When Should You Delete a Low Traffic Keyword?

Regularly researching keyword opportunities and expanding your term list is critical to the success of a paid search program, but knowing when to delete keywords with low potential is also an important consideration in managing PPC accounts at scale. Why Delete Keywords at All? In addition to the engines having finite account size limits, keyword lists, especially in large, mature accounts, can grow to an unwieldy size as new keywords are added throughout the years.  This can make it more difficult to quickly enact time-sensitive changes such as promotional copy additions and intraday bid adjustments.  Thousands of additional keywords translate into hundreds of additional adgroups and internet and search engine API bandwidth limitations restrict how quickly requests can be transmitted and processed.  This issue is compounded if keywords need to be duplicated in various parts of the account for testing or geotargeting purposes. Also, while you may not see much in direct costs associated with allowing keywords that aren’t generating many clicks to continue running in the account, these no click/low click-through rate keywords may be lowering your Account Quality Score and indirectly raising the costs per click of other keywords in the account. Perhaps worse than not pruning low impression/traffic keywords at all, paid search account managers may arbitrarily decide to delete keywords that have not had an impression or click within a time-period of X, where X is a nice round number like 30 days, 6 months, or a year.  We hope our analysis below will help account managers make a more data-informed decision, minimizing the negative impact this task could have on long-term sales. Before We Begin…
  • We assume keywords without impressions, clicks, and/or orders have been appropriately optimized (ex. for low/no impression keywords).
  • The data analyzed in this post categorized a single keyword with separate matchtypes and campaign settings (ex. location, language, networks, device, etc.) as separate, individual data points.
  • RKG, while balancing financial risk and rapid learning, tends to set fairly aggressive initial bids, so the windows between a keyword's early impressions and clicks may be biased a bit shorter compared to more recent data, but the early data is still a reasonable proxy for future performance patterns.  We are using the earliest data here in order to incorporate sales and cost figures generated over the full lifetime of the keywords.
Deleting Keywords Without an Impression in X Days For this part of the analysis, we aggregated the lifetime sales and costs of non-brand keywords from a bucket of well-established accounts (average age of 4 years) by the time between each keyword’s first and second impression. The following graph shows the relationship between that time difference and a keyword’s lifetime contribution as a percentage of total account sales and costs.On average we found an account would lose about 1.0% of long-term sales and save 1.1% of long-term costs if keywords that did not have a second impression within 17 days of its first were deleted. To be clear though, unless there are compelling reasons, we do not recommend deleting zero or low impression keywords before they have run for 365 days.  A full year helps to account for seasonality, even for the most specific keywords. It is not difficult to imagine a keyword such as “[Small Town] Valentine’s Day chocolates” launched on exact match in March not receiving any impressions until the following January or February. For the very conservative account manager, 99.99% of total account sales and costs will typically be generated by keywords with less than two years between impressions.  If you have terms that fit that bill, you should be pretty safe to free up that account space. Deleting Keywords Without a Click in X Days Using a similar approach to identify keywords with no clicks or low click-thru-rates worth deleting, we aggregated the lifetime sales and costs of non-brand keywords from the same accounts by the time between their first impression and first click. The following graph shows the relationship between that time difference and a keyword’s lifetime contribution as a percentage of total account sales and costs. This data shows that one would lose about 1.0% of long-term sales and save 1.1% of long-term costs if keywords that did not have a click within 150 days (5 months) of their first impression were deleted. However, the data also shows that it took keywords with about 2.5 years between first impression and first click to generate 99.99% of total account sales and costs, which was longer than the age of a few of the accounts in our group. Looking at the accounts individually, we found some had 99.99% of their sales/costs generated by keywords with less than 6 months between first click and first impression while it took other accounts’ keywords almost three years between first click and first impression to reach that 99.99% mark. Age of the Account Impact We quickly realized that the predominant factor for this variability was the age of the account. While the data for a head-heavy client in the early days looked very different than an account that benefited more from the tail, we saw a convergence of 99% of total account sales being generated by keywords that have had a click since their first impression at about 11% of the age of the account. We see a similar trend as our previous graph, but on a more representative scale. For the very conservative account manager, 99.99% of long-term account sales and costs will typically be generated by keywords with less than 70% of the account age between first impression and first click. Summary Remember that the information in this table is just a guideline based on the analysis of a subset of our client base. There is no substitute for a competent account manager that can identify the individual keywords that are increasing account management overhead, decreasing account QS, and otherwise adding only minimal value to the program. Special thanks to Mark Ballard for his contributions to this post.
Join the Discussion