Mind Your Language: Why AI-powered comparative linguistics is the perfect complement to marketing instinct

OPINION: Marketers have a wealth of planning tools available to them, but machine-assisted analysis of language has the potential to reveal the ‘unknown unknowns’, argues Ralph London MD Tom Winbow…

How can we really be sure that our interpretation of patterns and trends we see in viewing and listening habits, social media activity and everyday conversation are accurately informing crucial decisions relating to content creation and marketing strategy?

As agencies, media owners and brands we have a wealth of data sources at our fingertips, some of which we own, some of which are sourced via third parties. And most of us will already be using tools and services to extrapolate that data and inform strategy.

Without doubt the most important planning tool for an agency isn’t technology-driven at all – it’s personal intuition and human instinct, based on immersing yourself in a brand and its community in order to become what we at Ralph call a ‘super fan’.

But new techniques based on analysis of language using AI, emerging from the cutting-edge worlds of counter-terrorism and criminal intelligence gathering, are going to change the way we approach audience research.

In short, topics of conversation, words, grammar and sentiment can all now be measured from multiple sources to identify similarities or differences in the way people speak, enabling you to identify opportunities and the ‘unknown unknowns’ – new insights that perhaps would never have been expected.

In our view, the key elements of planning any kind of campaign, especially for social media, are tone of voice and authenticity, both of which allow you to communicate a brand in a way that resonates with its audience – at a fundamental level you need to find common ground with the people you want to engage with. Additional linguistic analysis is a great asset in that respect.

We stumbled across AI assisted comparative linguistics almost by accident. A friend of mine was working on a language thesis on use of the word ‘man’ in youth culture, and that got us thinking about how knowledge of linguistic trends within certain demographics across different markets could assist with our own projects.

For example, social and economic trends are the base levels when researching the 18-24-year-old demographic. As such, our default personal, intimate approach to campaign planning requires us to live and breathe social media, spending a lot of time engaging with the extremely passionate communities that form around a brand.

But AI, it turns out, is extremely good at identifying trends within the language of audiences, especially those you or a client may not know so well, whether because the product is entirely new, or if an existing product is being targeted at a new market.

You can drill down into that AI-gathered language data quite quickly and see what is important to people.

There are a few solutions out there and we’ve been experimenting with a platform developed by Relative Insight, which originally created its technology as a tool for identifying imposters in online communities as part of safeguarding and law enforcement efforts.

This objective translated extremely well into a marketing context: How do we ensure we don’t sound like the imposters in brand communications?

Relative Insight’s platform uses a ‘Relative Difference Score’ that allows us to quantify language in a way we couldn’t do before, in order to back up our initial assumptions and instincts about an audience. This is especially useful when poring over two datasets, i.e. when taking a US property to the UK – the technique is great for multi-market campaigns in the English language.

It also comes into its own if you’re dealing with an unknown audience in terms of sex, age and location. We’re doing some work now for a brand that has multiple editorial properties, and we’ve seen how people talk completely differently about the same subject in communities like Mumsnet vs Reddit.

The potential applications are vast. For example, retail brands can use AI-driven comparative linguistics to compare how customers complain differently over the phone, chat and through offline channels – then use that information to model language back to them.

It’s important to remember, however, that it’s still very early days. And this technology should never be seen as a shortcut to success or as a replacement for personal intuition and instinct when talking to brand audiences – but it will enable you to follow up on and confirm your hunches, especially when time is of the essence.