When a multi-location beverage retailer set out to understand what was actually driving delivery sales across its five locations, leaders recognized that a single blended report — one number covering two different marketing tactics — was no longer enough to guide spend decisions store by store. That recognition led to a marketing-mix-modeling pilot, built directly on the brand's own delivery-platform data.
The Opportunity
Two tactics, one number
Before the pilot, the brand ran two marketing tactics in parallel at each of its five delivery locations: Promotional Offers and Delivery Discounts. Platform reporting captured order volume and revenue at every store, but combined the effect of both tactics into a single figure, regardless of which one was actually responsible for the result — and regardless of the fact that the two tactics carried very different costs. Delivery Discounts cost the brand relatively little per order to run; Promotional Offers cost meaningfully more. Standard reporting treated their contributions identically.
This made it impossible to answer a basic budgeting question: at each individual store, was a dollar spent on Promotional Offers returning more or less than a dollar spent on Delivery Discounts? Leaders recognized that allocating marketing spend off a blended, store-agnostic number was no longer sufficient once the two tactics diverged this sharply in cost. What the brand needed was a way to isolate each tactic's true incremental effect, location by location.
The Solution
Two models, not one
The brand worked with Vizuro, whose marketing-mix-modeling platform ingested the pilot's delivery-platform order, promotion, and delivery-discount data directly, without requiring manual data exports from restaurant partners.
The analysis used two separate statistical models — Sales ($) and Order Volume — so each tactic's effect on order counts could be measured separately from its effect on how much customers spent, while controlling for seasonality and store-level differences in baseline order volume and average order value. Both models achieved a moderate, comparable fit — consistent with the natural noise of daily multi-store retail data.
The output is a location-by-location, tactic-by-tactic ROI score: how many additional orders each tactic generates per day, how much each adds to average order value, and what return each dollar of spend produces at each store. The platform is designed to refresh these scores automatically as new order, promotion, and delivery data arrives, so each spend decision can draw on current figures without a new analysis cycle.
The platform is designed to support the marketing team in three key areas:
Reduced Data Collection Burden: Direct, automated ingestion from the delivery platform means stores never need to manually export or prepare data for the model to run.
Store-Level Decision Support: Marketing teams get a location-by-location, tactic-by-tactic ROI score in place of a single blended report that can't tell them where a dollar works hardest.
Always-Current Recommendations: Scores refresh automatically as new order, promotion, and discount data arrives, so spend decisions are never based on stale numbers.
The Impact
A different answer at every store
Across the pilot locations, both tactics contributed to daily order volume, and the two separate models made it possible to see each tactic's effect on its own rather than as a single blended number. The analysis also showed that the two tactics influenced order volume and average order value in different ways.
On spend efficiency, a clearer pattern emerged: at most locations, one of the two tactics returned meaningfully more per dollar spent than the other, though the gap was not uniform across every store. The brand can now get a different, store-specific answer to whether a given tactic is worth the spend, rather than relying on one blended figure across the whole footprint.
Looking Ahead
Turning a store-level lens into a standing part of how the brand plans spend.
Looking ahead, the brand is positioned to extend the same two-model approach to additional tactics, channels, and locations as delivery data continues to accumulate. Rather than a one-time analysis, the ROI scoring is designed to run continuously, sharpening with every new order, promotion, and discount cycle. The ambition extends beyond this pilot: to make store-level, tactic-by-tactic ROI a standard input in how the brand plans its marketing budget, quarter after quarter.