It’s helpful to think of social data as something of an alternative energy source. Social data existed as word of mouth before it was digitized and archived. Just like an energy source, the value we receive from social data is dependent on the ways we process it, what we’re powering with it and the maturity of the technologies and expertise that surround it. Let’s look at three of the most common challenges preventing most companies from extracting full value from social data.
For anyone involved in social at the brand or corporate level, noise is a real problem—even if they don’t realize it. Noise is anything that makes gathered social data less accurate or relevant. If they’re lucky, the phrases they’re trying to monitor are unique, but that’s often not the case. Accuracy and relevance depend on countless factors, including:
- Anticipation of common misspellings and incorrect spacing—bizarrevoice, bazzarvoice, bazaar voice, etc.
- User intent—are they talking about you, or another meaning/application/usage of the term?
- Reliability of spam, bot and duplication filtering—to what extent is the data you’re analyzing originating from non-humans (like bots), automated feeds, duplication and spammers?
- Product specificity—are you differentiating between brand-level metrics (like mentions of company name), and product-level metrics? How do you know what product someone is referring to if they don’t use the model number?
Trend lines and word clouds look useful, but if the data behind them is filled with noise, they’re nothing more than pretty pictures.
Silos are great for storing grain, but not social data or the insights it can provide. Silos are a common side-effect of corporate growth, or built in response to legitimate concerns about data leakage. But social data can have an enormous positive impact when it’s actively and securely shared. Some of the most common silos exist between:
- Internal stakeholders—is the right information getting to the right internal audience to support decisions and actions?
- Tools and datasets—are the tools we’re using and the data flowing through them leading to a seamless, holistic view of our customers, market, share of voice, etc.?
- Identity—how are we resolving identity information between onsite activity, social profiles, phone records, inbound emails, etc., into a better understanding of real people and our multifaceted relationships with them?
Destroy the silos that prevent your company from knowing and doing more.
Lack of sentiment standards
Social sentiment analysis—the often automated, algorithmic classification of content by positive, neutral, negative, etc.—is an incredibly promising field. But the results of many automated sentiment analysis tools currently on the market are often extremely inaccurate or unhelpful. The potential convenience offered by automation is frequently offset by hours spent manually calibrating and spot-checking. The prevailing reason that automated tools have such difficulty with accuracy is that most social signals can’t be easily standardized. Classification into positive, negative and neutral buckets relies on text or behavior clues, and can be thrown for a loop by things like sarcasm, humor, nuance, comparative statements—in other words, the way we actually talk. There are social signals, however, that can be classified by sentiment with complete confidence. Things like binary thumbs up /down, Net Promoter Score and scaled ratings are all unambiguous indicators of sentiment. Even though text-reliant automated social sentiment analysis isn’t yet fully mature, it can still be hugely helpful when anchored to some of the standard sentiment indicators just discussed. For example, when you sort content by positive, negative, or neutral ratings, you can then look for text patterns and reoccurring themes unique to each sentiment group.