down:
=IF(C2=0,"",D2/C2)
=E2-F2-G2
Format rates as percentages. A blank rate means there is no eligible population; it does not mean a measured zero conversion rate.
| Campaign and segment | Conversion calculation | Conversion rate | Contribution after campaign cost (USD) |
|---|---|---|---|
| A, new | 27 / 900 | 3.0% | 1,260 |
| A, returning | 15 / 100 | 15.0% | 1,100 |
| B, new | 3 / 100 | 3.0% | 140 |
| B, returning | 135 / 900 | 15.0% | 9,900 |
| Total | 180 / 2,000 | 9.0% | 12,400 |
The dollar check is 30,000 minus 15,600 minus 2,000 = $12,400. Sum the account counts back to the assignment export, sales and variable costs back to the matched order file, and campaign costs back to the expense register.
The overall rate uses totals:
=SUM(D2:D5)/SUM(C2:C5)
Keep customer acquisition cost separate when the decision specifically concerns new customers. This example includes returning accounts, so its full expense denominator cannot automatically become an acquisition measure.
Aggregate results seem to favor B: A converts 42 of 1,000 accounts, or 4.2%; B converts 138 of 1,000, or 13.8%. The gap is 9.6 percentage points. Yet both campaigns convert new accounts at 3.0% and returning accounts at 15.0%.
The difference comes entirely from the displayed audience mix. B has more returning accounts. To illustrate that effect, apply each campaign's segment rates to an explicitly hypothetical audience split equally between new and returning accounts:
=0.5*H2+0.5*H3
=0.5*H4+0.5*H5
Both return 9.0%. These standardized figures describe the chosen mix; they are not additional observed conversions or predictions. B's realized contribution after campaign cost is still $10,040, versus A's $2,360, because its observed population bought more. That financial result does not establish superior campaign effectiveness.
Show outcome counts beside rates. B's new-account result contains only three conversions. One additional conversion would move that segment from 3.0% to 4.0%, a full percentage point. One additional conversion among A's new accounts would move its rate by about 0.11 percentage point, rounded to two decimals.
Neither sensitivity calculation is a confidence interval. NIST's guidance on proportion confidence intervals explains why simple normal approximations can be inaccurate with small samples or few outcomes. Ask an analyst to choose an interval or test suited to the design before describing a difference as statistically reliable. An interval also cannot correct biased campaign assignment.
For a planned comparison, use the dedicated marketing A/B test analysis workflow. Decide what difference would change the business action before repeatedly inspecting the results.
Upload your prepared campaign, customer and sales files; Covirage maps the columns, and its tools compute and check the group counts, rates and contribution. The external AI model explains the results with every figure traced to its rows, so the audience mix stays visible beside the outcome. Create a free Covirage account and bring a sample of your own data.
See sales insights for the supporting business context.
Start with descriptive counts and rates for the decision at hand. Split results by a relevant customer characteristic, such as new versus returning accounts. More advanced methods cannot repair duplicate records, missing outcomes or groups measured over different follow-up periods.
There is no universal minimum. Required sample size depends on the baseline outcome rate, the difference worth detecting and the uncertainty you can accept. Report both the audience count and the outcome count, and plan an experiment's sample requirement before collecting results.
A descriptive report establishes what happened within its recorded population. Campaign exposure may also reflect customer intent, sales attention or prior purchasing. Establishing incremental impact requires an appropriate experimental or causal design, rather than simply comparing customers who received different campaigns.
Use conversion to understand the frequency of the defined outcome, then connect it to contribution and campaign cost. Different order values, discounts and fulfillment costs can make a higher conversion rate less attractive financially. Keep both measures attached to the same customer population.