Forecasting for Search Marketing: Art + Science

By Adam Garcia, Paid Search Supervisor, Advertising Solutions

As search marketing professionals, we are often asked to predict the future. The prediction often comes in the form of a forecast, plan or perhaps a weekly click estimate. Fortunately, we have many tools available to us. Google’s Traffic Estimator is a great place to start. By entering your keywords into the first box, along with optional data such as maximum cost per click, daily budget, language preference and country, Google will provide an estimated clicks per day. Based on an average cost per click, also provided by Google, the tool shows what it might cost on a daily basis to bid on the keywords entered.

This is a great place to start. As with any tool though, especially one named “Traffic Estimator”, there are varying degrees of accuracy. Relying on any one tool can lead to an estimate or forecast that greatly over (or under) estimates the traffic and therefore costs. This can be hazardous to the health of any search marketing program (and the health of the search marketer managing the program if there are sales and order data expected with those clicks!).

Knowing that we can never be 100% accurate with forecasts, what we’d like to achieve is that gut feeling that tells us that we’ve got it. We’re comfortable with it. That gut feeling is telling us that we’re within 10% +/-. This is also known as a well educated guess. We can reach this point through triangulation. By using multiple tools to give us the various data points, we can be certain that the answer lies somewhere in the middle.

To that end, we need other reference points. Microsoft offers another tool to estimate traffic. The adCenter plugin for Excel 2007 provides another view of the traffic that can be found in the search universe. While this tool does require a username and password to login to the KSP server, it is an invaluable tool to search marketers when creating forecasts. The data provided by this tool is another data point in our quest to triangulate the data. is a third data set used for forecasting. This provides information on related searches to a particular keyword. This can be helpful when trying to determine what searchers are actually typing into the search box and how often. It often shows keywords or queries that we wouldn’t necessarily have insight into otherwise. It can show the number of searches as well.

Once all of this data is collected, it can start to be shaped into a forecast of sorts. Most search marketing forecasts show the standard metrics: impressions, clicks, click through rate, average cost per click, average position, and media cost (or spend). If this is for a direct response client, we would include conversion rate, orders, average order value, and total sales (or revenue). Return on Ad Spend (ROAS) and cost per sale (CPS) might also be included depending on the client. All of these are based on a number of assumptions which is where the “art” comes into it. What we need to do at this point is to take a pragmatic approach. By asking what the potential limits of a KPI might be, we can narrow the range even further. For example, if we have data that shows a particular keyword will generate 700 clicks per day, we would ask is 1000 clicks per day feasible? If not, then it’s something less than that, 800 perhaps. Conversely, we would ask, is 500 too low? By determining these limits, we can narrow the band to come closer to the answer.

Once the clicks are estimated, perhaps on a weekly basis, we would ask if the clicks are going to be exactly the same from week to week or are there extraneous variables that would make the clicks go up or down during certain weeks. By using another Google tool, Insights for Search you can determine seasonality and or trends in search volume. Using this tool, you can adjust and massage the data to match what might happen in the marketplace.

The final step is to set some assumptions. Unless you’re working with a good set of historical data, many assumptions have to be made in order to complete the forecast. Click through rate is usually an assumption. This allows you to back into impressions. Depending on whether or not the media spend was set, you can either back into media spend based on CPC or estimate the media spend based on an assumed CPC. Finally, for direct response clients, by assuming a certain conversion rate, you can estimate the number of orders. Sales or revenue can be estimated based on an assumed average order value.

By picking one (or two at most) of the metrics listed above, CTR for example, you can create scenarios: Best Case, Worse Case and Probable. (Or high, medium and low CTR scenarios.)

All of this research, planning, educated guessing, massaging, tweaking and altering can’t account for the variances in an auction based marketplace. Once the campaign is launched, any number of things can happen. Competitors can enter the market place and cause CPCs to rise and clicks to go down. Competitors can change their bids for keywords at any time. News articles and blog posts can create unexpected surges in traffic. Holidays can create extreme variances in search volume.

Knowing when and how to react to these changes is what makes us experts. Being able to forecast these metrics makes us artistic scientists.


Copyright © 2008 Resolution Media, Inc. All rights reserved.