This month Co-Founder, Rachel Clarke, talks all things AI.

In The House

Innovations and initiatives from Strat House clients and partners.

• Unilever helping women farmers to grow pineapples and their incomes
• Adidas takes on Samsung’s controversial ‘Night Owls’ ad in women’s safety campaign
• Unilever Study Reveals Influence of Influencers in Sustainable Behavior Change
• Dove are standing up for #KidsOnlineSafety
• Gong cha named best food & drink franchise at the 2023 Global Franchise Awards
• Find out why LinkedIn And Dove Are In It Together To Help End Hair Discrimination

There’s an AI for that.

Apple told us there’s an App for everything. Now that App is likely to be AI driven.  Last year, the industry news stories were often driven by the “metaverse”. This year, the hype is going to be around AI and I can guarantee that the trickle of stories from last year will become a flood of PR firsts making the headlines for brands. The question will be whether or not they have any lasting impact.

But outside of the press, AI is also making an impact on marketing processes.  Tools and platforms are being developed that use the technology to automate more and more of marketing. It’s been a ‘thing’ for a while, for targeting, emails etc, and it is spreading to the more creative part of the process. Some of this will be good, some may not show the negative impact until much later, but as with web3 technologies, the genie is out of the bottle and we need to understand its shape.  

The older models tend to work on your data sets, the information you have about consumers, customers, their behaviours etc.  They take a large set of data and identify patterns and commonalities that are used to drive further actions.  An example would be email tools that analyse all open and read activity and ensure the comms are sent out at the right time to optimise open rates.   These algorithms tend to have a logic to them that can be easily explained.  Chatbots, used to support customer service by providing common answers, have a logic plan behind them – there’s a lot of planned conversation trees all written beforehand.  There are SEO tools that continuously analyse what is working and suggest keywords for your content; other tools will analyse all your content to ensure it follows the rules you have set.

The newer models are different – they are a lot more like magic.  From a prompt with a few words they can produce images, or poems, or essay answers. They are based on a wide assortment of training data – often scraped off the web – that is fed into the mill and recombined for the output.  Which is why, when Heinz asked OpenAI Dall-E2 generator to create images of ketchup, the images looked like the Heinz bottle. It’s the same reason they asked people to draw ketchup bottles the year before – because it is an iconic image. So people use the image in the web, describe it as ketchup, which gets scraped as training data and outputted as an answer to what ketchup looks like. So we’ve moved from understanding what the data and information I have is telling me – or how to respond to it – to creating something that is new and not seen before, or at least standing on the shoulders of giants by combining, in new ways, everything that is fed into it.

An addition to the more number-driven email AI tools could be a subject line generator, that takes a wide set of your brand comms as training data and generates lots of different subject lines; it can then refine the predictive model based on responses.  The feedback loop should provide continuous improvement on how the content is written.

For anything that requires training, the output is only as good as the input.  There are enough horror stories of AI that responds with varieties of hate speech – because it is being trained on open data sources from the web and that type of speech is common.  As a brand, you’ll need to provide your own set of data, information and input for your bot of choice to ensure any output is on brand.  You’ll also need to remember to cost in ongoing training, as the data set will need to be updated on an ongoing basis.

We’re moving from the tools helping the creators – the writers and the designers – to possibly replacing them, or at least reducing team sizes and effort required.  Our projects have explored using AI to write product copy for ecom, to write social posts from short tweets to blog posts.  We’ve looked at ecom asset generation: “why do photoshoots when your AI can use digital models and digital clothes to create assets for every single product on every size you offer …and then use the tools to display the right size product to your digital store browsers?”

We know that marketing automation will be used to cut costs and people from the process. But we hope that by freeing up time taken to do repetitive jobs, we can use that to be far more creative and responsive to consumer needs and wants.  This year, we can take a look behind the headlines and see what new ways the tools may be able to help you.