Earlier this month, we talked about sentiment analysis. In this post, I’ll dive into the topic of adding meaning to the numbers. Having a nice number of total positive mentions doesn’t have to be representative of how each customer segment feels about your organization, brand, or products/services.

Adding Up The Numbers

As discussed earlier this month, the most important thing to discern in sentiment analysis is the uniqueness of sentiment, i.e. whether the sentiments monitored/measured are unique to a person or is distributed through influence or virality.

Throughout this post, I’ll use the following example. A company has monitored the social web and has filtered out the unique sentiments that mention something about the company. A total of 5,010 sentiments were found. The company found 1,765 positive mentions; 2,357 neutral mentions; and 888 negative mentions.

From a totals perspective, this looks quite promising. When the neutral mentions are disregarded, the company gets nearly twice as many positive mentions as negative mentions. However, it’s time to put the numbers into perspective.

Adding Customer Segments

The company in our example makes a distinction between four customer segments, based on demographics and usage: (1) business, (2) families, (3) young people, and (4) luxury. The monitored mentions have been mapped according to these customer segments and the results are in the table below.

You’ll see that the 1,765 positive mentions are still the eye-catcher of the table, but it gets more interesting when you convert the table into percentages:

In this case, some interesting figures become apparent. Each segment has something that requires the attention of the company:

• In the business segment, it’s the overwhelming amount of neutral mentions. Here, the company can work to improve those neutral mentions to becoming positive mentions.
• In the families segment, things look positive. If the company can convert the 37% of neutral sentiment into positive sentiment, it will establish a very strong position in the segment.
• Things don’t look so positive though in the young people segment. A whopping 54% of the measured, unique mentions are negative. The company clearly has to right the ship in that segment or it might lose a lot of future business.
• The luxury segment is interesting as well. Almost half the mentions are neutral, but there is almost an even division between positive mentions and negative mentions. In this case, the company needs to gain a deeper understanding of why the public opinion is so divided and also of how it can sway some of the “neutrals” to the positive side.

Of course, the sentiment measured on the social web doesn’t have to be representative of a segment’s total population. However, the overall segment is more strongly influenced by the sentiment within the segment than by sentiment outside the segment. For example, a person looking for a business-line product will look at reviews from business users of the company and not necessarily at reviews from those in the families segment.

Are you performing a cross-analysis of customer sentiment and customer segments? Gauge what the sentiment per segment is. Look at where you can improve your position as a brand and take that next step, which is reacting appropriately to the data to help boost your brand’s perception and eventually sales.

As always, if you have comments, experiences or nuggets of wisdom to share, feel free to leave a comment!

Photo credit: Megan Fizell, Tres Jolie Studios

The content on the social web is a rich source of information for people and organizations. This content can also contain a certain sentiment related to you, your brand or your products/services. In this post, I’ll be discussing the distribution of content—based on influence, virality and discussion—and the monitoring and analysis of positive or negative sentiment.

Access

When someone posts content on the Web, people will have access to it. This is a matter of fact and relates to all content and isn’t necessarily restricted to social media content, e.g. blog posts, status updates or user-created videos. It becomes social once distribution of content takes place outside the original creator’s direct control.

Distribution

The distribution of content can have an impact on you, your brand or your products/services if there is a clear sentiment involved. Sentiment is generally classified as positive, negative or neutral. In this post, I’m omitting “neutral”, because it doesn’t carry an opinion and in some cases, can even be positive because it enhances awareness.

I’d like to reflect on sentiment through three catalysts of distribution:

• Influence
• Virality
• Discussion

Influence

Influence is attributed to the person publishing and/or distributing a particular piece of content. That person’s influence is based on his/her authority (in knowledge, trust or experience) within his/her network.

Through influence, this network of friends/peers/fans is more likely to distribute the content that the influencer has published or distributed.

However, consider the following scenario:

Bill (INFLUENCER): “Don’t buy from <COMPANY X>. Their helpdesk kept me waiting for 45 minutes and then disconnected me.”

John (MEMBER OF BILL’S NETWORK): “I’ve heard from Bill that <COMPANY X> offers bad service.”

Paul (MEMBER OF JOHN’S NETWORK): “Have you seen this post from Bill that says <COMPANY X> sucks? Thanks for the heads, John!”

While this is a very simple example of content distribution, the essence remains this: three people are distributing a message that describes <COMPANY X> negatively. However, these aren’t three negative messages, but they’re three instances of the same message. The point is that distribution of content based on influence does not necessarily equal distribution based on agreement.

Virality

Content can spread like wildfire and earn a viral status, because they’re extremely funny, sad, absurd or well-made. An example is the “United Breaks Guitars” video from earlier this year:

People will spread content like this, but—as with influence—it doesn’t necessarily mean they share the sentiment portrayed in the content. In the “United Breaks Guitars” example, people may choose to spread the content because they’re amused by it rather than because they share a similar, negative sentiment about United Airlines.

Discussion

Agreement or disagreement over content definitely comes into play when we talk about distribution through discussion. Here, members of the network respond to the content by expressing their own stance. Let’s illustrate this by revisiting the two earlier examples mentioned under influence and virality:

Example 1:

Bill (INFLUENCER): “Don’t buy from <COMPANY X>. Their helpdesk kept me waiting for 45 minutes and then disconnected me.”

John (MEMBER OF BILL’S NETWORK): “I agree with Bill. <COMPANY X> once promised to send me a replacement product once and never did! They truly suck!!!”

Paul (MEMBER OF JOHN’S NETWORK): “That’s odd, I’ve never been treated badly by them. In fact, I’ve been a loyal customer of <COMPANY X> for 10 years!”

Example 2:

Dave: “Check out this video about United Airlines breaking guitars!”

Sam (MEMBER OF DAVE’S NETWORK): “That’s outrageous, but overblown in my opinion. United is one of the better airlines out there.”

Joe (OTHER MEMBER OF DAVE’S NETWORK): “Yea, I don’t like United. They lost two of my suitcases already!”

It becomes clear that in both cases, we’re not necessarily restricted to the distribution of a single message, but that multiple messages are involved. As a result, the sentiment of each message is unique as well and not a duplicate of the source.

This is the model I designed for content distribution and sentiment:

The model starts with a piece of content. The content is then accessed by people on the Web. It doesn’t matter whether this is free, paid, membership or freemium content. If it’s published, there will be people who have access to the content. After consuming the content, the viewer can choose to share/distribute the content. The distribution of the content can have one of three forms discussed in this post: influence, virality or discussion.

In case the content is distributed through influence and virality, it’s very likely that we’re dealing with the distribution of a single message (or sentiment).

When content is distributed through discussion, we’re more likely to see multiple messages (or sentiments) as people chime in with their opinions on the content or the parties involved in the content. Also, when content initially distributed through influence and virality spawn discussions, the original distribution of a single message becomes a distribution of multiple messages. However, the inverse is also possible. When content is initially distributed through discussion but someone publishes something that’s so strong that it branches into distribution through either influence or virality.

Sentiment Analysis

When you’re out to gauge sentiment about you, your brand or your products/services, it matters to know whether you’re dealing with a single or with multiple messages.

While it’s obviously not desirable to have negative content about you distributed through influence or virality, it helps to understand whether it’s the distribution of a single message seeded by the most influential members of people’s social networks. If that is the case, you might not have to deal with thousands or millions of negative members of the public but perhaps only a handful.

On the other hand, if thousands or millions of members of the social web are actively active in a discussion about you, it’s important to understand whether the messages (or sentiments) these individuals publish are unique and not a replication of someone else’s content/message/sentiment.

Making this distinction is still very difficult for machines even though we’ve come a long way in artificial intelligence (A.I.). It’s one of my focal points for 2010 to figure out how A.I. can contribute to sentiment analysis and other social business measurement/analytics topics.

There are solutions on the market that claim to be able to automate monitoring and analytics of brand mentions on the social web. However, to my knowledge, these tools are unable to distinguish between the distribution of a single sentiment and the distribution of multiple sentiments. In the case of the first example, these tools would count three negative brand mentions, while in reality, it is only one.

As a result, it’s most appropriate if sentiment analysis is performed by humans rather than machines. Humans can spot the difference between content distributed through influence, virality and/or discussion. Using sentiment analysis in this way will help you create a clearer picture of what is being said about you on the social web. It will also help you identify key influencers in your community’s social network. Sentiment analysis is not about keeping a tally of how many positive, neutral or negative sentiments you receive on the social web. A sentiment from a key influencer is more impactful than the sentiment of a lesser know person. Note that a key influencer’s sentiment is not necessarily more important but because s/he has more reach, the word-of-mouth effect is larger and thus more impactful.

Once you have an idea of what the content distribution looks like, how sentiments about you are built up and who key influencers or seeders are, you need to take action. This can vary from rather passive activities, like closer monitoring, to very active tasks, such as participating in the discussion or issuing a public response.

Go and find out what’s being said about you, your brand or your products/services. Identify whether these sentiments are distributed through influence, virality or discussion.

In the case of influence, find out:

• What the sentiment represented in the content is
• Who the key influencers are
• How members of the influencers’ network are distributing the message
• Whether you are dealing with a single sentiment or with multiple sentiments
• Whether discussion is being created out of the original content
• Whether it’s appropriate and how you can be involved in the content’s distribution

In the case of virality, find out:

• What the sentiment represented in the content is
• What is causing the content to go viral
• Who the key seeders are

• Whether you are dealing with a single sentiment or with multiple sentiments
• Whether discussion is being created out of the original content
• Whether it’s appropriate and how you can be involved in the content’s distribution

In the case of discussion, find out:

• What the sentiment represented in the content is
• How many unique sentiments are represented in the discussions taking place
• Whether key influencers are involved in the discussion
• Whether it’s appropriate and how you can be involved in the content’s distribution

If you have comments, experience or other nuggets of wisdom, let them be heard in the comments!

Photo credit: David Sutherland

It’s always important to consider the value, the relevance and the timing of your message when marketing on the web. The danger in online marketing—especially with the advent of social media—is the backlash that can arise when a certain marketing effort is not to the audience’s liking.

Some companies will enter a confrontation with the audience, while other companies will ignore the audience. But either way, there’s a moment of contact that you can seize. The amount of backlash you receive is the result of the influence of the starter. Wouldn’t you like to be able to make use of that influence to your advantage? There are more benefits to be gained in repairing any bridges that may be broken than to burn them down completely.

Understand the value of your messages to your audience. When things do go wrong, embrace the naysayers.

However, the need for real-time discovery has opened up the marketplace for new players who can specialize in this field. The best will be able to provide search feedback, keyword monitoring and intelligent recommendations.

• Search feedback will give you the real-time results for a certain query, which is particularly useful with current events or trending topics.
• Keyword monitoring allows you to monitor real-time activity on the words that are relevant to you. This will also be helpful in your search engine optimization efforts. This could also be a premium option that service providers can monetize on.
• Intelligent recommendations help you filter out the clutter by bringing you only the best and most relevant search results. This increases usability by decreasing overload. Furthermore, it’s a monetization opportunity for service providers to offer premium advertising space.

The key will be to find a tool that works for your search needs but prevents you from becoming overloaded.

• Content: Rewrite the content to be more about them than about you. While messages should cover what you’re promoting, they should also emphasize the benefits to users. Scrutinize your content and check whether its relevance is high enough for your audience.
• Form: Experiment with different types of headlines to different audiences. The headline of a press release should catch the attention of a journalist, but at the same time, it might not be enticing enough for a consumer. Play with the layout and the use of images to make your message more attractive or reader-friendly. Make use of consumer (eye-tracking) studies to see what areas should be focused.

Don’t be afraid to throw spaghetti at the wall and see what sticks.

Have you set up listening tools, such as Google Alerts, or are your agencies/partners listening to the feedback that is given in your consumer community? Are you actively searching for mentions of your brand/product/service or even actively reaching out to the community to provide feedback to you?

Your community of consumers has a lot to say and has a lot of platforms to say it on. Stay on top of the content related to you and take what’s being said into consideration.

Listen. Validate. Act.

Allow what people are saying to matter. Embrace and appreciate the fact that people are taking their time to talk about you.

I came across an interesting case recently where a client more or less demanded a service agency to invoice less—despite delivering all the services as described. Not only is this morally unfair, it’s disrespectful.

You get what you pay for. You pay for what you get. Don’t destroy a potentially solid long-term relationship by making the wrong one-time decision. Go for the best that your organization can get, the price will justify itself.

As David Schwartz alluded to in his book “The Magic of Thinking Big” (Amazon link): Go first class, you can’t afford not to.

But what if you had an actual economy? What if community members had access to an exchange to real or virtual currency?

As we’ve seen with virtual world Second Life, the presence of an economy has destroyed the concept of fair value and the social aspect of altruism. The result has been a community where the power lies with the rich and the rich protect their status by giving little and taking much. As a result, they impose a high barrier to entering the community’s elite and discourage the notion of sharing to those who enter the community later.

Play into your community members’ egos but don’t play into their potential greed.

However, there are also instances where the product/service is killed after the pilot wasn’t successful.

Re-evaluate. Re-adjust. Re-think. Re-design. What went wrong to warrant a pilot failure? Was the concept thus flawed that an unsuccessful pilot could be predicted? Did the pilot produce new information about the product/service that was overlooked? Success can be unpredictable but shouldn’t be discarded after a setback. The pilot isn’t an end in itself. It’s merely another means to an end.

Let the end guide you.