Misuse of Big Data Can Cost You at the Cash Register

The good folks at BloomReach shared some data with me recently. This data showed how “gift” oriented search queries differ from other types of queries. In the process of reviewing this I realized that it provides an excellent example of how drawing premature conclusions from data can cause you to make bad mistakes.

As background to the source of the data, BloomReach provides a product called BloomSearch that is in use by a large number of e-commerce web sites. The product enables those sites to scalably modify their product pages so that they can capture a lot more long tail search traffic, resulting in significant incremental revenue.

As a result of this, BloomReach has access to lots of information on how these sites perform. Let's take a look at a sample of the data!

This shows data for 9 anonymous e-tailers. We see the bounce rates shown for 2 different types of queries – “gift” and “non-gift”. BloomReach found that gift queries contain certain obvious terms like “gift” or “present”, or sometimes not so obvious things like “mother's day flowers” or “Valentine's Day chocolates”.

Some gift queries occur at the same time every year, while others are unpredictable and ongoing (i.e. birthdays and anniversaries). For example, “housewarming wine basket” is another example of a gift query that was included. “Non-gift” queries represent all other queries.

For 7 of the 9 e-tailers, the bounce rate went up. For e-tailers 1, 3, 5, 6, and 9, the increase in bounce rate was quite dramatic. So this means that these sites should not target gift queries, right? After all, an increase in bounce rate suggests a poor quality experience for the user. Ok, great, let's stop optimizing these pages for gift queries, they don't work on our sites!

Not so Fast!

Before we make a rash decision, let's look at some more data. Here is the conversion rate for gift and non-gift queries for the same 9 e-tail sites:

Conversion Rate by Search Query Type

This tells a very different story. The conversion rate on these queries increased for all 9 e-tail sites. In fact, for 7 of these etailers, conversion went up between 14 and 70%. Wow. Let's take a look at the data in bar chart format for emphasis:

Bet you're glad you didn't change your pages to stop emphasizing gift queries now!

Takeaways

I have 5 conclusions for you:

1. Lower bounce rates do not necessarily mean more conversion. In this case, bounce rate and conversion rise together for this class of queries.

2. We also see that data can be mis-used. No single measurement of user behavior is likely to be sufficient. You need to step back and look at the bigger picture before making decisions. We all understand that intuitively, but this data shows us a great example of one real scenario where you could make a mistake if you are not careful.

3. I believe the reason why these contradictory signals occur is that people who enter gift related queries already have a high degree of intent. Basically, they want to take immediate action. As a result, we see two things happening:

  1. If they do not quickly see what they want on the first page, they're out of there right away. They will be faster to draw a conclusion on whether or not you are going to be able to help them. You might say that they will be more judgmental.
  2. Since they are on a mission, if you have the goods they want, or something close enough, they will pull the trigger quickly. No need to go see another site, as you have helped them complete their mission.

In fact, Akanksha Baid of BloomReach says, “this same phenomenon that we see with gift queries can manifest itself for all queries with strong intent. For example, “products for clogged sinks”, “sinus relief medicine”, or “dresses for a beach wedding” each express a very strong intent. Determining which queries have strong intent and capturing that intent with relevant pages is key to improving conversion rates for e-tailers.”

4. If you weren't already passionate about landing page optimization, this is one more piece of data that underscores how important it is to invest heavily in that practice. People who are on a mission to buy something appear to make snap decisions. Make sure they see that they can get what they want quickly!

5. This also highlights just how difficult it would be for search engines to use “simple” bounce rate as a ranking signal. By this I mean the traditional Analytics bounce rate, where a user visits only one page before leaving the site. By itself, bounce rate has a lot of drawbacks as a ranking signal.

More sophisticated metrics such as people who return to the SERPs quickly, and then click on a different result make much more sense.

zp8497586rq

Comments

  1. says

    Hi Eric – Great stuff as always. I recently saw the quote “Every good metric deserves a counterbalancing metric” (maybe you said it?!).

    What I think your post also illustrates so well is the absolute necessity of analysis (in your third takeaway). Sounds obvious but sometimes in the harried publishing world, at least, we see people throwing out numbers as if the numbers tell you what to do. Numbers aid analysis, rather than replacing it – that’s my mantra. So thank you for this excellent illustration of the point.

  2. says

    Hi Eric -

    Excellent post. I think that your conclusions are right on target based upon the data your provided. I do have some questions, though, regarding the data itself.

    The numbers for bounce rate seem “off” to me. Specifically, the bounce rates across the different retailers seem abnormally low. Also, are the conversion rates that you share percentages? In other words, does your data suggest that each one of these nine retailers have a greater than 50% conversion rate for gift related keywords? Is the demoninator for the conversion rate visits or visitors? Either way, seems too high.

    Generally speaking, these numbers provided by BloomReach seem suspect to me and my “spidey-sense” questions their accuracy.

    Best,

    Yehoshua

    p.s. Great meeting you last week in London.

  3. says

    Great article, always looking and striving to find ways to decrease bounce rates. I have found having websites that serve some form of interaction or search function help this the most.