My post last week about the shortcomings of Klout got several thousand views and generated quite a bit of discussion. it also got me several e-mails from companies that claim to have built a better mousetrap than Klout. I haven’t reviewed these tools in detail just yet, but it appears that influence is a red-hot topic in PR and marketing circles right now.
Influence measurement is a natural evolution of conversation monitoring, a discipline that’s personified by Salesforce.com’s Radian6 tool and dozens of competitors. Monitoring is a solid practice that can keep you in touch with the topics and brands people are discussing online. Most tools now also provide some degree of sentiment analysis, which attempts to derive attitudes from comments. Sentiment analysis is devilishly difficult to get right, however. If a teenager calls something “sick,” it’s a compliment. Coming from a 50-year-old, it’s an insult. Most experts I’ve spoken to on this topic say that sentiment analysis tools are at best 70% accurate.
Topaz Partners has developed the "Circle of Influence" to depict the different factors that go into decision-making (Click to enlarge). TopazPartners.com
This isn’t stopping vendors from tackling the even more complex issue of Influence analysis. This goes beyond sentiment analysis to attempt to determine a person’s ability to drive action. The problem is that there are lots of variables and intangibles to influence that resist being boiled down to a single number. For example:
- What is action? A “like” or retweet is a form of action, but not necessarily one that leads to a decision.
- Online actions have different gravity depending upon the stakes and the effort involved. Writing a comment takes more effort than clicking a “like” button. Posting a blog entry referencing someone else’s words is more involved than writing a comment.
- Which actions really matter? I have yet to see a tool that can correlate influence with purchases or donations with any degree of certainty. We assume that conversation about a topic influences decisions, but are they the decisions we want? A lot of people have been talking about Hewlett-Packard lately, but I doubt it’s driving profitable sales of HP products.
- Influence is contextual. If I’m considering buying a Yamaha stereo and find a blog entry from someone who exhibits deep knowledge of the model I’m considering, that person may have a disproportionate influence on my decision, regardless of the number of followers or subscribers he has. The weakness of most influence analysis tools is that they abstract broadly, looking at things like reach and amplification. However, decisions are more likely to be influenced at a micro, rather than a macro level.
One of the most illuminating books I’ve read on this topic is Influencer Marketing by Duncan Brown and Nick Hayes. The authors argue that the influence of media in general, and social media in particular, is greatly overrated. They count no less than 50 kinds of influencers, ranging from resellers to academicians to government officials. Most of them have little or no online visibility, but their knowledge, leverage and/or connections make them enormously influential. What’s more, the larger the purchase, the greater their influence tends to be.
I don’t agree with everything Brown and Hayes say, but I commend them for resisting the urge to oversimplify. Their basic message is that influence and audience are two different things. Celebrities can have huge audiences but little power to affect decisions. Conversely, people with very deep knowledge can have small audiences and great influence. Seth Godin said it well: Small Is the New Big.
In the mainstream media world, audience was associated with influence because we had few tools to understand the true dynamics of decision-making. Our natural tendency is to apply this same metric to online conversations. The danger of this approach is that social media is more about quality than quantity. In the same way that early automobiles had steering mechanisms that mimicked reins, we are applying old assumptions to a new medium. I’m not saying that influence measurement tools are inherently unreliable, but they are attempting to measure what may be immeasurable. Just be skeptical.
The best takeaway from this: “I’m not saying that influence measurement tools are inherently unreliable, but they are attempting to measure what may be immeasurable. Just be skeptical.”
Skepticisim, critical thinking, may be our best tool in today’s wild world. Automating the measurment of influence trys to make an art a science. This diminishes the value of our better instincts and such non-trivial personal strengths as knowledge, experience, intuition. It is a fool’s errand. Numbers can lie and liars can number.
“Numbers can lie and liars can number.” <--You have a gift for turning a phrase!
Hi Paul,
Good article and great thinking on the subject. Trying to measure anything related to things that influence human behavior is certainly difficult. Your points about influence and reach (or audience or popularity) being different is an important point, one that I try to emphasize to brands all the time. That’s why I think any automated measurement of influence must be used smartly, not as a silver bullet, and must first start by looking at things topically: someone who is popular gets a lot of engagement when they speak about wine making, does not make them influential about sports cars.
I do think that the tools for measuring influence can be helpful to uncover things, but the user shouldn’t use them blindly as an “all seeing” final authority on the subject. I agree with the statement that not all social actions lead to decisions – which is why we (Radian6) focused on enabling our users to adjust the weightings of various social actions, essentially tuning the influence algorithm for their situation and determining which social actions are more important indicators of influence (i.e. if you think a “like” is worth less than a comment, then you should give comments a higher weighting). I always felt that there was no perfect algorithm, so users needed flexibility to adjust.
On your point of correlating influence with purchase decisions… that can be done actually, but only in part. We had customers asking about this, so we added the ability to incorporate web analytics data back into our influence score, such as a purchase or other website conversion event. This way you see which influencers sent you traffic that actually converted and appropriately incorporate those actions into their influence score along with the other social actions (likes, comments, retweets, etc.).
However, as you say, it is only a tool and will never be perfect. What about all those who were influenced but made their purchases offline or through some other means where they left no digital breadcrumb trail to follow? What about the fact that my decision to buy was influenced by the fact that I keep seeing/hearing about some thing in multiple places so it wasn’t a single influencer but the overall conversation level that is causing me to consider buying? What about….? No tool will ever capture every real world scenario of what influences people’s thinking or buying decisions, but they can be very useful when used appropriately.
Good to see you writing on the subject. Hope things are well with you, Paul.
Cheers,
Marcel
Salesforce/Radian6
Marcel:
Thanks for your detailed comments and for the update on what Radian6 is doing to give your customers flexibility in tuning their views of influencers they monitor. You’re right that in some cases online activity is a perfectly legitimate measure of influence. You’re also right that in other cases it’s almost useless. What’s important is that customers understand the difference and are able to make intelligent choices. The market needs education on this issue, not more grades and numbers.
Hi, Paul. I enjoyed this post and the previous, and share many of the same points. We use Radian6 where I work for general social media monitoring, but have elected to not use the influencer tools for many of the same reasons (though I will give them a second look). I find the sentiment rating to be horrific at best – I like your example. I agree with your numbers, but conversely (70% *inaccurate*). This is no slight on Radian6 – I think many who reply on NLP suffer the same fate. True human sentiment scoring is better, but also slower and more expensive. I also looked into two other tools (Socmetrics and Traackr) and will be moving forward with one of them to help understand who are the influencers in our space. But all the while, no matter what tool I use, I will take every results with a grain of salt. I think if users in general can keep in mind that a tool helps a process (and it isn’t “the answer”), then it can be useful. A cup or two of skepticism into the recipe helps as well.
Alan: A personal anecdote in that area. I was doing some sentiment analysis for State Street Bank and noticed a huge spike in negative sentiment. Upon looking at the source information I discovered that the spike was caused by coverage of a shooting on State Street in Boston. For some keywords, the domain of potential mentions is so large that it’s nearly impossible to automate the process.