The last month I saw a very significant breakthrough in the application of AI to human language processing. OpenAI, a nonprofit research firm backed by Elon Musk, developed a revolutionary AI model that can read, understand and write, almost at the competency level of humans.
OpenAI considered this breakthrough so dangerous that they stopped short of sharing the code, and instead, released a limited version. In this article, we’ll take a sneak peek into this landmark, and explore how such advances in AI-driven natural language processing can be leveraged by enterprises.
What Makes this Breakthrough so Remarkable (and potentially dangerous)?
OpenAI trained a large-scale language model by showing it 8 million pages from the internet, and the model then used this learning to generate the next word in a given sentence. However, to the researchers’ surprise, it got so skilled that it could extend this and write pages of prose, when primed with just one leading sentence. Here’s one of the many examples shared by the researchers:
While the model used isn’t novel, the landmark here is the performance level it reaches when trained with exceptionally large sets of data. To contrast, imagine a human who has to read all of Wikipedia’s English articles and use just that knowledge to write creatively on any assigned topic. Add to it the need to answer questions in reading comprehension style, to translate languages, or to summarize essays. This is what the AI model has been able to do.
While all this is cool, it became distinctly more concerning when the model displayed a knack to pick up context, adapt to a human’s writing style, and even get creative by introducing fictional characters, or imagined facts. Just think of the fake news that could be churned out at an incredible scale — and with easy access for a malicious actor looking to spread disinformation.
Perhaps as intended by OpenAI, this news has made the conversation around AI ethics even more intense. With recreation of this capability just a few months of trial-and-error away, the industry is grappling with the kind of safeguards that will be needed to keep it from getting into more sinister hands.
How Can an AI with Exceptional Literary Skills Help Enterprises?
In the backdrop of this development, it is pertinent to explore ways that such capabilities can be put to good use in enterprises. In spite of steady advances in AI’s abilities to process text, pictures, and videos, most organizations still have a huge dependence on numbers and structured data.
While financial dealings, business transactions, and operational updates can be quantified and computed upon, the same cannot be said of human interactions. With natural language being the free-flowing mode of communication amongst people, the spoken and written words contain a treasure trove of information. And today, this remains largely under-leveraged.
Whether it is periodic customer surveys, chatter on social media, feedback on review websites, interactions through contact centers, or ongoing communications with customer service professionals, all these touch-points are peppered with vital clues that can help answer the million-dollar question, “What do customers really want?”
However, many enterprises use archaic approaches to customer survey and digital listening programs. Textual feedback from these programs is often subjected to superficial text analytics that don’t go beyond simple text summaries, frequency counts of words, or naive sentiment analysis. These squander valuable customer signals, falling short on intelligence and actionability. This is akin to sending carrier pigeons in the age of internet and emails.
Here are 5 ways that AI, using advanced analytics, can help transform customer experience by reinventing customer feedback and listening strategies:
1. Listen better with adaptive surveys
Customer feedback programs such as NPS, VOC, or CSAT often use standard survey questions. With hyper-personalization and targeted marketing now being the norm, the use of a canned list of questions for thousands of customers is dated and inefficient. These surveys include open-ended questions prompting ‘Tell us more’, when in reality, it’s rare for a customer to give detailed feedback off their own bat.
Adaptive surveys, on the other hand can use targeted questions and smart follow-ons to probe customers for deeper and contextual responses. When a customer writes about bad order experience, the language model can query them for specific details of the process that fell short. If their response is too generic, it can gently nudge users to share something more actionable. By diving deeper with fewer questions, adaptive surveys can drive up overall response rates.
2. Listen continuously and across channels
Customer survey programs are usually triggered at periodic intervals, or based on predefined events. However, in this day and age, with most people being perpetually online, banking solely on customer feedback that trickles in on a quarterly or annual frequency is shortsighted.
It is crucial to listen continuously and leverage multiple channels to keep the conversation going. A quarterly CSAT survey could be augmented by monthly customer interviews, with online chats triggered by micro-events such as order confirmation, and by social listening that’s always-on. What’s critical in this multi-channel approach is to use analytics to contextualize the feedback, and to integrate insights for a comprehensive story that is actionable.
3. Understand the deeper intent in language
Most customer survey analytics in enterprises are limited to a shallow treatment of the textual feedback — word clouds showing the frequency counts, dictionary-based summaries, or naive sentiment models. They slice and dice the text, but end up losing precious intelligence about the deeper customer intent.
State-of-the-art AI models can take in tens of thousands of comments to identify themes that are on people’s minds, and further call out ones that directly contribute to them being a promoter. By detecting emotions, they can summarize all the comments into a digestible one-paragraph summary. Wouldn’t it be insightful to find that, say, product installation is the top influencer of satisfaction ratings, and that the recent increase in turnaround times are ruffling new customers?
4. Piece together an integrated view of the customer and product
While such a multi-channel strategy with deeper analytics can start surfacing new signals, it can also quickly get overwhelming. To promote actionability of the many insights uncovered, these signals need to be blended together to present an integrated view. This poses practical challenges since the various data streams have neither a common audience base nor clear integration points.
This can be solved by aggregating the insights at defined levels of ownership or logical hierarchies, such as products, brands, or geographies. Then the messages must be contextualized based on the channel, since a random complaint on Twitter needs different treatment than a ticket logged with the call center. Finally, the insights must be presented as a coherent narrative with visual storytelling, abstracting out analytical complexity for average users.
5. Enable organizational design to pave way for an integrated CX strategy
While we are at a stage of technology and data feasibility for tapping into customer signals from across disparate channels, the organizational structures might not be ready for it just yet. In mid-to-large sized enterprises, initiatives such as Customer NPS, CSAT, Social listening, Contact Center Intelligence and Customer Advocacy may well belong to different business units. Attempts to transcend these traditional turfs are often met with resistance.
An integrated strategy to manage customer experience calls for deep collaboration amongst the many teams operating with different ownership and incentive structures. Hence, such an initiative must be enabled by the right organizational design. A good start is to bring in executive-level sponsorship, define strategic goals and lay out a framework for partnerships amongst teams.
Summary
In summary, the current enterprise practices of customer feedback and social listening offer a lot of scope for improvement. There are opportunities in the administration of surveys, in the integration of signals from multiple channels and in the maturity of analytics that can be performed on the gathered data.
While there is ready availability of the analytics firepower, executives need to enable this strategically by planning the right organizational design.
About the Author
Ganes Kesari is a Co-founder and Head of Analytics at Gramener. Ganes advises leading enterprises on data science for business outcome, and helps NGOs adopt AI models in their conservation efforts. He is passionate about the confluence of machine learning, information design, and business value and is on an endeavor to simplify data science to help everyone understand its true potential. He is an Industry recognized speaker in Data science and is recognized as a top writer in ‘Artificial Intelligence’ on Medium.
Sign up for the free insideBIGDATA newsletter.
Speak Your Mind