Every step is a SAS program
Each step writes a SAS program from what the screen holds and shows it under "The program", so you can read exactly what will run.
Jenner Studio application · Version 0.4.0
An application for Jenner Studio, published by Jenner Analytics.
Marketing Mix takes your own sales, media, price and weather files to a budget plan in seven steps: the weekly data, the baseline, the drivers, your geo experiments, three models, their comparison and the plan. Every step writes a SAS program from what is on the screen and shows it under "The program", and every run lands in your Jenner Studio project with its log.
Needs Jenner Studio, which is part of Jenner. Plans are on the pricing page.
Marketing Mix, step by step
Data
Sources · Checks · Model table
Estimate the baseline
In 0.4.0Choose the drivers
In 0.4.0Read the experiments
In 0.4.0Fit the models
In 0.4.0Compare the models
In 0.4.0Plan the budget
In 0.4.0§1 How it works
Nothing happens out of sight. Each step is a program you can read, and each run is kept.
Each step writes a SAS program from what the screen holds and shows it under "The program", so you can read exactly what will run.
Each run lands in your Jenner Studio project together with its log, next to the rest of your work.
The explanations are sentence templates filled in from the run's own results.
§2 Step 1, Data
The first step has three tabs: Sources, Checks and Model table.
Tab 1
Add .csv, .xlsx or .sas7bdat files, or tables in a library. Each source is added as facts, a lookup, a calendar or an experiment.
Each source is profiled: rows, column types, missing values, ranges and distinct values. For dates, the profile says how far apart they are: a day, a week, a month or a quarter.
| Column | Role | Type | Missing |
|---|---|---|---|
| week_start | time | date, a week apart | 0 |
| store_id | place | text | 0 |
| units_sold | outcome | number | 14 |
| tv_spend | spend, channel TV | number | 0 |
| shelf_price | price | number | 3 |
Example profile of one source (invented data)
Marketing Mix guesses the role of every column, and you can change any of them.
Tab 2
Marketing Mix looks for the problems that spoil a model. Each finding is marked "must fix" or "worth a look".
Tab 3
The last tab writes the weekly table a marketing mix model starts from.
Marketing Mix comes with example data for a simulated manufacturer: sales, products, stores, media, competitors, search, weather, weather stations, a calendar and two geo experiments, with the column roles already set. It is simulated, so it belongs to no real company.
Version 0.4.0 takes your files from the first checks to a budget plan, and keeps every result as a table in your project.
§3 Steps 2 to 7
All seven steps are in version 0.4.0. Each result is a permanent table, kept in your project unless you choose another library, for a workflow or a notebook to carry on from.
02
In 0.4.0Two methods, either or both: the classic baseline engine and a state-space model (PROC UCM). Checks on the baseline, and the lift each kind of promotion adds (PROC MIXED).
03
In 0.4.0Give each column of the model table a use. Marketing Mix writes the drivers table, with carry-over (adstock), lags, the price index, season terms and a dummy per place, and flags channels that run together and high variance inflation.
04
In 0.4.0Read your geo tests: difference in differences, a permutation test and a synthetic control. The estimator is chosen on fake tests before any real test is read, and each test gives a lift and a return per dollar withheld.
05
In 0.4.0Three models: MMM 1, regression with carry-over (PROC AUTOREG); MMM 2, Hill curves and adstock (PROC NLIN); MMM 4, Bayesian and calibrated by your experiments (PROC MCMC). A long run shows its progress, and can be stopped and resumed.
06
In 0.4.0Fit, return on ad spend (ROAS) and marginal ROAS by channel, response curves, and each experiment replayed through each model. The pick follows a rule shown on screen before any model runs.
07
In 0.4.0Each model that converged plans the same total spend across the channels, within bounds you set. A channel the models disagree on is held where it was, and a cautious alternative sits beside the recommendation.
Still to come: three more models, MMM 3, the two-layer model and Meridian.
§4 What you need
Marketing Mix needs Jenner Studio, which is part of Jenner. It installs into Jenner Studio and shows up there. Plans and prices are on the pricing page.
§5 Related
Our free video course on the Marketing Analytics page takes a simulated retailer from raw store sales to a budget decision.
Watch the free course →If you would rather have us build marketing mix models on your data and deploy them in your own infrastructure, that is our Marketing Analytics service.
See the Marketing Analytics service →Marketing Mix runs in Jenner Studio, part of Jenner. Plans and prices are on the pricing page.
Tell us about your data and what you want to measure, and we will get back to you.