Data Science

The Missing Link of Influence: The Potential to Influence, and Be Influenced

In my previous posts, I defined influence and discussed why brands don’t seem to understand digital influence. Today, we are ready to talk about the missing link in the influence industry. This article builds on the previous two, so I would recommend reviewing the following posts if you missed them earlier:

  1. What is Influence, Really? – No Carrots, No Sticks, No Annoyance, No Tricks
  2. Why Brands STILL don't Understand Digital Influence?


What’s Wrong with the Current State of Influence Analytics?

Last time we explained why nobody can actually measures real influence on social media. So influence vendors must build models that predict someone’s influence in order to compute their influence score. The problem is that most of these influence models focuses on the influencer. Nearly all models are focused on estimating the influencer’s social capital. Therefore an influence score is merely a prediction on the influencer’s potential to influence.


school_leader_800_clr_5247b375.pngHowever, we also know that no actual influence occurs until the influencees produce a measurable action. Without the influencees, there is simply no influence. This is what is surprising and mind boggling to me. If real influence depends so much on the influencees, why are most vendors still so focused so much on the influencers?


It is partly because there are significantly fewer influencers than influencees, so it is much easier to collect, analyze, model, and compute the activity data from the influencer than from the influencees. It is not difficult to spot an influencer, because they are the ones that typically stand out right in front of your eyes when you look at their activity data. However, it is much harder to identify the influencees. They are pervasive. They easily outnumber the influencers by hundreds or thousands of times. Influencees are also inconspicuous despite their abundance. Their activity data are weak and typically masked below the noise level.


However, a valid influence model must consider the influencees. In fact, the accuracy and effectiveness of any influence scoring algorithm depends on how much influencee attributes it considers and how close these attributes are to the influencees. But again, no one measures whether an influencee’s is actually being influenced (i.e. real influence data on social media do not exist). So these influencee attributes can only help predict the influencees’ potential to be influenced.


The Big Missing Link of Influence

So when does real influence take place? To be absolutely rigorous and correct, no one knows for sure. But we do know when influence will occur with high probability. That is when the influencer’s potential to influence is aligned with the influencee’s potential to be influenced.


weak_link_800_clr_4405b300.pngBecause all that influence vendors have is a score that is indicative of the influencer’s potential to influence, there is a missing link in the influence industry today, and it is the influencee’s capacity to be influence. A couple years ago when I introduced a simple influence model, I found four categories of attributes that characterize the influencees’ likelihood to be influenced:

  1. content relevance
  2. timing (or temporal relevance)
  3. channel alignment
  4. confidence (a.k.a. trust)


Today, several influence vendors have implemented timing and channel alignment, and to a much lesser extent relevance. However, they have implemented these as attributes of an influencer, whereas they should be attributes of the influencee. For example, influence vendors treat timing as an influencer attribute, which indicates when the influencer’s frequency of communication changes. But the timing that I talk about is an attribute of the influencees, and it characterizes the temporal window within which the influencees are susceptible to being influenced.


No wonder, brands don’t get influence, because I don’t even think the influence vendors get influence. In reality, an influencer’s potential to influence has little correlation with his frequency of communication.


For example, President Obama’s potential to influence is the same regardless of whether he communicates or not. His actual influence (i.e. how many people he actually influenced) does change depending on how much he communicates. But influence vendors do not measure actual influence; they can only estimate someone’s potential to influence. By including the timing factor as an attribute of the influencers just tells me that even the influence vendors do not understand the difference between the potential to influence and real influence.



  1. Most influence vendors focus on measuring the social capital of the influencer, but real influence occurs when a change in thought or behavior is produced in the influencee (i.e. person being influenced)
  2. This is the missing link of influence: the link from the “potential to influence” to the “potential to be influenced.” Real influence can only occur when there is an alignment between these two. In fact, this is the minimum required state


Influencer marketing has huge potential. But as an industry we are far from realizing this potential. Partly is because influence is a challenging concept that involves much more than just the influencers themselves. Moreover, there are many big data and analytics challenges in accurately estimating someone’s potential to influence. But first, influence vendors need to start incorporating more attributes of the influencees into their model in order to improve the accuracy of their influence score.


Next time, let’s dig deeper into the algorithms that score influence. In the meantime, if you also do research on digital influence, I’d be happy to discuss your findings here.



Michael Wu, Ph.D.mwu_whiteKangolHat_blog.jpg is 927iC9C1FD6224627807Lithium's Chief Scientist. His research includes: deriving insights from big data, understanding the behavioral economics of gamification, engaging + finding true social media influencers, developing predictive + actionable social analytics algorithms, social CRM, and using cyber anthropology + social network analysis to unravel the collective dynamics of communities + social networks.


Michael was voted a 2010 Influential Leader by CRM Magazine for his work on predictive social analytics + its application to Social CRM. He's a blogger on Lithosphere, and you can follow him @mich8elwu or Google+.


Lookig forward to the follow up article. This reminds me of the top down and bottom up paradigm. Nice read, thanks!
Data Science

Hello Orlando,


Thx for the comment. Glad you find it interesting.


Yes, many of my models are developed using both top-down and bottom-up reasoning. And then validated independently. So you are totally correct.


Stay tuned and see you next time.


Hi, Mike! 

Like you are back to influence topic Smiley Happy

You know in social science influence does mean ability to cause action (Solis is sociologically educated - As I too - and in sociology influence is similar to power definition).


And your categories does not reflect this impact of influence (action), but only conditions, that may cause action with some probability. Or are this impact "packed" somewhere inside these categories? 

Fantastic read. Thank you for making that distinction. 

Data Science

Hello Andrei,


Thanks for the comment and glad to see you here.

As you might know. My blog got relocated and got a new URL. But thanks for coming back.


With respect to the definition of influence. Please take a look at my definition here:

What is Influence, Really? – No Carrot, No Stick, No Annoyance, No Trick


Some misconceptions of influence is that, not all abilities to cause actions are true influence. For examples, those described in the linked article above (i.e. carrots, sticks, annoyance, tricks).


Moreover, not all influence requires an action. If someone changes your mind, opinion, or sentiment about something, they’ve influenced you, while you have not taken any action. The action requirement only makes it easy for people to measure. But there are a lot of things that are  not measured that are important. For example, as people move down the purchase funnel, they change their mental state: unaware --> aware --> interested --> desire --> then finally action. The early stages of the funnel may not result in any action. But they are still important. And marketers have ways to infer these changes in awareness, interest and desire, even though there is nothing that actually measure people’s change of mind. Action is only the most profitable kind of influence at the bottom of the funnel.


If you read my last article, you should know that no platform actually measures influence. People can only measure the probability of potential influence and claim those to be real influence. I just like to make this point clear.


Hope this address your question.

Thank you for asking the question.


Data Science

Hello Sharlene,


Thank you for the nice comment. Glad you find this distinction interesting.

More post about influence coming soon. Stay tuned and hope to see you again next time.


Thx, Mike!

For me action is not last stage on AIDA journey (purchase). All change in mind, opinion, or sentiment can be represented in some action. Negative tweet or advice tweet are results of changed opinion and sentiment. You may hear smth about performative arrpoach in language science, when words are "real" actions. 

Data Science

Hello Andrei,


Thank you for continuing the conversation.


Yes, definitely the action in AIDA is not the last stage of the consumer journey. Some even say that the funnel is really a cycle. That is also true. No one model completely describes all the peculiar behavior of every consumer. These are all just models.


Although changes in mind may result in some action, such as recommendation, as you mentioned, not all changes in mind result in those actions. So simply counting negative sentiment is not sufficient to “measure” the early stages of influence. They must still be inferred. Moreover, a recommendation may also come from consumer after the purchase, and there is not a good way to determine whether someone has purchased the iPhone 5 if they simply tweet “iPhone 5 is great, I totally recommend it.”


Although most people will probably say something negative about their bad experience, a lot of people may not say any words when they are happy about something. So marketers still need to infer changes of mind, which means change of mind is still important and valuable, not just action. Action simply make it easy for someone to measure it.



Michael, I'm a huge fan of your work and this article is no exception :-) However, if I did have to ask a question it would be...

"Is this the right conversation to be having?"

Even if we had the perfect influence algorithm, which we clearly don't, influence can never be represented as a single value. Its always contextual, relational, temporal, combinatorial, ... And not only that , but its also totally dependent on the question you are asking. For example, if I am looking to buy a smart phone l am not likely to trust an Apple "influencer/expert", however If I was looking for a good case for my Apple smart phone he might be my first choice. For one question, he is indeed an influencer but for the other he's not.

I had a wee rant about this on my own blog:

Data Science

Hello Marie,


Thank you for the nice comment and being a fan of my work.


I apologize for the late reply. I’ve just return from NYC, Boston, and Philly. So it’s been pretty hectic last week dealing with scheduling and meeting in the midst of Sandy’s aftermath + the nor’easter storm. And now I have to deal with my exploded inbox. Anyway.


Your question is definitely very valid. Context is everything. It determine whether one person will be susceptible to another’s influence. As such I’ve said over and over again that “there is no such thing as a universal influencer.” Influencers are topical specific, channel specific, and as you mentioned, context specific. As such there shouldn’t be a single score that characterizes a person’s influence.


If you look at my influence model, it actually try to capture this context specificity. The models basically says that there is a high probability of influence occurring when ALL 6 factors are high.

  1. Credibility
  2. Bandwidth
  3. Content Relevance (the right information)
  4. Timing (the right time)
  5. Channel Alignment (the right place)
  6. Trust/Confidence

Basically factors 3, 4 and 5 provide the context that you are talking about. If the influencers are talking about a different content, then he’ll have a different influence score on that topic. Even with the same topic, his influence score actually changes over time and differs on different channel.


That is the type of influence algorithm that I envisioned. Maybe that shouldn’t be call an influence score anymore, b/c it is not just one score. They are scored based on the context. People recommender may actually be a better term. Whatever people call it, really doesn’t matter to me, that is a marketing exercise. But I am a scientist. Ultimately, the algorithm that works must considers the contextual factors, and that seem to be missing in the industry now.


Thanks for the link to your discussion about people recommendation. It’s a good read.

Hope to see you again next time.


Thanks for the response Michael Smiley Happy And glad to hear you guys are all in one piece after Sandy. It looked like one hell of a storm.

Leaving the whole copying/influencing debate to one side, I agree with you that the focus on influencers is short-sighted. The holy grial of all this is determining when the subject is influenceable because until they are, all the influence in the world is going to have no effect.

Data Science

Hello JohnDodds,


Thank you for the comment.


You are right on. That is precisely the point that I feel that most influence vendors are missing. How influenceable the audience are is equally important as how influential the influencer is. When it comes to the propagation of influence, both of these quantities has to be high in order for real influence to happen.


Thank you for confirming this point.

I hope to see you around next time.

Michael, you should check out this link:


The link will take you to a report on the subject of, "Quorum Responses and Consensus Decision Making".


My reading of the report suggests that what you are attempting to model inside a social network naturally occurs inside the animal kingdom. 


Abstract: "In the absence of centralised control, arriving at a consensus depends on local interactions in which each individual's likelihood of choosing an option increases with the number of others already committed to that option. The resulting positive feedback can effectively direct most or all group members to the best available choice."


I think the "mass" in mass markets is incredibly important. We use other people's decisions to shape and inform our own. 


Best, David


Data Science

Hello David,


Thx for the comment and the linked article. 


Although I know this phenomenon, and the dynamics of quorum response is very interesting, I don't think it is directly related to what I am trying to model here.


I am trying to model what does it take for influence to propagate, but this articles talks about the likelihood of a group to make the optimal decision base on the other members of the group that is already committed to that decision. I can see how it might be the case if the interaction between the members are fairly strong.


However, this might be a mechanism to make the group more susceptible to influence when more members of the group is being influenced, and if we make the influenced members very visible to others and create some social pressure to conform. Nevertheless, I will take a look at it in greater detail to see if it is indeed applicable. Maybe I'll find something interesting.


Thank you for the suggestion.

See you next time.


Hi Mike. 

Yes action is one way to measure smth. Some big retailers register cunsumer's emotions and gesture by decision making on store shelves.

All types of changes are goinfgin mind, but not all are expressed - and measurable on valid base. Today we know that 90% of WOM is not digitalized - and so direct not measured (we can measure it a posteriori by surveys). Influence is reciprocal process as you pointed, but on both sides exist problem of expressivity and measurability. Clear a lot of these changes stay unmeasured  if people don't have good will to express that in structured way. 

I'm waiting your next articles and answers on these difficult questions )

Data Science

Hello Andrei,


Thank you for the comment, unfortunately, I have a bit of hard time understanding your English. But let me try to answer it anyway.


Yes, there are a lot of changes that are not measurable. That is precisely nobody really measures influence. If it can be measure through survey a posteriori, there is still not guarantee that the consumers will respond to such survey, So even if it can be measure, it doesnt' mean it will be accurate. However, this doesn't mean those un-expressed changes cannot be inferred from other measurable quantities. That is why we must build a model to estimate / predict people influence. And when you do that, you need an independent validation data for validate the model.


Next articles will come, but I may switch to another topic and come back to influence later.There are so many good stuff I want to share.


In the mean time you can take a look at my TechCrunch article on this subject: "The Irony of Measuring Digital Influence."

Check it out and let me know what you think.

See you next time.


Thx for great article Mike!

Sure your Irony is not nice to hear for most vendors. But it is truth, that they play with media metrics without some external validation based on independent measure as you state.

These services are growing on trend of social PR, where audience reach as such is a "metric" of succes. But when you have reach it does not mean you have ROI from PR. And even content consum (retweets or likes) are not strong signals user have read and valued retweeted or liked content.

A lot of things occur beyond media and they must be counted to demonstrate real value from these media quasi actions. Klout and another vendors have no or little tools to demonstrate real value gained from big accounts and opinion leaders when they wiil find it metrics iside media world of today. 

Data Science

Hello Andrei,


Thank you for the conversation, and recognizing the truth and reality out there.


Yes, there are too many hidden assumptions in influence metrics. So when it is unvalidated, we can't really say anything about it's accuracy and validity. That is why brands need to understand these critical concepts around influence before they can even talk about ROI. 


Obviously when events take place off media, then the problem is even harder.



Lithy Awards 2017

The winners in digital CX have been crowned!

See the winners!!