How a food delivery app partnered with MASS Analytics to run automated MMM updates every quarter, and now checks and adjusts its media plans between yearly rebuilds.
- At a glanceSector
- Food delivery, direct to consumer
- Scope
- One food delivery app, quarterly model rescoring and a yearly model update
- Engagement
- Ongoing
Always-ON measurement, delivered on the MassTer platform.
The challenge
A food delivery app serves a fast-moving audience, so its marketing team wanted to make more tactical decisions: check whether a media plan worked, then adjust the next one. A marketing mix model can support that, but only while it stays current. Once the team built its models, each update meant redoing the whole job, because its data came from several sources that each needed their own pipeline. This case study therefore shows how the app and MASS Analytics moved to quarterly rescoring and a yearly update.
Every update meant starting again
Before the change, the app’s models followed the traditional approach. When the team wanted fresh numbers, the work began again from the start, which takes time and effort. As a result, the longer the wait, the older the read on media performance, and the less the team could use it for tactical decisions.
“We wanted to judge our media plans between yearly updates, and adjust them while they were still running.”
A member of the app’s marketing team
Two problems behind the slow updates
- Updates meant starting again. After the team built the models, it could not simply refresh them. Instead, each update meant redoing the whole job, which cost time and effort every time.
- Every data source needed its own pipeline. The app’s data came from several sources, so the team had to build data pipelines again and again to make each one ready for modeling.
MASS Analytics answered both problems with two routines on one timeline, shown below. Each period lasts 3 months.
| Period | Routine | What happened | What it gave |
|---|---|---|---|
| Period 1 | Build | MASS Analytics built the model and set up the data process. | A working model and a reusable data process |
| Periods 2 to 4 | Rescoring | Each quarter, MASS Analytics rescored the model on new data. | A mid-term plan each quarter, plus a check on the last one |
| Period 5 | Update | MASS Analytics rebuilt the model for new variables, and proxies estimated their impact. | A refreshed structure for strategic planning |
| Period 6 | Rescoring | Quarterly rescoring resumed on new data. | Tactical planning back on a quarterly cycle |
Table 1: Three models replaced one. Paid media, pricing, competition and seasonality stayed in the specification with standard MMM treatment.
The solution
MASS Analytics proposed two ways to keep the app’s models current, each built for a different kind of decision. First, a rescoring every quarter would refresh the model on new data, for mid-term planning and tactical decisions. Second, a model update once a year would revisit the coefficients and structure, for strategic planning. Finally, a set of data pipelines, defined once and reused for every update, would take over the repeated data work that made updating slow.
Two routines for two kinds of decision
Rescoring is the quarterly routine. The MMM model takes in the latest data and estimates the effect of the media activations that ran since the last check. As a result, the team gets fresh results for campaign planning, and it uses each rescoring to see how earlier plans performed, which shapes the plan after it. Meanwhile, the yearly update is the strategic routine. Here, MASS Analytics prepares the data again, but this time it rebuilds the model itself to account for the changes the app made during the year. It adds new variables, and proxies estimate their impact.
A data process built once, then reused
The first build took the most data work, so the data engineers spent time learning the app’s data structure and sources, and they defined the transformations its data needed. That upfront time paid off because it gave the team the understanding to prepare data quickly for every later rescoring, and it made the setup ready to scale. Since the pipelines automated the data step, automated MMM updates became practical every quarter.
Results and Impact
The app now plans its media in quarterly cycles, and it checks each plan once it has run. In addition, its revenue forecasts matched actual revenue at 89% accuracy, the clearest sign that the model keeps pace.
89%
forecast accuracy: predicted revenue against actual revenue
3 months
rescoring cycle for mid-term plans
1 year
full model update for strategy
A plan checked every quarter
With regular model maintenance in place, marketing performance measurement became a quarterly habit: the app builds mid-term media plans, and it promptly checks how they ran. Those findings then become the base for fine-tuning tactical media strategy at the next rescoring. As a result, budget management became adaptive and responsive: the team plans, checks and adjusts each quarter, and the yearly update keeps the model’s structure in step with the business.
Forecasts that matched what happened
Because the team rescored the MMM model often, MASS Analytics could compare what it predicted with what actually happened. In fact, predicted revenue matched the revenue the app actually realized at 89% accuracy. However, a gap remained, and it usually came from outside the model: time between a suggested plan and its execution, last-minute budget changes after delivery of results, and competitor activity.
Want quarterly updates without the full rebuild?
MASS Analytics can set up the same cycle for your models: quarterly rescoring, a yearly update, and data pipelines that make both quick. Talk to our team.
Book a Demo Read the MMM Guide
