Automation in customer experience & market research - the future of surveys?
Gavagai CEO interviewed by CX Network for an article called “Do AI and text analytics bring the death of surveys?” (2018)
Q: What is next in terms of customer experience measurements? Is it all about verbatim and social media commentary or is there still an unchartered territory you are looking into?
I think solicited open-ended responses as well as unsolicited social media commentary will continue to be the most valuable sources of deep insight for the foreseeable future.
By leveraging the power of automation - enabled by AI-driven NLP - much larger data samples of unstructured data can now be analyzed in a fraction of the time this used to take.
We are now trying to take automation to the extreme, so that anyone - without any prior skills - can get operational and prescriptive insights from raw text data in just one single click, and we just launched a beta version for 46 languages.
We are also exploring several innovative ways of approaching this data. I can mention two initiatives:
We are working on how to setup a system for so called “continuous polling” which will collect data from open sources online and produce statistically representative polling results based purely on spontaneous unsolicited commentary.
This may have significant impact on how polling is done in the future.
Secondly we are developing a way to tackle vast online data streams for automatic and transparent topic extraction, allowing the linking of unstructured natural text to existing knowledge bases and to the underlying social graphs of web chatter.
This allows the detection of causal social network communities and events, the integration of various and different knowledge bases, and comparison of data-driven and knowledge-based analysis in the same interface.
Q: In layman's terms, we had Natural Language Processing and codeframes as the 'traditional' or 'early' stages of text analytics - how does this new generation (of which your business is a part of) differ from this and how much more insight does it bring back?
Traditional NLP relies on rule-based processing of text, which today is much too clunky and slow to manage the volume and extreme variability of user generated content.
Most of this has also been limited to English and translations to English.
In order to take on today’s data, and satisfy the required quality of analysis, we must move to a native and scalable approach which is language agnostic, insensitive to language variability, and fully unsupervised.
This is what we have been working on at Gavagai for the last 10 years, based on 20 years of research.
This approach has allowed us to perform analyses in virtually any language and launch domain specific analysis models for research in a matter of minutes instead of weeks or months.
Q: NPS proponents are known to say you only need to ask NPS & an open-ended why question. As a leader in the space, what are your thoughts on this. Would you agree, that one question can give you all the insight you might need?
In principle, yes. If you are able to gather enough volume of answers, you will be able to determine with high accuracy what is driving customer satisfaction - on a much more granular level - based solely on unstructured text responses.
In fact we are now able to predict the NPS score based entirely on what is written in the response, rendering the actual NPS score unnecessary.
And interestingly, responses also contain information about the likely future direction of the implied NPS, on an individual respondent level.
Skipping the numerical scoring part might have important implications.
In the first place, it gets rid of one step in the process, which is always good when trying to mitigate respondent fatigue, and the single open-ended question corresponds better to modern behavior, in as much as the response is like any other text or voice message, possibly delivered via the mobile phone.
In the second place, it eliminates a possible source of bias, namely the inclination people have to write motivations that are designed to fit their already given numerical score.
The most direct way to find out what people think is simply to ask them, without going through some intermediate step of introspective gauging,
It turns out that people are very inconsistent when numerically assessing their own opinions or beliefs.
Q: And lastly, do you feel like there's still a place for structured data in customer feedback or is it on the way out?
With a few exceptions, there will be no need for structured data, beside the metadata, in typical customer feedback.
But, there will still be a place for structured data and scale questions, in so much as it is still what people are familiar with.
Ultimately, it is the cost and business value of the customer insight that will determine the need for structured data.
As soon as the newer approaches reach maturity and become more widely used, the closed-ended approach will be less commonly used, in my view.
The Gavagai Explorer is the market's best and easiest tool for automated AI-driven insights from text data, with native support for 45+ languages. In the Gavagai Explorer you can upload and analyze ¼ million texts (reviews, complaints, feedback) in one go. No licenses. Pay as you go, and for what you need - when you need it. Follow the leaders in digital transformation and automation, with outstanding Return on Investment (ROI) for deeper and clearer insights.