Simple Linear Regression: Modelling Revenue from Ad Spend
linear regression, OLS, ad spend, revenue, R-squared, slope interpretation
TL;DR: Revenue = ₹77.9k + ₹0.31 per ₹1 ad spend (n=200 weeks), slope CI [₹0.16, ₹0.45], R² = 0.08 (weak fit), residuals look normal. Decision: the association is real but small. Each ₹1k of ad spend is associated with about ₹306 more revenue, less than the spend itself.
Answer
Method: Simple linear regression (OLS).
Estimate: ₹0.31 per ₹1 and CI ₹0.16, ₹0.45.
Data: Marketing campaign, variables ad_spend, revenue, n = 200.
Action: Each ₹1k of ad spend yields about ₹306 of revenue. Review the budget before adding spend.
Case
You manage digital advertising for an e-commerce platform and want to quantify the relationship between weekly ad spend and revenue. Specifically: For every additional rupee spent on ads, how much revenue can you expect? Is the relationship strong enough to justify continued investment? What portion of revenue variation is explained by ad spend alone?
Dataset
Synthetic sample from weekly marketing data (Schema A).
| Variable | Label | Value |
|---|---|---|
spend |
Weekly ad spend | ₹ (rupees) |
revenue |
Weekly revenue | ₹ (rupees) |
n |
Number of weeks | 200 |
| Range | Spend range | ₹0 to ₹100,000 |
Method
We use ordinary least squares (OLS) regression to model revenue as a linear function of ad spend. The regression equation is:
\[ \text{Revenue}_i = \beta_0 + \beta_1 \times \text{Spend}_i + \epsilon_i, \]
where \(\beta_0\) is the intercept (baseline revenue when spend = 0), \(\beta_1\) is the slope (the revenue difference associated with 1 more rupee of spend), and \(\epsilon_i\) is the error term assumed to be normally distributed with constant variance.
The slope estimate with standard error: \[ \hat{\beta}_1 = \frac{\sum(x_i - \bar{x})(y_i - \bar{y})}{\sum(x_i - \bar{x})^2}, \quad SE(\hat{\beta}_1) = \frac{s}{\sqrt{\sum(x_i - \bar{x})^2}}, \] where \(s\) is the residual standard error.
The coefficient of determination (R-squared): \[ R^2 = 1 - \frac{SS_{\text{res}}}{SS_{\text{tot}}} = 1 - \frac{\sum(y_i - \hat{y}_i)^2}{\sum(y_i - \bar{y})^2}. \]
We report 95% confidence intervals for both intercept and slope using standard t-distribution critical values.
Calculation
Visualisation
Results and Interpretation
The regression analysis reveals a statistically significant positive relationship between ad spend and revenue. The estimated slope is ₹0.306 (95% CI [₹0.164, ₹0.447]), meaning that for every additional rupee spent on advertising, revenue increases by about ₹0.31 on average, b = 0.306, SE = 0.072, t(198) = 4.26, p < .001 (R Core Team 2024).
The intercept of ₹77,899 (95% CI [₹69,517, ₹86,281]) is the estimated baseline revenue when ad spend is zero. Read it with care, since the model describes the observed spend range of ₹0 to ₹100,000 and nothing beyond it.
Model fit: The R² value of 0.084 (8%) indicates that ad spend explains about 8% of the variation in weekly revenue (adjusted R² = 0.079). The other 92% comes from factors outside this simple model, such as seasonality, competitor actions, organic traffic and product mix. The residual standard error of ₹29,674 shows how widely weekly revenue scatters around the line.
Statistical significance: The F-statistic, F(1, 198) = 18.12, p < .001, confirms the model is significant. The relationship is real but small.
Practical significance: Weeks with 1 extra rupee of spend show about ₹0.31 more revenue, and the 95% CI [₹0.16, ₹0.45] stays below ₹1. Even read as a causal return, which these observational data cannot establish, extra ad spend would bring back less revenue than it costs, before any other costs are subtracted.
Decision framework. The relationship is significant, yet the marginal return falls short of the spend, and association alone is weak evidence of cause. Before making budget decisions: (1) Check residual diagnostics for violations of regression assumptions, (2) Consider confounding variables (e.g., seasonality, external events), (3) Calculate net profit impact after subtracting ad costs and other variable costs, (4) Test whether the relationship holds at different spend levels (non-linearity), (5) Consider building a multiple regression model with additional predictors for better prediction accuracy.
Residual Diagnostics
Diagnostic interpretation: The residuals vs. fitted plot should show random scatter around zero (no pattern). The Q-Q plot should show points following the diagonal line. Deviations suggest violations of assumptions (heteroscedasticity, non-normality or non-linearity).
Assumptions
Simple linear regression assumes:
- Linearity: The relationship between spend and revenue is linear (check with scatterplot and residual plot)
- Independence: Weekly observations are independent (no autocorrelation across time)
- Homoscedasticity: Constant variance of residuals across all levels of spend (check residuals vs. fitted plot)
- Normality of residuals: Residuals are approximately normally distributed (check the Q-Q plot, less critical with n = 200)
- Accurate predictor: Ad spend is measured accurately
- Unconfounded predictor: Other important predictors are uncorrelated with spend
Limitations
This analysis leaves out:
- Causality: Correlation stops short of proving that ad spend causes revenue. Reverse causality is possible (higher revenue allows more ad spend)
- Confounding variables: Seasonality, promotions, competitor actions, product launches, organic trends
- Non-linear relationships: Diminishing returns at high spend levels, or threshold effects at low spend
- Time series structure: Temporal autocorrelation, lagged effects (ads may influence future weeks)
- Interaction effects: Ad effectiveness may vary by channel, audience segment or time period
- Costs: This models revenue, not profit. Must subtract ad costs and variable costs to assess true ROI
Recommendations for improvement: - Add time trends and seasonal dummies to control for temporal patterns - Use multiple regression with additional predictors (organic traffic, email campaigns, competitor spend) - Test for non-linearity with polynomial terms or segmented regression - Consider time series models (ARIMA, VAR) if temporal dependencies exist - Conduct controlled experiments (randomized ad spend levels) to establish causality - Calculate contribution margin and customer lifetime value for profit-focused analysis
Use the below format to cite this page
Sharafuddin, M. A. (2025, October 25). Simple linear regression: Modelling revenue from ad spend. Flair Marketing Intelligence (FlairMI). https://flairmi.com/blog/posts/05-simple-regression.html
@online{sharafuddin2025-regression,
author = {Sharafuddin, Mohammed Ali},
title = {Simple Linear Regression: Modelling Revenue from Ad Spend},
year = {2025},
date = {2025-10-25},
url = {https://flairmi.com/blog/posts/05-simple-regression.html},
langid = {en}
}
References
Citation
@online{ali_sharafuddin2025,
author = {Ali Sharafuddin, Mohammed},
title = {Simple {Linear} {Regression:} {Modelling} {Revenue} from {Ad}
{Spend}},
date = {2025-10-25},
url = {https://flairmi.com/blog/posts/05-simple-regression.html},
langid = {en}
}
Comments