E-Handbook: Social media analytics applications live and die by the data Article 4 of 4

peshkova - Fotolia


Better sentiment analysis can bolster customer data analytics

Customer data analytics are easy to gather in the social media era -- but they can be misleading if based on sentiment analysis culled from automated social media monitoring.

It's only with years of hindsight that the shift in customer sentiment analysis can be fully appreciated. There was a time decades ago when the only ways to gather insight on what customers wanted and how they felt about a brand were by knocking on their front door and asking, cold-calling them on their wall phone or sending them a survey with a self-addressed stamped envelope.

In hindsight, all that labor intensity brings a smile. The troubling truth is, those methods got great results: earnest, accurate information about a brand's success in the marketplace, whether it rated up or down, how much and the reasons why. It's easy to assume that today's ready access to customer sentiment would mean that the resulting customer data analytics are more accurate. But that's not always the case.

Social media to the rescue

The sea change in customer sentiment has been that the enterprise no longer has to go out in search of it; customers blurt it out all day long on social media. Uncounted gigabytes of information pour into Facebook, Twitter and their various counterparts at a staggering rate. It's a simple matter, then, to simply scoop up all that data and ingest it, to achieve a better view of one's brand in the marketplace.

Entire software platforms now exist to do just that: It's called social media monitoring (SMM), and the vendors offering new ways to gather and format customer data analytics are growing in number. There's Zoho, BuzzSumo, Sprout Social, Conversocial, HubSpot, Brandwatch, Digimind, Hootsuite, Buzzlogix, SentiOne, Talkwalker -- and that's just scratching the surface. It's not even necessary to buy a separate product to achieve this, in some cases: SMM is built into many CRM platforms.

The trouble is, it's often not very useful.

Chats aren't stats

The first (and biggest) problem with sentiment analysis culled from SMM is the casual assumption that chat stats are the same as survey results. They aren't.

What makes survey data so valuable is that the brand-specific data can be correlated to demographic data. It is difficult if not impossible to convert customer sentiment into actionable insight for the brand if demographic data is absent. Scraping brand sentiment data from social media is easily accomplished, but more often than not, only the most casual assumptions about the demographics of the chatting customers can be inferred.

It is difficult if not impossible to convert customer sentiment into actionable insight for the brand if demographic data is absent.

The impressive functionality of SMM itself is partly responsible for some of the misunderstanding of its utility. SMM tools crawl selected sites continuously, indexing them as they go -- and that indexing can be configured by the user and made available for very focused queries, pulling out brand mentions, selected keywords, phrases and emotional trigger words. All of this data then feeds into sentiment analysis. These are such cool tricks that the resulting customer data analytics seems more structured than it really is, for analytical purposes.

One thing that can be usefully culled from this process is head-to-head comparisons among brands: Brand X can gather chatter on its own brand, while also scooping up numbers on Brands Y and Z, to get an idea of which brand is top dog in the market. That's fine, but it's an awful lot of trouble to go to for information that is more easily gathered elsewhere: Competitors usually know, as a matter of course, where they stand in market share, to a tenth of a decimal point.

No, what's needed from social media is specifics about customer likes and dislikes. The stats aren't that useful.

The 'I know, right?' factor

Another problem with scraping customer sentiment in social media is a lack of control over groupthink. Social media forums tend to draw like-minded people into mutual validation.

If a particular make and model of automobile, for instance, is a topic of conversation in social media, and someone has a gripe about it, others are often inclined to pile on, chiming in about what they don't like, just to be included. It may well be that they like their car just fine, and wouldn't have thought twice, had they not happened across the conversation. This often shows up in text analysis, when new versus repeated information is charted over time: The entropy of new sentiment falls off rapidly, indicating that there is more 'Me, too' in the exchange than authentic contributions (and it will be no surprise that it happens most often in political forums).

Both negative and positive sentiment can be skewed in this way: Social media conversations about a brand have no controls in place to mitigate this phenomenon.

Understanding the conversation

All of this is not to say that good customer data analytics can't be culled from social media; on the contrary, social media offers a treasure trove of data that was nearly impossible to obtain in the past. It should not only be a priority, it should be considered indispensable.

Better analytical results can be derived from SMM data when the social data collection is more proactive. Automated social data collection is very useful, but stronger analytics are produced when the data can be enriched.

Brand ambassadors, for instance -- loyal customers easily incentivized to participate -- can actively initiate brand discussion in social media. This boosts the usefulness of the customer data analytics and the insights gleaned from it through the creation of more structured responses. Brand loyalists can establish parameters in discussion of products and services, increasing the specificity and personalization of responses to questions such as "First-time user or repeat buyer?" while eliminating noise.

Where this proactivity exists, SMM results are better still. And it sure beats going door to door.

Dig Deeper on Data science and analytics

Data Management
Content Management