The Recast Process
How to build robust, statistically rigorous Causal Media Mix Models.
Recast is designed to provide all the information you need to validate your media mix models both before and after you start using them for ongoing planning and budget optimization.
Recast does this with a best-in-class configuration process, rigorous model performance checks and ongoing model calibration.
Phase One: Initial Model Configuration
In order to get stable and robust results, Recast’s underlying Causal Media Mix Model must be configured to match how
your business and marketing program operate. No two businesses are the same and the purpose of model configuration
is to make sure that Recast is configured to match the realities of your marketing program.
Deep-Dive Business Review
Meet with Recast data scientists to discuss the key drivers of business and marketing performance. Discuss questions like:
- What KPIs are most important to your business?
- How are marketing channels defined and operated?
- Do you have other evidence of the incremental impact of marketing activities?
Model Configuration Call
Next, Recast is infused with prior knowledge of your business. This ensures your model will return actionable results that align with your general understanding of your business.
Phase Two: Model Performance Checks
The Recast platform is designed to help you build confidence in your model’s results before you start taking action with it.
There are a number of performance checks your model must pass before it is delivered:
Prior Consistency Checks
Media Mix Models are highly complex, and this check ensures that the model’s priors are plausibly consistent with the data that Recast is seeing. The prior consistency check validates that your assumptions about your business performance are plausible and compatible with your actual data.
Parameter Recovery Checks
After initial model configuration, parameter recovery checks serve as the ultimate model validation test. This check asks the question “if we know the ground truth, can the model find that ground truth?” and is key to ensuring the model’s reliability and effectiveness.
Stability Loop Checks
Stability loops are designed to validate the robustness of the model results. During the stability loops exercise, Recast runs the same model configuration on different subsets of the data to check how much results change when the underlying data is varied slightly.
This set of rigorous tests ensures that the model’s output doesn’t bounce between different possible (but inconsistent) parameter checks. Ensuring stability is hugely important before you make decisions with your model!
Predictive Accuracy Checks
Out-of-sample predictive accuracy checks show how well the model can predict a KPI of interest on data that it hasn’t seen before. This step helps validate the model’s output, provides evidence that true causal signals have been identified, and starts to build trust in the model’s forecasting ability.
This is an extremely important check in the model building process. If the model fails to capture major trends in “future” data, it probably does not have a good read on the past.
Phase Three: Action & Iteration with Production Model
The work doesn’t stop after initial performance checks are passed and your model is put into production. The out-of-sample
forecast accuracy of Recast models is automatically validated in-platform every week, and you can also work with the Recast team
to design an experimentation roadmap that will calibrate the accuracy of your model over time.
Taking Action with the Model
Once configuration is complete and your model passes stability and robustness checks, it is put into production. At this point, you’ll begin routine meetings with your dedicated account manager and data scientist to review the production model’s insights and create a roadmap of actions you can take to improve marketing performance and your model’s accuracy. We call this process of action and iteration an “incrementality system”.
Using Recast to Operate an Incrementality System
Recast provides all of the tools you need to engage in this “incrementality system” right out-of-the-box. Broadly, this approach looks like the following:
- Using Recast’s planning and forecasting tools, your intuition as a marketer, and channel-level insights to build an initial marketing budget.
- In tandem, creating an experimentation roadmap that’s focused on channels with relatively high uncertainty in your model or channels that are strategically important for other reasons.
- Executing experiments and using their results to validate the output of your model and calibrate its future runs.
- With a more calibrated model, results from experiments, and refreshed insights from new model runs, optimizing paid media budgets across channels.
- Then, repeating this process!
By following this system of action and iteration with the help of your dedicated Recast team, you can both calibrate your model and improve the performance of your marketing organization over time.
Schedule a Recast demo
Learn how Recast can help you eliminate wasted marketing spend, optimize your channel mix and plan for the future with reliable forecasting.
Not ready for a demo? You can always reach us at: info@getrecast.com
Recast is SOC 2 compliant and a badged Meta Measurement Partner.
