What are people saying about your brand online when you’re not listening? And how much customer feedback are you missing out on?
These are key problems that one North East-based company is hoping to solve.
“There’s no business that isn’t being talked about somewhere,” Steve Erdal, director of linguistics at wordnerds, explained to BusinessCloud.
“You know there’s information out there that’s useful, but pinpointing it and getting it in an understandable state is difficult.”
While the likes of Google and third-party apps are very good at finding specific words and phrases across the internet, they don’t offer interpretation.
In contrast wordnerds, which began life during a hackathon for Nissan, is determining what people actually mean by applying text analytics.
Erdal explained that around 80 per cent of the data that exists online is ‘unstructured text’ and deriving meaning from it would be near-impossible with human labour alone.
“It’s not a question of is that data out there, it’s a question of how do you do something useful with that data when it’s a firehose firing at a teacup?” he said.
Unlike the numbered data in sales figures and website analytics, which can be quickly graphed, a summation of people’s written opinions takes far more time – and smarter technology.
But even if a business does put humans up to the task of trawling through blogs, reviews and comments, this can create its own problems. “When that type of thing is attempted, bias still tends to sneak into it,” explained Erdal.
“If you have an opinion about what your products are like, or what people think of your product before you start, that will come out in the ones you choose to highlight.”
Erdal, who has previously worked with Newcastle University and as an agency copywriter, admits that current systems for ‘sentiment analysis’ often fall short. For example, scanning a business’ reviews for predefined ‘negative’ words often leads to false positives, particularly because the meaning of words is always evolving.
“We felt that there was a misunderstanding by a lot of these tools about what language is and how it works. People’s vocabulary changes considerably, but the underlying grammatical structure of how the language works are quite similar.”
On eCommerce websites the company’s ‘stargazer’ product can analyse large sets of existing reviews to reveal insights that might otherwise fall between the cracks.
“We bring those reviews together and we give the information on what people liked and didn’t like, and how they would improve the product, and where people are buying it and how things are going with delivery,” he said.
Instead of finding keywords, the AI analyses words that are being used in the same context as other words to create topics, and from that topic a meaning can be deciphered.
While the company hopes that the AI will one day be completely self-sustaining in this task, it is already finding insights without the help of the team.
“It finds things we would never have thought to look for,” he said. “A customer said a product that they bought was ‘toast’, meaning broken.
“I always thought that toast was warm bread, and would never have thought of looking for that word!”
The company’s technology can be put to the task of interpreting social media posts, backlogged stacks of documents or internal company feedback.
wordnerds is currently focused on brand perception but is looking to expand into analysing the way organisations speak and how they differ from their competitors.
“We know, as human beings, that the power and the beauty and the emotion and the sentiment of language come from how the words interact with one another. Language is endlessly interesting so there’s always going to be more stuff that we can do,” he said.
“Every conversation that we have with smart people, they come up with another three things we could do with this.”