AnalyticsForecasting SEM Activity

Forecasting SEM Activity

Forecasting goals, targets, and budgets for coming months or quarters is neither easy nor quick, but you must get your forecasts as accurate as possible to make sure you and your client can agree on reasonable targets. Here’s how.

The best account managers have such deep client relationships that they become a direct extension of a client’s internal team. Along with that comes responsibility, often in forecasting goals, targets, and budgets for coming months or quarters. It’s not easy, and it’s not quick, but you’ve got to get your forecasts as accurate as possible to make sure you and your client can agree on reasonable targets.

For a typical SEM forecast, the best place to start is by formulating a linear run rate based on recent data. Typically this will be last 7 days, last 14 days, or last 30 days.

The proper look-back period will depend on the specifics of the business.

For businesses with sparse conversions, longer look-backs are generally necessary to get enough data to form reliable averages. Some businesses are more volatile and require a shorter look-back as things change quickly.

The right look-back to begin with is something you have to determine based on your knowledge of the business.

From the client side, generally clients only want to see projected conversions, costs, and CPA (or ROAS), but at this first step, you should project out all metrics so you can adjust the next steps to accommodate various possibilities.

For this first basic calculation, just take your top-line data (for all summed fields, shown in blue), divide it by the number of days in that data, and multiply by the number of days you’re projecting for. The calculated fields (CTR, CVR, etc.), shown in orange, will just be the same as in your total.

The below example is projecting for March:

projections-for-march

Adjusting For Day of Week

Most businesses have a pretty dramatic difference between weekday and weekend activity, and some have strong day-of-week patterns for all days. Analyze the trend for your business and adjust your projections to align with the days of the week upcoming in your projection period.

In our example, business weekdays follow one trend and weekends another, so you’ll adjust projections to account for the fact that March has 10 weekend days and 21 weekdays.

For some projections to be sound, projecting each day of the week individually makes sense; if this is the case, you’d want a longer look-back (probably at least 4 weeks so each day has four data points you’re averaging against).

Additionally, consider any holiday days in the month and adjust accordingly; for holiday-based businesses, this may mean a flood of activity before and on holidays. For most businesses, though, holidays typically see activity at or below what weekends deliver.

This March there are two holiday Sundays: St. Patrick’s Day and Easter. For these two days, we can expect all traffic and conversions to be 20 percent lower than the regular Sunday average.

day-of-week-forecast-adjustments

The numbers aren’t dramatically different, but this refined projection shows 84 fewer conversions expected than the pure linear run rate, and that could make a big difference to a client.

Adjusting for Seasonality

All businesses have some seasonality. For some, such as a business selling Christmas tree ornaments, it is very strong. But you should always investigate the less obvious seasonality as well.

Internet activity in general dips in the summer just because people spend fewer hours logged onto computers. However, a bathing suit business likely bucks that trend.

If your business is fairly mature, you can use the previous year’s results to adjust for month-over-month or week-over-week seasonal trends and use those deltas to adjust your projection numbers. For this account, we can look at the month-over-month 2012 numbers – but the deltas are more a reflection of expanded test initiatives done last spring and not seasonal activity, so they really won’t help for this.

month-over-month-deltas

If you weren’t in business last year or your business has matured significantly from the previous year, Google Trends is your friend here. Pull the data on your core head term (that single term is usually enough, though you may want to test against a couple core terms) and use the percent deltas on the numbers Google provides to adjust your week-over-week or month over month numbers accordingly.

google-trends-forecasting-all-inclusive-vacation

Say we were in the highly seasonal all-inclusive-vacation business. You can compare your two data periods in Google Trends and get an average volume number.

Here it gives 81 for January compared to 52 for March, so a 36 percent drop in impressions can be expected. Starting with the day-of-week adjusted data, you can adjust your impressions by this expected seasonal drop.

day-of-week-with-seasonality-forecasting

And then assuming CTR, CPC, CVR remains even, calculate all your other numbers and a pretty different picture emerges!

adjusted-march-forecasting

Adjusting the Data For the Things YouWill Do

Now you need to account for the things that you’re going to do to the account. Here is why you kept carrying all the cost and impression data with your projections. This part gets into more subjective data points sometimes, but you can make reasonable projections from the data you have available. The possibilities are endless, but here are a few common scenarios.

  • lost-is-budgetThe client was budget-capping last month and wants to run full speed next month.In this case, pull the budget impression share numbers from Google for your projection period and adjust your impressions to align with 100 percent budget impression share. If CTRs CPCs etc stay the same, you can get a reasonable guess of how much activity will increase. Add 2 percent to your impression and calculate from there.
  • You’re launching new keywords. These can be tricky as you have to rely on an additional data point – Google’s estimated keyword searches, which can be unreliable. Instead of using the sum of this for all keywords in a launch, just use the base “head term” within the launch to calculate expected impressions. Assuming average CTRs and CPCs, you can factor in the impact of these launches.

keyword-searches-data

    In this example, if we launch a keyword set around “all inclusive bora bora” vacations, we can expect 12,000 additional impressions.

  • You’re launching in a new metro. Look at your data for a similar-sized metro from the previous month and add it in. (TIP: It’s in your dimensions reporting.)

Additionally, it may make sense to look at the long-term trends in the account to gauge the impact of your ongoing optimizations and the changing marketplace and adjust to the averages and the expected trends. (Example: CPC has been consistently dropping in the account for the last 8 months, and if you project that trend to March it shows the expectation that it will be about $.40 lower in March than in January. You can do the same for all metrics, not just CPC.)

That said, if the “trend” isn’t particularly consistent over time, the linear average projection won’t be reliable. Make sure to execute your optimizations on a consistent set schedule, and for some metrics that are highly variable or influenced by inconsistent things, don’t use this method.

avg-cpc-trends-linear

forecasting-with-cpc-trends

The Gut Check

The final polish is the gut check. This can account for changes you’re making to the account that don’t have data points to draw from, like the addition of sitelink extensions.

Here you can sometimes tweak based on your personal knowledge of past changes in this or other accounts, but often it just requires a final bump up or bump down based on gut.

When you do this, try and remember that human beings (even you!) are irrational, eager to please, and overly optimistic. You will regret it later if you don’t temper your “gut” with some worst-case-scenario thinking. In fact, the argument could be made to always round your numbers down a bit in order to more reliably exceed expectations instead of setting yourself up to fail.

Delivery

Typically, because so many factors are in play, it’s smart to deliver the client a range of projections or a couple of different projection outlooks based on different scenarios. Whether your client is amenable to this or requires one hard number, it’s important to message these forecasts realistically.

Smart analysis of historical data can be a great predictor of the future, but we don’t have true crystal balls. The reality is that events we have zero control over (such as a new aggressive competitor entering the space or a weather event) can seriously disrupt the results, so make sure your client gets that message and forecast away!

Resources

The 2023 B2B Superpowers Index
whitepaper | Analytics

The 2023 B2B Superpowers Index

8m
Data Analytics in Marketing
whitepaper | Analytics

Data Analytics in Marketing

10m
The Third-Party Data Deprecation Playbook
whitepaper | Digital Marketing

The Third-Party Data Deprecation Playbook

1y
Utilizing Email To Stop Fraud-eCommerce Client Fraud Case Study
whitepaper | Digital Marketing

Utilizing Email To Stop Fraud-eCommerce Client Fraud Case Study

1y