Professional Services · Marketing Analytics
Know what every marketing dollar returns, with models your own team runs.
We build marketing mix models (MMM) on your sales and media data — Google Meridian, Bayesian MMM, and regression with adstock and saturation — prove them with geo experiments and independent code, and deploy the best one in your own infrastructure. The models are SAS-language programs that run in Jenner, so your analysts, including a team that already does its marketing analytics in SAS, re-run and re-train them every quarter with no further services.
Fixed-price and time-and-materials engagements.
§1 The offer
Media mix modeling done properly once, then handed over: your team owns the software, the trained models and the pipeline that runs them, with no ongoing service subscription.
The edition with 200+ procedures and no data limit, on macOS, Windows or Linux. It runs SAS-language programs, including the models we deliver. One year of new releases and support is included.
A set of models fitted to your own sales, media, price and promotion data, with the budget recommendation they support. Your team re-runs them on new weeks and re-trains them as the market moves, without calling us.
We connect the models to the data you already have and run them where you already run things: your cloud or your own servers, on your scheduler, with results delivered to your BI and planning tools.
§2 How it starts
We look at your data and your infrastructure first, so the estimate is a number we can stand behind.
01 /
A working session with your marketing, analytics and data teams: what you sell, which channels you buy, what data you hold and where it lives.
02 /
A written scope, timeline and price for the models, the integration and the hand-over, as a fixed price or as time and materials.
03 /
We prepare the data, fit the models, check them against independent code and your experiments, and agree the results with your team.
04 /
The models go live in your infrastructure, your analysts run them with us once, and from then on they run them on their own.
One price for an agreed scope. The right choice when the data is in good shape and the deliverables are clear.
A fixed-price engagement →Billed for the work done. The right choice when the data needs work first, or when you want to explore before you commit.
A time-and-materials engagement →§3 What you get to see
Four figures from our marketing mix modelling course, made with the same models we build for clients.
FIG. 1 — Response curves by channel
How much each channel adds as you spend more, and where it stops paying.
Source: invented data for illustration
FIG. 2 — What moves revenue
Every factor ranked by how much a 1% change moves revenue, media and non-media alike.
Source: invented data for illustration
FIG. 3 — A geo test, read correctly
Switch TV off in a few markets and the lost sales show up, then run on after it comes back. We measure both.
Source: invented data for illustration
FIG. 4 — Market by market
Which channels work the same everywhere and can be planned nationally, and which vary by market.
Source: invented data for illustration
The engineering review takes one conversation. You leave with a scope and a price.
§4 What we build
We fit several model families — including Google Meridian — on your data, test each one the same way, and deploy the one that performs best.
M1
Log-linear regression with adstock, so a TV flight keeps working in the weeks after it airs, and autoregressive errors, so a run of good weeks is not mistaken for media.
M2
Hill saturation curves fitted channel by channel on top of adstock: diminishing returns, and the response curve that tells you where the next dollar does the most.
M3
Markets borrow strength from each other, with a media effect for every market, so you see where a channel works harder and where it can be planned nationally.
M4
Geo-test results enter the model as priors, so the history and the experiments agree on one answer, with an interval for every number.
M5
Test design and analysis: difference in differences, synthetic control and placebo checks on fake tests, so a measured lift comes with an honest margin of error.
M6
How much of TV's and video's effect arrives through people searching for you, so search is not credited with sales the other channels created.
M7
Google Meridian implementation without a separate stack: Google's open-source Bayesian MMM runs from inside Jenner and is tested on the same footing as the others — the same holdout weeks, the same experiment calibration, the same budget optimizer. When it fits your data best, we deploy it into your infrastructure like any other.
What you would have sold anyway, and how much each promotion added on top.
What each channel returned on average, and what the next dollar would return.
Spend against return for every channel, with the point where it saturates.
The best split of the same budget, within the limits your planners set for each channel.
Price, distribution, competitors, weather and media, ranked by how much each one moves the result.
§5 Google Meridian
Google Meridian is Google's open-source Bayesian marketing mix model. We run it from inside Jenner, calibrate it with your geo experiments, and test it on the same holdout weeks and the same budget optimizer as every other model we fit.
When Meridian is the best model for your data, we deploy it into your infrastructure on the same schedule and with the same hand-over as any other: your analysts re-run it and re-train it themselves.
§6 Why you can defend it
Before a ROAS figure reaches your planners, it has been checked three ways.
Every model is fitted again with independent Python and R implementations, and the coefficients have to agree.
We run each model on simulated data where the true answer is known, and show it finds that answer before it touches yours.
Where you have geo tests, the models are calibrated with them; where you do not, we design the test that settles the question.
§7 What we hand over
Jupyter notebooks your analysts can read top to bottom and run again, each step explained in plain words.
Every program behind the models, commented and organized so your team can change it.
Fitted to your data, with the settings and checks to re-train them on new weeks.
Where to move the money next, by channel, with the expected return.
A recorded tour of the models and notebooks for everyone who joins the team later.
Start with an engineering review. We look at your data and your systems, and come back with a scope and a price.
Tell us what you sell, which channels you buy, what data you hold and where it lives. We will come back to arrange the review.