TL;DR version: Based on this study:
- Click-through rates were significantly higher without DKIs (10-65% higher)
- Conversion rates were significantly higher without DKIs (300%+)
- This study did not manage to give a definitive conclusion on the impact on the cost per click when using the DKI
The online marketing community seems to have a love/hate relationship to Dynamic Keyword Insertion (DKI). While I generally do not recommend using it, there may be cases where it is very helpful and where the advantages outweigh the disadvantages. This is particularly the case, when time and resources is the constraining factor.
- Fast to make relevant ad copy
- Limits the number of ad groups needed to stay relevant
- May give weird ad copy
- May lead bad and irrelevant traffic to the site
A word of caution for those using or looking to use DKIs before we go into the test data; if you are not careful, you may end up attracting unwanted traffic! In fashion e-commerce, there are three categories: Apparel, footwear and accessories. With the mind occupied by fashion, you may decide to bid for the keywords apparel, footwear, and accessories. However, it turns out accessories can mean a lot of different things and when we used DKIs, we would end up having ads that are relevant to all of them. Even if they search for audio accessories or auto accessories.
Not only did we waste money on driving irrelevant traffic, the bad ads also affected out quality score (landing page experience went down), which increased the CPCs. It may sound obvious that accessories was too broad a keyword to use, but in many cases, it is difficult to find alternative interpretations of the keywords one bids for when your mind is fully occupied with the topic.
These learnings led me to experiment with DKIs to see if my disregard for DKIs was grounded in just one recent, bad experience (availability heuristics) or if they could actually be justified from performance. I decided to test three hypotheses:
- Will DKIs cause higher click-through rate through more relevant ads?
- Will traffic from ads with DKIs have the same conversion rate as non-DKI ads, despite the risk of misaligned expectations to the landing page (as in the example above)?
- Will DKIs result in decreased CPC (as a function of higher relevancy and hence better QS)?
My test sample is data from 3756 pairwise similar ads with the only difference being the line including the DKI or alternatively a static line using the keyword in the ad group, but not necessarily the entire search query. Each ad group would only include very similar keywords (plural, singular, uses with/without hyphens, typos, etc.) The ads had combined over 100.000 clicks and close to 5 million impressions. I tested for DKIs in either headline or description line 1.
Before applying these numbers to your own campaigns, bear in mind that the account was fresh, so there was no optimization in negatives, etc.
Hypothesis 1: Click Through Rates
The data for the test showed that click-through rates generally were higher when ads did not include DKIs, no matter if I included them in the headline or in the description line 1.
Click through rate impact with and without DKIs
|DKI?||Headline||Description Line 1|
Interestingly, already before optimization, the CTRs without DKIs were superior. This is particularly interesting given that ads with a fixed text should benefit more from continuous optimization that DKI ads (given that there are more characters to optimize).
Hypothesis 2: Conversion Rates
When I tested the impact on the conversion rates, the picture was very much the same as what I found above. Conversion rates were higher without the DKI than with it for every single test. As I can’t make the conversion rates public, I have presented the relative performance of the ads without DKIs to the ones with.
Conversion rate impact with and without DKIs
|DKI in:||Headline||Description Line 1|
|CR impact without DKI vs with DKI||3.2x||3.1x|
Throughout the testing, ads without DKIs consistently had a significantly higher conversion rate. Here is of course the challenge that ads with DKI may appeal to the wrong users – the users who are not interested in our products. With that in mind, optimization (particularly with negatives) could potentially impact the conversion rate of DKI ads more than of the non-DKI ads.
Hypothesis 3: Cost per click
Besides from volume of traffic and the quality of traffic, the last important attribute of traffic from any ad is the cost. Many defenders of DKIs argue that DKIs should give the highest possible relevancy given that they mirror the search query. My stand point has always been that I find it hard to believe that Google would make a mechanism that would simply mirror the search query and then give it the full value of matching the query since the offering will always be non-tailored and the risk of mismatching is big. Google generally doesn’t tend to reward the lazy.
CPC impact with and without DKIs
|DKI in:||Headline||Description Line 1|
Counter to my expectation, cost per click did indeed go down when the DKI was placed in the description line 1. This could be explained by the fact that ad groups used were all quite long tail and therefore – in most cases – very specific to the product. On the other hand CPCs were lower without DKIs in the headline.
Conclusion: DKIs do not have a positive impact overall!
This study is only of ads in one industry and for e-commerce. However, the findings are in line with my expectations (or fears) prior to the study, with the exception of the impact on CPC for KPIs in the description line 1. That said, I would still argue against the use of DKIs, as the decrease in CPC doesn’t outweigh the decrease in CR. Overall CPA is still better in the case with no DKIs at all.
Mind you that the impact of DKIs may differ across industries, business models and countries, so it is hard to draw any universal conclusions, however the data seems to back up my initial intuition about DKIs. I still suggest being very careful in the use of DKIs and avoid them where they do not add major benefits.
Please feel free to share your own experiences with using DKIs!
Good luck with your continuous testing!