Decoding Customer Feedback - Robots to the Rescue!
07 March 2019
Researchers fear open-ended data because the traditional approach to dealing with it is broken:
You ask an open question, gather some verbatim responses then pay someone to read through them one by one and manually categorise them against a codeframe. This approach is time-consuming, slow, expensive and sometimes error-prone.
As a result, over the years, researchers have shied away from using open-ended questions in surveys because of the hassle and effort involved with dealing with the data they yield.
However, this means the industry is ignoring a critical source of real insight – open-ended questions represent a unique way for customers to express their opinions in their own words, saying what they really think. By removing those comments, we are essentially muffling the customer’s voice.
Furthermore, increasingly, researchers also have to deal with a wide variety of other text sources (social media, customer reviews, emails, contact centre conversations etc.). Using traditional thinking, researchers become quickly become overwhelmed.
With all the advancements in AI in recent years, surely there must be a way for computers to help?
In 2019, surely it must be possible to simply throw all this text at a computer and have it spit out a set of useful insights and actionable findings?
Well, AI can help massively speed up the process, but our experience shows clearly that human support is still necessary to achieve high-quality results. Despite what you might hear, it’s simply unrealistic to think of AI as a single black box that can magically make sense of this mass of data for us, all by itself.
At Digital Taxonomy, we’ve spent the last few years looking at this challenge, and one of the most important things we’ve learned is that there isn’t a single approach that works for all situations. There are a number of complicated and interrelated factors at work (e.g. length of text, cleanliness of text, historical data). So, the ideal approach depends largely on those factors.
In order to avoid the limitations and inaccuracies of the single “magic black box” approach, we have instead developed a “layered” approach to AI, with different layers performing different functions. These layers can be switched on or off depending on requirements to tailor the system to each use case.
Using this approach, our clients are benefitting from massively increased coding productivity, with no loss in quality, and are at last able, cost-efficiently, to unlock the value of open-ended text.
Back to Blog