top of page
Writer's pictureBearson

The science of engagement and the death of copywriting

Figurative language, the limbic system, ethical-AI, and bourbon.



“I’m really sorry Andy but I think I may have killed copywriting last night”.


The memorable words of my friend and AI-innovator extraordinaire, Paul Howarth. This was 2 years ago, way before the generative-AI hype tsunami. At the time, as well as waxing lyrical as copywriter, I was a director at his company Akumen.


Who is Akumen? Akumen is a world-leader in ethical-AI and Natural Language Processing (NLP). This means Akumen mines new seams of understanding, insight, and value from the ever-growing mass of global human dialogue. In a metaphorical nutshell, Akumen turns words into well-informed plans that work.

Anyway, on with the story.


3 weeks before this murderous admittance, the universe had chucked one of its innovation grenades over the fairly rickety fence that is the edge of Paul’s consciousness and it had gone off. It scattered fragments of genius all over the garden of his mind. Paul had spent a few hours pottering about, picking up the pieces before knitting them all together into a new idea for a kick-ass software product.


Paul explained the idea was to use Akumen’s software to identify figurative language in human generated text datasets. Examples of this are ecommerce reviews, message boards, and social media. This had not been done before. Sure, there are hundreds of ‘ai-powered’ Part of Speech tagging models that will try and identify the core components of a sentence – nouns, verbs, adjectives, prepositions, adverbs, etc. This is very useful in the world of NLP to identify language usage trends and meaning. But it is nothing new.


“Okay, so why is it a good idea for us to identify metaphors, similes, and analogies?” I asked.


“Because symbolic language is a key to your deeper consciousness Andy!” Paul replied.

I knew I was in for an education. I made myself comfortable.


Paul explained that figurative language is generated by several parts of your brain. The prefrontal cortex, temporal lobes, and parts of the left hemisphere help with language generation. They are supported in this noble undertaking by the limbic brain, specifically the amygdala. The amygdala generates the emotional aspects of our verbal responses.


“What’s really interesting is that there is a school of thought that suggests some types of emotionally charged statements, that are created by the amygdala, are not filtered by our consciousness. They escape from our brains before they can be altered or toned down. Getting access to unfiltered emotions through analysis of figurative language will provide a huge leap in accuracy and depth of understanding about our behavioural drivers. Because, as we all know Andy, emotions drive behaviour.”

“Sounds good. How do we prove it?” I asked.


Luckily, we had recently been introduced to Mr Bill Harte. Due to his extensive global experience, one might surmise that what Bill doesn’t know about marketing, sales, branding, and innovation doesn’t actually exist. He’s even written a book about it: The Agile Innovation Playbook.


With Bill’s help an idea was hatched.

Let’s see if we can improve purchase intent in a target audience by improving the words within a brand’s existing marketing copy.

Step 1: Choose a product. Something we all like and about which there are lots of public domain comments.

Meet Bulleit Bourbon, originally from Kentucky USA, and owned by Diageo. The brand was re-launched in the UK in 2006. You may already have met.


Bottle of Bulleit Bourbon on a balck and white explosion background


Step 2: Obtain lots of public feedback comments about Bulleit Bourbon, and bourbon in general, as an authentic sample dataset.

Using clever Web XB software we collected over 26,000 comments from multiple product review websites in the UK and US in about 4 minutes. We called this dataset “The Consumer’s Voice”.


Who is Web XB? Web XB is a disruptive market research agency that helps their clients collect, tag, and analyse target audience textual data with a suite of cutting-edge software tools and market-leading expertise. This includes using Akumen’s emotion analytics tagging and a ground-breaking conversational voice bot for data collection at scale and hi-fidelity.

Step 3: Use the ‘Consumer’s Voice’ dataset to build and refine the ‘part of speech’ and figurative language models to be used by the Akumen software.

It took the Akumen R&D team 3 days to do this. Pretty damn quick.


Step 4: Run these models over existing Bulleit Bourbon copy to see how well it scores in terms of common language used.

A description of the taste experience of Bulleit Bourbon from a front page of the brand’s website was analysed against the most common language used across the 26,000 reviews in the Consumer Voice dataset. Considering the likely size of the Bulleit Bourbon’s branding budget, some of the findings were rather shocking...


Website landing page for Bulleit Bourbon with highlighted paragraph of words
Bulleit Bourbon's website copy - I've highlighted the section of copy we analysed (©Diageo - https://www.diageo.com/en )


Words like ‘smooth’, ‘dry’, and ‘oak’ scored well, meaning they were used frequently by consumers. If words in your copy are common parlance within your target audience, the chances are any content published that includes these familiar words will be well received and have a positive impact. All good so far.


The copy scored less well with the words ‘gentle’ and ‘toffee’. Akumen’s software identified that ‘subtle’ was 3 times more likely to be used by consumers than ‘gentle’, and ‘caramel’ over 10 times more than ‘toffee’.


But wait, it gets worse. Imagine if some of the key descriptive terms in your brand’s copy were so far off the mark that your target audience did not use them once in their own descriptions of their experiences with your product. Not once in over 26,000 unique reviews.


Ladies and gentlemen. Fellow bourbon drinkers. May I introduce to you, some for the first time perhaps, the words ‘nutmeg’, ‘satiny’, ‘tones’, and ‘spiciness’. No hits. Nul points. A total waste of content space.


A paragraph of words describing the taste experience of Bulleit Bourbon, highlighted in different colours
Mixed results for the language used in the brand’s own copy.

The software also floated up a handful of other words and phrases that occurred frequently in the Consumer Voice dataset, but not in the brand copy, including the figurative expressions ‘up front’ and ‘a touch of’.


Step 5: Use these linguistic insights to create an improved piece of copy and A/B performance test against the original with an independent research panel.

This is what we came up with. Simply swapping out terminology from the original copy that scored poorly, replacing it with words and figurative phrases shown to be more relevant and familiar to the target audience. Both pieces of copy were then sent to be independently tested with a panel of 502 18+ year old consumers.


The results were sobering:


Two paragraphs of words with a bottle of Bulleit bourbon between them.
The modified copy scores almost 25% better with the test audience simply by changing some of the words in an AI-informed manner.

The score is a statistical comparison of expressed ‘purchase intent’ as a result of reading the copy. This scientifically informed improvement of copy from a major global brand increased consumer purchase intent by 24%.


Bill, being the most familiar with this space, had something to say about this:

“Let’s assume Bulleit is a 40,000 case brand and it makes £40 per case =£1.6m profit. Assume profits go up in line with sales (unlikely but possible), the upside is 24% of 40,000 = 9,600 x £40 is £400,000 extra profit. Less £150,000 for the analysis then that’s 166% RoI for the first year! Every year after that the RoI increases.”

A massive result enabled by some clever, but fundamentally simple and easy to apply, Artificial Intelligence thanks to Akumen.


This was the reason Paul was apologising to me, his friendly copywriter professional. A genuinely considerate dude, he was concerned his most recent innovation was going to reduce opportunities for the scribblers amongst us.

Well-designed AI continuing an inexorable march of progress over the increasingly irrelevant endeavours of us fallible humans! Eeek!!

I thought on it and met up with Paul the next day.


My response was “Thank you guv. Not for your apology, as it’s unwarranted. Thanks for creating a game-changing tool that I can use to do my job even better. It’s not replacing me. It’s enabling me to guarantee the impact of my copywriting products for my clients. It’s making my life easier. Nice one.”


So, there you have it. An alcoholic example of the absolute power of a well-formed AI tool being leveraged by some smart humans. Here at Bearson I provide wit and verbal vigour, otherwise known as copywriting and brand-hacking, to my very cool clients using all the tools at my disposal. Sometimes this includes AI solutions (some that are not even on the market yet), and sometimes it doesn't. But it always includes using my curious brain and my particular set of skills.


Whatever you need, Bearson can sort it for you.



 

Footnotes:


To avoid confusion perhaps inevitable due to the raging AI-hype levels at present circulating the zeitgeist, this article is not about generative AI – your ChatGPT’s and BARDs. This is about ethical and rule-based AI which is far more reliable than generative AI.


It should be noted that the testing Akumen performed did not confirm or deny the posited benefits of being able to track the use of figurative language to reveal subconscious emotions and behavioural drivers. Since this project Akumen has continued development of their emotional, cognitive, and behavioural ontologies (frameworks of understanding) delivering the ability to identify and analyse deep and diverse indicators of psychometric designations across segmented groups. Akumen is particularly concentrating on supporting live-saving initiatives in mental-wellbeing and healthcare.



If you want to know more about anything mentioned in this article, please drop me a line.











bottom of page