Projecting Performance for PPC Campaigns

If you’ve been in the PPC world long enough, you’ve heard some version of this question many times: 

“Can you predict how many conversions we’ll get from this campaign?”

In this article, I’ll offer advice on how to answer the question, along with cautions to keep in mind along the way.

Use Platform Projections with Caution 

Most ad platforms offer some ability to predict performance based on a list of keywords or audience criteria. Google and Microsoft Advertising each have their own version of a Keyword Planner tool, allowing you to find suggested CPCs and volume for individual keywords. 

You can enter a list of existing keywords and find suggestions for new keywords to incorporate. You can then create a plan, which will predict metrics based on the set of keywords you include.

However, note that this data is based on the past performance of these keywords, and search volume can vary heavily month-to-month and even day-to-day. Particularly with low volume keywords, you’ll often find these projections to be inaccurate, so present the numbers with a disclaimer that actual performance will vary.

Google Trends can also be a good source to get very high level data on how frequently certain keywords and keyword themes are searched, as well as identifying seasonality. Once again, be cautious of putting too much stock in exact numbers, but use the data as a general guide.

Look at Historical Data

Generally speaking, your own historical data (for brands that have it) is going to be the best source of truth for future campaign predictions. Your ad copy, your website experience, and even the general appeal of your product to potential customers are the factors that ultimately drive performance, even beyond any across-the-board volume that may be true of a set of keywords or an audience.

If at all possible, base predictions off the following criteria:

  • Similar audiences to the campaign you’re projecting performance for
  • Similar set of keywords
  • Time of year, to account for seasonal bid changes
  • Geography and other demographics
  • Type of offer (don’t project performance for a bottom-of-funnel leads-focused campaign based on numbers from a top-of-funnel webinar campaign) 

Give Ranges, Not Exact Numbers

Even if an ad platform is projecting a specific number of clicks and impressions, you should temper expectations by presenting ranges of potential volume. For instance, try setting your bid higher or lower in Keyword Planner to project various scenarios. 

If you show your client an expectation of exactly 549 clicks and 26 conversions, they’ll be more likely to expect those exact numbers (or very close to them) and challenge performance if they receive fewer clicks and conversions. However, if you present a range of 500-700 clicks and 20-40 conversions, the range better implies that you’re not providing an exact prediction and that results can vary.

For example, see how LinkedIn does this by showing ranges for all potential numbers once you’ve entered in audience data. 

Use Campaign Learnings to Inform Future Projections

As you run a campaign after providing projections, compare performance to what you predicted. You may find discrepancies resulting from factors such as increased competition driving up CPCs or better/worse conversion rates than anticipated on the landing page.

Use the differences you identify in order to build in buffers when providing future ranges for projections, and make a note of external factors to be aware of.

While there may not be a perfect solution to predicted performance, the need for projections will likely be an ongoing request that PPC managers encounter. Use the tools available to you for looking up potential volume, but rely on your own historical data over the platforms’ predictions, and add disclaimers to note that no projections can be close to perfect.

What tips do you have for developing projections? We’d love to hear your thoughts in the comments below!

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.