Stupid online measurement

Digital marketing is an uncertain environment. Apart from the new channels and modes of interaction with consumers, there’s also the ability to record, measure and report on everything. From mouse clicks to Facebook chats, everything that happens online is recordable, measurable and reportable. If we didn’t know it before (and most of us did), a number of intelligence agencies have shown us that recently. Since we’ve now entered the Age of Big Data and data mining, there’s good reasons for recording as much information as you can and processing it ready for use. However, no human organization can possibly use, or even want, all of that data all of the time. To misquote a great man:

“You need some of the data some of the time, but you never need all of the data all of the time.”

I discussed this with Bryan Eisenberg, one of the fathers of web analytics, who’s expressed similar concerns in his Smarter Data Manifesto (see: http://www.bryaneisenberg.com/the-smarter-data-manifesto/), told me:

“Don’t measure anything that you can’t find a direct line of sight back to your financial statements. You manage by what you measure so focus on those things that affect the bottom line directly and over the long term (brand metrics).”

What you need or don’t need is going to depend on your objectives, which will in turn determine what actions you take. So it’s axiomatic that you shouldn’t measure stuff you can’t do anything about. So don’t measure form performance if you’re not going to change the form, and don’t measure Facebook activity if you don’t have a Facebook campaign. In addition, there are a few things which people like to measure which are a complete waste no matter what your objectives are and no matter what you can action.

Social activity is the most blatant area of difficulty. At the foundation of web analytics is a void we can never fill. We can never really know what sort of person is behind the behavior web analytics records because people don’t visit websites, devices do. If you have a look inside the data your web analytics system is recording, you’ll see that the “people” it counted were really just IP addresses and system identifiers. You hope that system was being operated by a human, but you can’t tell. You hope it was just one person, but if people swapped over in the middle of the session, you’ll never know. It’s also possible that there was no person behind the device, there’s plenty of robots out there driving browsers now.

These robots have always been around, but it’s never really been a major problem before now. Before the days of social media there wasn’t much to be gained by writing a program which pretended to surf the web like a person. However, these days poor web analytics practice has created an entire industry designed just to serve up useless data. The problem stems from counting volume without any reference to quality or segmentation. The vast amount of traffic on the web disguises how varied is the mix within that volume. Raw totals are rarely of any use – the lack of detail disguises so much they might as well be wrong.

I’m going to suggest five things you can stop measuring. By all means continue to gather the data, but don’t bother reading the reports. These cover popular metrics in social media and search marketing which people spend a great deal of time and money chasing, but are a waste of time.

Don’t count Twitter followers

Let’s start with counting Twitter followers. It’s a completely pointless exercise. Let’s think about exactly what the number of Twitter followers represents. It is nothing more than the number of Twitter accounts which have connected. It’s not the number of people actively receiving Tweets. Up to half of all Twitter accounts are inactive while many are just spambots. People even buy fake spambot followers. For example, it is estimated that two-thirds of the twitter fans of President Obama, Justin Beiber and Lady Gaga are fake. So we have to ask: why would someone purchase fakes on a system whose sole function is to send text to people? Why would anyone want to pay money in order to talk at a machine? The only reason someone would do this is because many people think the number of followers means something. That’s fine if you’re a low-level blogger and want the naïve press to think you’re someone worth talking to once in a while. However, the raw number of followers is of no use for digital marketing. Not only has this misperception by the press spawned the Twitter spambot industry (as you can see at buytwitterfollowers.org [LINK: http://www.buytwitterfollowers.org] ), it’s now reached the level where an anti-spambot industry is developing, as we see with companies like Statuspeople [LINK: http://fakers.statuspeople.com/Fakers] and Twitter Audit [LINK http://www.twitteraudit.com/].

The only real way to assess the impact of Twitter on a brand is to use conversation analysis systems, such as Transana [LINK: http://www.transana.org/] to determine the sentiment being expressed. A group of genuine human Twitter followers will include detractors as well as supporters, so even if you do count active genuine humans, you still need to understand their feelings to understand what impact Twitter is having on market sentiment. The social influence estimator, Klout, takes a small step in this direction when it considers retweets and mentions, but even these can be robotically generated and it still doesn’t take into account what people are saying about you within those mentions.

Don’t count Facebook Likes

Likes are everywhere. Clicking them may indicate a real engagement with your brand or maybe you just amused them for a moment. There have been plenty of attempts to calculate the value of a Like. Estimates by companies which sell social media marketing range from $3.60 to over $100. Estimates by companies doing their own social marketing of the value they achieved range from 20c to $1. Estimates by independent web analysts tend to put it at zero. As Forrester Research’s Augie Ray [LINK: http://blogs.forrester.com/augie_ray/10-07-08-what_value_facebook_fan_zero] put it:

“the answer is zero — unless and until the brand does something to create value”

The vast majority of every website’s traffic is transient. This means the vast majority of people who click the Like button will never return to the site of their own accord. If you want to get value out of them, you need to actively do something. Given the diverse range of people and their intent when Liking you, there’s little you can do which will appeal to them as a single homogenous mass except offer them cash. If you want to do anything more sophisticated than that, you need to segment them and respond accordingly. This means the important numbers are how many fans you’ve got in each segment, not some grand total.

Forget Social Mentions

I’m sure it’s no surprise by now that I’m not a fan of social mention reports. A social mention is nothing more than the use of a keyword in some form of social media. The problem is, again, one of context. Without knowing how the word was used, in what context, to express what sentiment, a mere count is meaningless. It is possible, with some serious effort, to tune social mention systems to surrounded phrases and other contextual clues, but it requires a great deal of effort and few social mention systems are up to the task.

Who cares about Link Count?

Search engine optimization is another area where there’s a great deal of confusion. SEO consultants tend to fall into two opposing camps depending on how they think Google decides how to rank sites. One camp favors content while the other favors links. The link group hold that what’s most important the number of sites which link to you, so many chase a link count – the total number of websites (or pages) which link to a brand’s site. As we saw with Twitter, this created a spam industry. In this case the culprit was link farms – websites which existed solely to provide thousands or even millions of links to their customer’s websites. Once again, we can see that this is based on Google’s founding premise that a link was created by a human who had reviewed and approved the content at the other end of the link – an endorsement. The people behind the search engines aren’t stupid, some of them are even ex-rocket scientists, so once the link building industry arose, Google commenced a cycle of changes designed to filter out what it calls “low value” links. For example, Google counts the total number of links coming out of a site, and spreads the site’s total value amongst them. This means a link from a site with only five links is 200 times more valuable than a link from a site with 1,000.

Unless you have very few links, and can check them manually, or are scrupulous about your link building activities, your link count will mean nothing because it covers different types of linking sites, each with a different value. A single number tells you nothing about the impact on your listings. A large number of poor links can even count against you. In such cases, it can be better to have a few dozen hand-built links than thousands of link farms. If you want to count links, you need to segment them into different channels, different types of site so you can determine the value of the link each site provides.

Search Engine Visibility

The problem with search engine optimization is that it is incredibly fragmented. Activity within a search engine is segmented by search phrase or keyword. Most sites will have thousands or even tens of thousands of different keywords delivering traffic to them. The site’s ranking can vary for each and every one. Search engine optimization is therefore about improving the individual rankings of each of these phrases. This makes assessing the overall state of a site’s listings very difficult. It’s hard to make sense of a report containing a wall of phrases and their eve changing rankings, even harder to gauge progress. The industry’s response has been to develop the Search Engine Visibility Index, a single number based on a website’s rankings for many phrases. There’s no rigid standard determining how this is calculated, but the basic idea is that a #1 listing is worth a certain percentage, a #2 listing is worth something less, and so on down to #10 (no one cares what happens on the later pages). The calculations are set so that if you had all of your keywords at #1 you’d have a score of 100%.

Unless your keywords are all of equal value (and they’re not), this doesn’t tell you much. Many performance indicators, including bounce rate, form abandonment, average order value, engagement and conversion rate vary from search phrase to search phrase. I’ve seen site’s performance metrics destroyed because they were mistakenly ranked #1 for a set of popular phrases which had nothing to do with them. It resulted in huge volumes of inappropriate traffic which drove bounce rates through the ceiling and conversion rates through the floor.

You can only focus on a limited range of keywords at a time. It’s kind of comforting to measure changes in the search engine visibility score for these, but it’s not really much help for actually doing anything.

CONCLUSIONS

There are two threads running through the problems with these metrics. The first is the assumption that onlione activity represents human activity. That was true when the web started, but it hasn’t been true for years. It didn’t really matter until social networking arose and it became worthwhile creating software which could pretend to be human. The other thread is that of totals. Total numbers, for many web metrics, aggregate so many different segments that they become meaningless. Such numbers cannot serve as a platform for planning action, and if you’re not going to use these numbers to do something, why bother to collect them in the first place.

Remember – you don’t need all of the data all of the time, you just need the stuff you can use.