How to harness social data to drive real-time competitive advantage
In the era of post-truth and fake news, how can brands gain real-time competitive advantage to drive business outcomes?
The answer resides in social data and social listening combined with the capability of artificial intelligence (AI).
How this is done is the subject of this article.
Unstructured real-time social data is the only source of data fast enough to keep up with the ever-changing consumer behaviour and landscape. If properly harnessed it can drive business outcome. However, social data isn't without its challenges. Here are some of them:
Social data is vast and unstructured
Language is complex and nuanced
Meaning of words and some sentences can be ambiguous and often require context
Slang and sarcasm have to be decoded and contextualised
The traditional keyword-based approach to social listening may overlook important implicit information
Sentiment is often only a starting point to meaningful and measurable insight
Social data is a noisy environment with many voices including: the general population, content creators, influencers, broadcasters, brands, vloggers, bloggers, publishers, and writers. Added to this environment are different languages, dialects, cultures, and countries.
Coupled with the above is the spread of misinformation and fake news from dubious sources with political motives. Also, misinformed persons with sincere intentions are also spreading unverified information that unsuspecting recipients assume to be facts. The term post-truth aptly describes the condition we find ourselves in since the proliferation of the internet and social media.
Real-time brand tracking, of brand attributes in different languages, offers the answers to harnessing social data to drive business outcomes. What this means in practice is to deploy AI capabilities through the classification of language to enable predictive, actionable and measurable customer behaviour.
It requires a rethink of existing brand tracking tools. The current approach of conversation search queries (ie. keywords based approach) isn't sufficient. Language is complex. Words can have implicit meaning. We require greater sophistication and powerful tools to unearth not just what people are saying about brands but more importantly, what they mean by what they are not saying. This may sound counter-intuitive. However, this is the very essence of human language. People can infer meaning from what isn’t said. For machines to deliver meaningful insights that can influence consumer behaviour in real-time, it too needs to have at the very least, a language intelligence that can rival humans in order for it to understand human behaviour and motives.
For example, if a brand wants to find out whether its product is trusted by the general public, it wouldn’t make sense to use social listening search queries like 'trust + brand’ keywords. People don’t usually speak in simple and direct terms such as: "I trust 'x' or 'y' brand because it makes my skin glow". Instead, you may 'hear' trust expressed in the following way: "I use 'x' product because it makes my skin glow". This is a simple trust statement. However, the word trust isn't explicitly used. It is implied or inferred. As such, brands require an intelligent social listening tool and approach that can deduce inference from language at scale, by analysing tens of millions of social conversations, in real-time to accurately interpret meaning. This can lead to actionable insights and accurate predictions about customer behaviour.
New technologies will very soon be available that will enable brands to have predictive consumer insights at their fingertips.
How a ride-hailing company uses social data to drive business outcome
A popular ride-hailing company has classified language for its machine learning algorithm in much the same ways that humans classify language. They have moved away from just sentiment analysis*.
They have classified language incorporating the nuanced attributes* relevant to their brand. In the past, machines required a binary definition of words and meaning in order for it to successfully identify and execute a command, like a simple search query. To derive real-time actionable insights, machines have to accurately infer meaning from sentences that are implicit, ambiguous, humorous or sarcastic, etc. It has to learn to make meaningful distinctions between different types of expressions and take into account the contextual use of language. With the capabilities in Artificial Intelligence*, significant inroads have been made in this area, and it continues to develop at an accelerated pace.
Sentiment analysis - the process of contextually mining text to identify and categorise the subjective opinions expressed by the writer. Normally it is used to determine whether the writer's attitude towards a particular topic or product, etc. is positive, negative, or neutral.
Nuanced attributes - having or characterised by subtle and often appealingly complex qualities, aspects, or distinctions (as in character or tone).
Capabilities in Artificial Intelligence
It’s important to understand the four essential capabilities that constitute Artificial Intelligence as we progress into this discussion.
Natural Language Processing (NLP) - is the domain of artificial intelligence that deals with the interaction between computers and humans using the natural language. The ultimate objective of NLP is to read, decipher, understand, and make sense of the human language in a manner that is useful and valuable.
Knowledge representation - is the domain of artificial intelligence that focuses on designing computer representations that capture information about the world that can be used to solve complex problems. It incorporates findings from psychology about how humans solve problems and represent knowledge in order to design logical statements that make complex systems easier to design and build. It relies on logic to model reasoning.
Automated reasoning - is the domain of artificial intelligence that focuses on understanding reasoning capabilities in computer systems. The aim of automated reasoning is to design computer systems that can reason completely automatically without human involvement.
Machine learning - is the domain of artificial intelligence that seeks to design a system that improves and gets better over time by learning from previous interactions without being explicitly programmed.
The ride-hailing company built a 'trust model' that can be deployed to social data. 'Trust' has to be defined and agreed upon. It's critical that there is a common and universal agreement of what 'trust' means within an organisation that actively uses social listening to determine and measure its own brand equity amongst its consumers.
A taxonomy was established to classify and define all of their brand attributes. A trust framework* was developed for a common understanding that articulated their brand values, purpose, people-first approach, customer experience, safety, privacy, customer support, insurance, driver background checks, etc.
Trust framework - is a generic term often used to describe a legally enforceable set of specifications, rules, and agreements that govern a multi-party system established for a common purpose, designed for conducting specific types of transactions among a community of participants, and bound by a common set of requirements.
The customer's interpretation of 'trust' compared to the driver's, and to the employee's of the ride-hailing company all had to be understood from each vantage point. Using their trust framework and brand attributes, their computer programmers can review similar social conversation data and code for it consistently so that their machine learning models can learn from those algorithms coherently. An added complexity is around localisation. The way something is expressed varies from country to country, and many times from region to region too. Cultural nuances also come into play. Therefore, the ability to classify language organically to capture and codify national and regional variations in expressions is critical for effective machine learning.
Triaging and decision science were applied to recurring negative trends, crisis or one-off news stories in order to achieve better, faster and cheaper insights when compared with traditional approaches. This enabled immediate leadership decisions to be taken where necessary to mitigate potential negative brand attributes impact, or to reduce its effect if it had already occurred. This can all be done in real-time. A great deal of effort was invested in the root cause analysis of negative customer experience. Within the data records, there are multiple opinions of an event, both positive and negative in order to get to a level of actionability to improve the customer experience.
Opportunities can also be leveraged in real-time when brand listening, AI, brand attributes and trust framework are all coherently working together to drive business outcomes.
The use of metrics such as real-time brand trust score is a good indicator of a brand's equity at any given moment. This will become ever more essential in the social era where brand value can easily be damaged by a single negative social post that goes viral.
Categorising social data into various specialisms such as market research, brand tracking and customer experience, etc will open up lots of opportunities for brands to gain actionable, real-time and predictive insights to drive behaviour. This was unimaginable only a few years ago.
#AI #brandtracking #sociallistening #brandwatching #digitaltransformation #robotics #automation #artificialintelligence #naturallanguageprocessing #knowledgerepresentation #automatedreasoning #machinelearning