The marketing mix model explains 92.6% of sales variance across 200 market observations (R² = 0.9263, Adjusted R² = 0.9251), with an overall fit that is statistically significant (F = 820.621, p = 0). The highest marginal ROI channel is radio_spend, delivering approximately 0.204 incremental sales units per $1,000 of additional spend. Shifting 10K from the lowest-ROI channel to radio_spend is expected to generate a net sales lift of 1.99 thousand units.

Each bar shows the estimated incremental sales lift (in thousands of units) per $1,000 increase in that channel's budget, controlling for all other channels. The highest-coefficient channel is radio_spend (estimate = 0.2041), meaning each additional $1,000 of radio_spend spend drives roughly 0.2041 thousand units of incremental sales. 2 of 3 channels have a statistically significant effect (p < 0.05); channels whose confidence intervals cross zero have uncertain or negligible incremental impact.

Marginal ROI is expressed as incremental sales units per dollar of additional spend at current budget levels. radio_spend leads with 0.204 sales units per $1,000 spent, while newspaper_spend delivers 0.005 units per $1,000. The top channel outperforms the bottom by a factor of 40.8x, suggesting that reallocating budget toward radio_spend would increase total sales without changing total spend.

Each curve traces predicted sales as one channel's spend increases from zero to its observed maximum, holding all other channels at their mean spend. In a linear model the slope of each curve equals the channel's regression coefficient, making steeper slopes directly comparable to marginal ROI. The channel with the steepest response curve is radio_spend — additional spend in this channel generates the largest incremental sales per dollar. Channels with flat curves (near-zero slope) contribute little marginal lift and may warrant budget cuts.

This table shows the recommended budget reallocation: shift $10K from newspaper_spend (lowest marginal ROI) to radio_spend (highest marginal ROI), keeping total spend constant. The expected net sales lift from this reallocation is 1.99 thousand units — achievable with no additional budget. Channels with positive spend_change receive the incremental budget; channels with negative spend_change are the source of the shift.

Baseline sales (36%) represent the sales volume that would exist even with zero advertising spend — driven by brand equity, word-of-mouth, and other non-modelled factors. The largest advertising contributor is tv_spend at 37.1% of total predicted sales. Channels below 5% of predicted sales have limited influence and may be candidates for budget reallocation toward higher-contributing channels.

Each point represents one market period, comparing the model's predicted sales to the actual observed sales. Points that cluster tightly along the 45-degree diagonal indicate accurate predictions across all budget levels. The model explains 92.6% of sales variance (R² = 0.9263) with an average prediction error of 1.39 thousand units (RMSE). Systematic deviations from the diagonal — such as consistent over-prediction at high spend levels — would suggest non-linear channel responses not captured by the linear model.