by Raghu Siddani, VP Data Strategy and Analytics, Broadbeam Media
When you invest time and effort to measure the impact of media buys, the real value of this attribution is knowing what to do next.
For thoughtful performance advertisers, forecasting key performance indicators is an integral part of the planning process. Traditional direct marketers look for a balance between immediate performance and building their brand and tracking lower funnel metrics that are closely tied to profitability is just the first step.
A key tool in building forecasts for lower funnel metrics is building marketing mix-like models that are not only descriptive of the recent past but look forward with a reasonable degree of accuracy. Effective forecasting can result in critical and quantifiable benefits but requires careful attention to detail and significant data discipline.
Why bother with a forecast?
Accountability. For most clients today, conversations on media performance are no longer restricted to sales. Consumer journeys are complicated and marketing impact requires a holistic outlook of the business. Creating effective forecast frameworks provides for a credible understanding of the client’s business and contributions of marketing activity to business growth.
Scalability. With the explosion of various forms of digital media and the proliferation of smart TV’s, there’s lots of places to invest a media dollar. However, returns on marketing investment differ drastically by channel and traditionally have a non-linear relationship with key performance indicators. An essential element of creating effective forecasting models is to carefully estimate potential diminishing returns from increasing media spend. Appropriately identifying the size of the audience reachable through each channel as well the cost and availability of next eyeball is critical to understanding the response of profitability drivers to changes in levels of marketing investments.
Seasonality. A critical component of any robust forecasting model is to accurately estimate seasonality. Whether we are looking to forecast demand or accurately predict sales, understanding annual patterns can identify key times when spending marketing dollars can be more effective. Seasonality can provide valuable information that could be used to make spending decisions weekly, monthly or quarterly.
Interaction between response drivers. To be accurate, forecasts need to understand and incorporate the interplay between marketing drivers. Typically, media channels do not drive demand or sales in isolation but rather collectively impact response. Models incorporating these interactions provide a comprehensive view of the business that are forward looking with a greater degree of confidence. A common example of such interaction is when TV drives Paid Search which, in turn drives key performance indicators of the business. Similarly, marketing activity is planned around seasonal patterns to maximize impact.
Once significant drivers of the business are accurately incorporated, scenarios can be built that run multiple spending options across marketing channels.
Scenarios. We’re building models to make investment decisions about the future. Scenarios can provide an estimate of where the business would be in a span of a few weeks, months or quarters. Forecasting models can provide a quantitative measure of the impact of non-marketing drivers as well. These drivers are typically not in the control of the business and action can be taken to provide guardrails against ineffective spending or opportunities to maximize the returns on marketing efforts.
How do you make sure your model works? Methodology.
Data Integrity. Not surprisingly, the most critical aspect to building usable forecasting models is not the mathematics of the modeling, but ensuring data integrity that feeds them. Without clean data, forecasting models could potentially provide direction that is detrimental to the business. To ensure data is accurate, the first step in the modeling process is data discovery where data is confirmed and finalized. This includes identifying and vetting a candidate list of potential influencers of the business that include marketing and non-marketing drivers such as pricing, competitive activity and product launches.
Modeling + Updates. While data accuracy is critical to any modeling effort, tracking and anticipating changes to data is needed to build usable forecasting models. Along with choosing appropriate models, changes in data over time can have an impact on model integrity as well as accuracy of forecasts.
Changes in the data collection and storage processes can have a significant impact on forecasts. This not only affects estimates of marketing efforts but can change key performance indicators used in the model as well, leading to forecasting failures.
Selecting appropriate adjustments to key data inputs (KPI’s + model drivers) as well as model granularity choices can have a significant impact on model stability and results. For instance, choosing an appropriate categorization of product lines as well as aggregating DMA level information to reflect business activity can have a significant impact on model results.
Changes in business choices can have a significant impact on all metrics and need to be appropriately identified and adjusted for. These change points may need adjustments both within and outside the forecasting model.
Any effective forecast is dependent on estimating how data inputs will behave moving forward. For example, how historical seasonality extends into the future is a critical input for an accurate forecast.
While digital channels can target audiences individually, reach has historically been driven by traditional media channels especially TV. In the current landscape, opportunities for efficiency typically lie with digital and opportunities for scale with channels like TV. However, that is expected to change with spending in digital media now exceeding traditional channels.
Forecasting requires diligent, complicated work for results that aren’t 100% accurate. But it still more than pays for itself.
While forecasting the future with complete accuracy is a near impossible task, robust accounting of data in hand and intelligent assumptions can lead to usable models and insights. The key to using forecasting effectively is to carefully integrate data discipline with robust methodology into the decision-making process for the business. As with any data-related field, collaboration across business functions leads to the most flexible forecasting tools that bring efficiency and scale to media planning and optimization functions.