‍What is OutPoint?

OutPoint is an automated media mix modeling (MMM) platform that empowers high-growth brands to grow revenue and reduce wasted spend. OutPoint's predictive models recommend how much to dial-up or dial-down ad spend across channels and offer a roadmap for unlocking future growth.

A media mix model is a statistical engine for measuring the impact of different advertising channels on a brand's overall performance. The basic idea behind a media mix model is to estimate how much each advertising channel contributes to the quantitative success of a marketing campaign.

OutPoint uses advances in machine learning and media mix modeling to help consumer growth teams map ad spend to revenue based on incremental lift and predicted performance. We help answer the two hardest questions marketers face: are you spending effectively, and where should you allocate your dollars to get more revenue?

OutPoint differentiates from more expensive consultancies and complex solutions by making MMM accessible and automated. We offer:

1. Fast time-to-value: Simple, fast, and secure data onboarding processes (a key constraint to most MMM projects)
2. Comprehensive coverage: The capability to model a diverse array of online and offline media and revenue sources (for example, understanding the lift of retail media like Amazon)
3. Automated insights: ML-powered recommendations that can improve in accuracy over time

OutPoint's cleanly-delivered insights  help a marketer unlock "the big three" of marketing measurement: attribution (what is the economic impact of my online and offline channels?), incrementality (what is the lift above the baseline?), and marginality (what will my next dollar earn me?).

How does OutPoint work?

The OutPoint predictive model is the engine that powers the core product, using regression-based analysis to determine the efficiency and scalability of each advertising channel. We use various techniques to model the relationship between revenue and one or more marketing inputs (e.g., spending): elastic net regression, bayesian regression, and a weighted ensemble model. These techniques help us uncover how increasing or decreasing spend can impact revenue, thus allowing us to recommend how much a client should increase or decrease spending to maximize revenue. We build these models using machine learning libraries in Python and custom research developed over years of client deliveries and model simulations.

How do you onboard my data?

We use daily time-series data to model changes in media spend to changes in revenue (or some other optimization event) over time. OutPoint makes it easy to ingest the daily ad spend and revenue data required to build your media mix model. For any platforms integrated with OutPoint, you can follow simple 1-click steps to connect your accounts. For channels and revenue sources that OutPoint does not and can't integrate with, you can upload a CSV file to OutPoint, enabling virtually complete coverage of any spend or revenue source.

‍

‍

‍

How does OutPoint predict return?

There are two types of models that we train: direct models and lift models. Direct models are simpler and represent the relationship between a single channel’s spend and that channel’s direct revenue (e.g., UTM tags for that channel). Lift models are more complex, they represent the relationships between all advertising channels and a specific type of lift revenue, such as organic/brand revenue or email revenue. The following graph illustrates how direct and lift models map spend to revenue:

Independent channel-level correlations between ad spending and revenue lifts inform our total return models. First, we train an ML model to compute a return value for 'Direct Return,' allocating and predicting credit for 'direct response' multi-touch UTM sessions associated with the channel spend. Next, we train the ML model to predict 'Brand Lift' and 'Email Lift' aggregated return and allocate partial credit between channels using causal inference techniques.

The last step is to take all of the direct and lift models we have trained and use them to make predictions and recommendations. We distinguish between offline and online channels when making predictions, where offline channels do not have any directly attributable return and are thus fully dependent on lift models for their revenue predictions. Online channels have both direct and lift revenue. The following equations represent how offline and online channel revenue predictions are calculated:

offline channel A  revenue = organic lift(channel A spend) + email lift (channel A spend)

online channel B  revenue = direct B(channel B spend)+organic lift(channel B spend) + email lift (channel B spend)

‍

We use three features to improve model accuracy:

1. Bayesian priors: Our Bayesian regression framework considers multiple distribution probabilities and allows us to apply prior beliefs on channel inputs. For more information on how we use Bayesian priors, see the following post.
2. Transformations: Revenue and spend data is optionally transformed through smoothing, time series decomposition, outlier correction, etc
3. External Factors: OutPoint incorporates external factors such as promotional periods (BFCM), consumer sentiment, economic indicators, etc.

‍

‍

Why use MMM?

The benefit of MMM over click or pixel-level attribution aloneis that, as a top-down predictive model, the primary inputs required are changes in spend and changes in the overall outcome, typically revenue. OutPoint ingests a historical set of signals to make the models actionable, and once we have enough of these signals, we can infer the estimated efficiency of each channel, allowing us to construct an estimated return curve for each outlet. Since our models rely on inferred "cause and effect" vs click-based attribution alone, OutPoint media mix modelling (MMM) can capture the value of economic lift from less-attributable media such as Influencers, TV, and Radio, which do not always provide impression or click data at a granular level.

While MMM as a general methodology does not require granular click data, OutPoint's models will incorporate first & last click data if relevant as an input improving modelled inference around cause and effect, specifically for direct response-oriented channels like Paid Search.

‍
Marginality and Diminishing Returns

OutPoint's product can predict the various spend levels where spending will increase marginal customer acquisition cost (CAC). Being aware of the rate at which marketers see diminishing returns helps determine optimal spend levels for profitability or faster, more predictable growth.

For each channel, we predict the direct revenue, if applicable, the organic lift revenue, and the email revenue over the time period using the actual channel spends over that time. We also predict the direct and lift revenues for an evenly spaced range of channel spends (usually from 0 to the maximum historical channel spend) to create a performance curve that shows how revenue predictions vary over different spends. 

We make direct and lift revenue predictions for the current baseline spend (calculated as the average spend over the last 14 days) for each channel. And finally, we find the optimal channel spends for both maximizing profit and maximizing revenue and make direct and lift revenue predictions as well.

‍

‍

External Factors

External factors such as sales, variations in consumer sentiment, purchasing power, etc. may affect revenue and have nothing to do with advertising spend. Thus, it is important to be able to model the effects of these factors and use them to remove non-advertising driven revenue or explain inflection points that may occur. OutPoint's MMMs are designed to account for external factors such as seasonality and pricing (e.g. Black Friday, Cyber Monday), showing a clearer picture of what is happing beyond average metrics.

‍

‍

The "Open Secret"

Scaled companies like Wayfair, Netflix, and Amazon have already been investing heavily in the above areas and the concepts behind MMM. OutPoint’s purpose is to make these types of approaches accessible to level the playing field and give growth stage companies the tools they need to enter their competitive landscapes.

Google and Facebook have also published open-source code and research validating the methodologies of incrementality and MMM, making similar models available to help researchers build their own tools.

• Google: An R package for causal inference
• Facebook: Robyn

Limitations of MMM

MMM is not a silver bullet and models should be used in addition to experimentation and more granular daily analytics tools to "triangulate" ROI. For example, there is a constraint around granular user-level attribution in MMM. A brand can use MMM to infer that a particular level in television spend likely drove a specific dollar amount in incremental revenue, but not the specific incremental user. Some business environments also deal with an extended purchase journey or a lot of noise in the buying process.

More Information On Incrementality + MMM

• Netflix Research: Incrementality & Attribution
• Greylock: Building and Measuring Sustainable Growth Marketing
• Harvard Business Review: Marketing Analytics 2.0
• Mobile Dev Memo: Media Mix Models are the Future of Mobile Advertising
• Google Research: Challenges and Opportunities in Media Mix Modelling

OutPoint is a suite of data science tools used by high-growth B2C marketers to allocate paid media budgets optimized for incremental lift and improved marginal returns.

Contact us to schedule a demo or meet with the OutPoint team to learn how you can use MMM to lift your business.

🚀Sign up, or email rob@outpoint.app if you are interested in learning more.