Predictive Modeling
Efficient Frontier is the only Search Engine Marketing (SEM) solution provider with the ability to harness the synergy of modeling, optimization and automation to forecast the performance of your campaign. Our forecasting hierarchy comprises:
- Keyword level forecasts: our Click Models and Revenue Models predict the number of bids, number of clicks, CPC and revenue at every position for each keyword in the portfolio.
- Portfolio level forecasts: which predict the tradeoff between spend and the business objective. Since the tradeoff between volume and ROI/CPO is critical for most businesses, portfolio level forecasts enable advertisers to pick their sweet spot - the optimal ROI. When an advertiser has several business objectives, portfolio level forecasts can be combined to calculate the combined tradeoff between spend and the business goal.
An Efficient Frontier Predictive Model Simulation forecasts optimal search spend.
The power of this approach enables the search marketer to spend more time on strategic decision-making in areas such as budgeting and campaign growth, and less time on tactical issues such as keyword bidding decisions.
Estimates of key performance metrics - such as CPC, number of clicks, and revenue of every keyword in a campaign - comprise the critical information on which any SEM solution will operate. If these estimates are not accurate, then any method to optimize an SEM campaign for performance will not be successful. Given the variability in the marketplace, it is no surprise that this is a challenging task.
There are three main factors that determine the quality of a model: the availability of historical data, the capability to properly model tail terms, and adaptability of the model.
Availability of Historical Data
Needless to say, if you were to rely on a day’s worth of data to predict performance, day-to-day market fluctuations would make your predictions very inaccurate. A good model needs to incorporate the trends seen in a week’s or a month’s worth of data. Long-term historical data is also needed to predict performance of keywords that may perform well during a particular season but then decline.
At Efficient Frontier, we have achieved an extremely large pool of historical data covering multiple verticals. This enables us to build accurate models. Additionally, our extensive research of aggregated vertical data allows us to improve model accuracy even further.
The Ability to Model Tail Terms
The majority of terms in a search campaign are tail terms, which typically get less than one click per day. Due to the lack of historical data, sophisticated data aggregation techniques are required to build models for tail terms.
At Efficient Frontier, we use finite mixture models to build models for these keywords. This method, delivers better results because click and revenue data from several keywords are clustered to build a common model.
Adaptability of the Model
Search engine marketplaces are influenced by numerous factors, including news events, competitors, user behaviors and search engine algorithm changes that affect CPC and click and revenue volume of a keyword on a daily basis. In order to maintain accuracy, keyword models must adapt extremely quickly to these marketplace shifts.
At Efficient Frontier, we build adaptability into our click and revenue models. First, we build these models on a daily basis for every keyword, and every match type for every client. That means we build over 66 million individual models on a daily basis. This ensures that keyword trends are identified and responded to immediately.