Twitter KPIs, Including The Twitter Quality Score (TQS)

We hear an awful lot about Twitter, don't we? @Oprah is on Twitter, with more than a million followers. CNN Breaking News has 1.4 million followers. It seems like you can solve all of your marketing woes by simply having a presence on Twitter. And then, you're reminded that customers buy from you because they like the merchandise you sell.

Of course, there's the rest of us, the 3% of the United States population that have a Twitter presence, and have somewhere south of a million followers.

@minethatdata is a tad north of 500 followers. But even with a modest audience like mine, there are many interesting metrics / KPIs that can be derived from the profile of those who follow me. Let's see what the faithful 500 look like:

Median Number Following = 281.
Median Number Followers = 273.
Median Number Updates = 158.

These are humble numbers, folks. And there's nothing wrong with humble numbers. Few people publish numbers like this, nobody would pay any attention to Twitter if this were the published promise of the tool. The numbers reflect the reality of the Twittersphere, not the reality of the Twitterati. And it's just fine, isn't it?

Let's put the number of people my audience follows into decile cutpoints.

• 1st Cutpoint = 39.
• 2nd Cutpoint = 73.
• 3rd Cutpoint = 123.
• 4th Cutpoint = 190.
• 5th Cutpoint = 281.
• 6th Cutpoint = 406.
• 7th Cutpoint = 624.
• 8th Cutpoint = 993.
• 9th Cutpoint = 1844.

Similarly, let's put the number of followers my audience has into decile cutpoints.

• 1st Cutpoint = 33.
• 2nd Cutpoint = 67.
• 3rd Cutpoint = 114.
• 4th Cutpoint = 186.
• 5th Cutpoint = 273.
• 6th Cutpoint = 386.
• 7th Cutpoint = 544.
• 8th Cutpoint = 888.
• 9th Cutpoint = 1321.

At the 7th cutpoint, the number of followers and the number of those followed begin to diverge. It's darn hard to encourage large numbers of people to follow you! Also notice that the top decile for followers has at least 1,321 followers ... a credible number, but not what the Twitterati might have you believe (the top user had more than 600,000 followers ... @zappos).

Here are the decile cutpoints for the number of updates:

• 1st Cutpoint = 8.
• 2nd Cutpoint = 25.
• 3rd Cutpoint = 54.
• 4th Cutpoint = 91.
• 5th Cutpoint = 158.
• 6th Cutpoint = 254.
• 7th Cutpoint = 370.
• 8th Cutpoint = 687.
• 9th Cutpoint = 1,111.

There's a lot of diversity here. We have folks who update infrequently, we have folks updating daily, and we have folks updating ten times a day.

Let's see if we can draw inferences from the metrics / KPIs.

I created a very simple model, trying to predict the number of followers based on how many people others follow, how many updates a person has, and the order in which the individual chose to follow me (1 = 1st, 500 = 500th). Here's the equation:

• Followers = 55.5 + 0.788*#Following + 0.161*#Updates - 0.128*Order.

Let's discuss the business intelligence embedded in this equation. Remember, one key reason we build models is to gain business intelligence, not necessarily to predict the future.

• For every 100 individuals you follow on Twitter, you'll earn 79 followers.
  
• For every 100 updates you have on Twitter, you'll earn 16 followers.
• Each additional follower you earn tends to have fewer and fewer followers ... to be expected as a social media tool grows in popularity.
  

Your mileage may vary ... just do the work and see what the data tells you!

I was very interested in the relationship between updates and followers ... each blog post I write results in 1.5 followers. Each Twitter update results in 1.1 followers. For my audience, each update results in 0.16 followers. No numbers are good or bad ... they're just interesting to track and to think about, and they are representative of the objective of the user.

The relationship between followers and following folks is interesting. The data strongly suggest that if you want to build an audience, you follow other individuals, the current "best practice". I do not execute this strategy, simply because I want to "test" alternate strategies.

Let's create a new KPI, called Twitter Quality Score (TQS), calculated as (#Followers / #Following). The "best" Twitterers, in terms of content, should have more Followers than those Following them, leading to a Twitter Quality Score (TQS) of 1.01 or greater. Here are the decile cutpoints for the TQS:

• 1st Cutpoint = 0.49.
  
• 2nd Cutpoint = 0.60.
  
• 3rd Cutpoint = 0.67.
  
• 4th Cutpoint = 0.77.
  
• 5th Cutpoint = 0.84.
  
• 6th Cutpoint = 0.93.
  
• 7th Cutpoint = 1.03.
  
• 8th Cutpoint = 1.21.
  
• 9th Cutpoint = 1.64.

Clearly, it is hard to build an audience ... 70% of my followers follow more people than follow them.

The data suggest that if you want to look for quality, look for a Twitter Quality Score of at least 1.50, if not greater. If you see a TQS value of this magnitude, then the author is probably publishing quality content that is appreciated by the Twitterati. Pay close attention to a TQS of 6.00 or greater, these values are in the top 2% of my following.

Finally, I look at every individual who follows me on Twitter. Here's what is interesting:

• Those who hire me are Catalog CEOs, then Retail CEOs, then Online Marketing CEOs.
• Those who follow me include Entrepreneurs, Web Analysts, Members of the Vendor Community, Online Marketers, Data Miners & Business Intelligence Analysts, Retail Marketers, E-Mail Marketers, and last = Catalog Marketers. The relationship is nearly opposite of the folks who I am purposely speaking to, in order to make a living.
  

Twitter may be a flash in the pan, it may be a valuable marketing tool, it may be a place where folks chat. Regardless, Twitter is a wonderful laboratory to experiment in. Build the KPIs for your own following, or do this type of study with your favorite Twitterer to see what the data tells you. No Twitter metrics are good or bad. Simply explore the data, folks!

Complimenting Web Analytics With Customer Insights

Your CEO makes a rare and unexpected appearance at your cube. Her appearance causes you pulse to increase, and causes your mouth to become dry. You quickly close out of the browser that is looking at The MineThatData Blog, and spin your chair in the direction of the inquiring executive who earns twenty times your salary.

Your CEO asks a simple question. "Why is our business missing expectations by ten percent?"

Here is where incomplete knowledge hinders the humble web analyst. The CEO might want to hear an answer like this: "Customer acquisition activities are meeting expectations. However, existing customers are not visiting the site as often as last year, not repurchasing at the same rate as last year, and are purchasing merchandise at lower price points than last year, contributing to our shortfall. We believe lower e-mail open rates are contributing to lower levels of site visitation."

Your CEO does not want to hear this: "We are seeing a fourteen percent decrease in organic traffic, partially offset by a seven percent increase in traffic from paid search. Conversion rates have actually improved, from 4.07% last year to 4.13% this year. This increase occurred because the mix of new and existing customers skewed toward existing customers, coupled with minor improvements in shopping cart abandonment. Our CPA improved, from $29.48 last year to $28.97 this year, illustrating improvements in response to portal advertising and favorable changes in the mix of partners in our affiliate program. E-mail open rates continue to decrease, in accordance with well-documented industry trends, resulting in a significant decrease in overall productivity."

The latter response actually demonstrates superior knowledge of the business than the first response. The first response tells a story that is easier for the executive to digest and act upon.

All too often, web analytics folks have an incomplete view of the business, caused by a failure of web analytics tools to view customer behavior across time. Web analytics tools are great at demonstrating what happens within a visit. These tools do a terrible job of illustrating what happens to one customer over time. Couple an incomplete view of the customer with an overly technical dissertation of business results, and you have trouble.

Web analytics gurus would be well served to forge partnerships with customer insight analysts. You know who these people are. They are the folks who sit on a different floor of your building, writing bizarre programming code in some obscure language called "SAS" or "SPSS". These are the people the CEO visited back in 1994 when a catalog wasn't meeting expectations.

The goal of this partnership is to combine metrics across platforms, so that all analytics individuals have a unified understanding of customer behavior, and can develop a more complete story about customer behavior.

There are a series of metrics that could be generated by this forged partnership. Let's explore some of these metrics. The following list is by no means exhaustive. Feel free to add metrics in the comments section of this discussion.

Probability Of Visiting Site, Existing Customers: Compare the rate at which last year's customers are re-visiting the site this year. If 94.2% of last year's customers are visiting the site this year, compared with 92.4% last year, you know that you are getting people to at least visit your website at better rates that last year.

Visits Per Existing Customer: Among those who do come back to your website, how many times are they visiting this year, verses last year? If existing customers who do come back to the website visit 7.42 times this year verses 9.82 times last year, you might have a problem with marketing efforts to drive customers back to the website.

Retention Rate: Web analytics tools are not well-suited to illustrate customer behavior across time. Work with the customer insight team to measure the difference in retention rate. For instance, assume that year-to-date, 63.2% of last year's buyers have purchased again, whereas last year, 68.8% of prior year buyers purchased again. This tells you that your existing customers are less loyal than last year. You can compare this metric with visits per existing customer. In this case, customer visit rates may be causing the reduction in retention rate.

Orders Per Buyer: Analyze how many times each of your retained customers purchased this year, compared with last year. If retained customers purchased 3.26 times this year, verses 3.09 times last year, you know that the customers who are purchasing are actually increasing their loyalty.

Units Per Order: Measure how many items a customer purchases, when they buy something. As an example, assume that this year, customers are purchasing 2.48 items per order, whereas last year, customers were purchasing 2.31 items per order. This suggests that customers are willing to purchase more merchandise, a good thing!

Price Per Item Purchased: The price of each item is an important metric to measure. If the price of an average item purchased is $38.42 this year, and was $42.99 last year, you know that customers are purchasing less-expensive items, offsetting the fact that they are purchasing more units per order.

Average Order Size: This is a simple multiplication of Units Per Order by Price Per Item Purchased. In our example, the AOS is $95.28, verses $99.31 last year. Customers are spending less per order, because they are purchasing less-expensive items than last year.

New and Reactivated Customers: This is a simple count of the number of first time customers, and number of reactivated customers (those who haven't purchased in several years. Assume that this year, 125,000 new customers and 35,000 reactivated customers purchased, whereas last year, 85,000 new customers and 25,000 reactivated customers purchased. This tells you that there are significant improvements in getting new and infrequent customers to purchase from the website. The analyst needs to take this a step further, measuring the orders per buyer, units per order, price per item purchased and average order size for these two audiences.

By combining information from a web analytics tool with information from the customer insights warehouse, a more complete story can be told regarding customer behavior. In this case, the problem seems to exist among getting existing customers to come back and visit the site on a frequent basis. The web analytics team and the customer insights team can work together to understand the reasons that are causing this problem.