A great many words have been committed to the record lately about the multi-industry-disrupting potential of AI. Ironically, it’s likely that many of those words were ‘authored’ by an AI programme (anyone else growing suspicious of articles that use too many bullet points yet?).
For marketers – buffeted by myriad macroeconomic headwinds, and understandably unsure of where to place their next bets – a panacea in the shape of AI certainly sounds good. But then it also sounded good half a decade ago, when “AI” was last the darling of innovation. The “automation” gains promised by that last wave of AI innovation delivered at best mixed results; research published this year shows that 74% of marketers still think that their digital advertising campaign delivery “involves manual processes that are often time-consuming” (Smartly.io).
Before we get too deep into our thinking on the current uses for and drawbacks of AI in marketing though, it’s perhaps only fair that we first check in with ChatGPT. OpenAI’s globally-famous, headline-grabbing chatbot offers the following definition of “marketing AI”:
“AI (Artificial Intelligence) in marketing [is] the use of intelligent algorithms and machine learning techniques to analyse and automate marketing tasks and processes. AI in marketing can help companies to better understand their customers, personalize marketing messages, optimise advertising campaigns, and improve the overall efficiency of their marketing efforts.”
So far, so fair enough.
Yet, and as excitable as the current hubbub around generative AI (sometimes justifiably) is, the emergence of technologies such as ChatGPT on the scene isn’t even half the story. AI’s role in the everyday operations of a marketing team has been developing for the best part of a decade now. In this article, we’ll look at the how, why, and what of artificial intelligence in the context of marketing.
You can’t spell “marketing” without “A” and “I”
While the zeitgeist may have been captured by those recent advances in conversational AI, the emergence of useful AI technologies for marketers is not in itself new news. A cursory scan across the enterprise marketing architectures of today reveals several use cases of how AI is already being used in marketing.
Predictive Analytics: AI algorithms can analyse large amounts of existing customer data to predict future behaviour, allowing companies to tailor their marketing campaigns accordingly.
As this element of the technology progresses (and intertwines with concepts such as hyper-personalisation) you’ll hear increased mention of the concept of a “digital twin”: i.e. a virtual customer profile that facilitates a brand being able to better anticipate the outcomes of a 1:1 interaction with a set of customers – both individual and at scale – before the brand has initiated that contact.
Chatbots: AI-powered chatbots can handle customer queries and provide personalised recommendations, improving customer satisfaction and retention (although not to the scale and ‘generative’ extent of e.g. ChatGPT).
Content creation: algorithms can generate content such as blogs, social media posts, product descriptions etc., reducing the need for human writers (but not necessarily human editors).
Customer segmentation: AI can segment customers based on their behaviour and preferences, allowing companies to create targeted (and retargeted) marketing campaigns that resonate with very specific groups.
Ad optimisation: once programmed by a flesh-and-blood marketer (a process that itself is an unavoidably time-consuming task) AI can be trusted to optimise advertising campaigns. Whether analysing data in real-time to make adjustments to ad targeting, switching up (or out) elements of the ad creative, tweaking bidding strategies to better realise the initial goal of the campaign group, or even dynamically adjusting product pricing based on in-market circumstances … AI removes the burden of constant attention that would otherwise rest with a performance marketing expert.
Aura AI – A Picture’s Worth >Thousand Tags
In the recent past, it has been both the premise of AI-powered software and the promise of its developers that AI technology can help marketers to become more efficient, effective, and data-driven in their approach. And it’s along those very same lines that StoryStream’s own AI – aka ‘Aura’ – operates.
Named for its ability to quickly capture the distinctive qualities of an image – i.e. to encapsulate and encourage their aura – Aura is StoryStream’s AI-powered content curation assistant. Using over 70,000 tags, Aura categorises large volumes of imagery content at incredible scale – picking out the defining compositional elements of any image and then using pre-defined rulesets to sift the best from the rest.
The result is immediate insight into the compositional and qualitative elements of any image; and a heavily enriched set of criteria through which to search for imagery. An example: a fashion brand is looking for a dozen or so images of its latest collection being worn out and about by customers. The brand is typically tagged in 200-300 pieces of content each week, with a wide range of its products represented. To manually filter through six months’ worth of content in finding the dozen or so images that depict a specific item being worn in an outdoor setting would take a lengthy amount of time – which is where Aura’s usefulness presents itself.
By programming a couple of rules into the search, Aura will surface only those customer-created images that feature the product in that particular context. And by adding a rule around the quality of the image, that brand could also choose to only be shown high-quality items too.
So not only does Aura help brands to better organise the User-generated Content flowing into the StoryStream platform – it creates a huge time and resource-saving for the marketing teams using it! (Read how the Mazda marketing team were able to save a full 755 days of manual work categorising, sorting, and approving content – at a total value of £187,500 in equivalent resource hours using Aura here.)
With AI’s role in the marketing stack thus far being largely optimisation-focused, the shift to “generative” (i.e. creating something ‘new’) is understandably exciting for marketing professionals. Yet while it feels true that Generative AI has the potential to revolutionise many of the creative industries, it also raises ethical concerns about the potential misuse of AI-generated content and the implications for intellectual property rights. Let’s take a deeper look…
A Quick Guide: Generative AI
Generative AI is a type of artificial intelligence that uses algorithms and deep learning techniques to generate new data that resembles data it was trained on. In other words, it can create something new, like images, music, text, or even entire websites, based on patterns and features it has learned from a set of training data.
The details: Generative AI programmes typically use a neural network architecture called Generative Adversarial Networks (GANs), which consist of two networks: a generator and a discriminator. The generator creates new data based on the patterns it has learned from the training data, while the discriminator evaluates whether the generated data is real or fake. Through an iterative process of training and feedback, the generator improves its ability to create realistic and diverse new data.
Complex in theory … but the outputs are often startlingly (read: uncannily) familiar.
From creating art and music to generating new product designs (and even virtual worlds!). Here are a couple of recent examples of generative AI in action:
Text generation: creating new written content – e.g. articles, stories, poetry – based on a set of training data. The example of a definition (for “marketing AI”) created by ChatGPT at the beginning of this article is essentially a level-up from a classic search engine result, but outputs from generative text engines can be much more creatively intensive (try asking for a story in the style of Aesop’s Fables, for example, and you’ll likely receive a perfectly serviceable bit of bedtime reading).
Some brands are already realising value from text-Generative AI, principally by using it to scale their content marketing efforts quickly and effectively. And there have been some early movers in the martech space (think: Jasper AI, Frase.io) too – utilising the combined powers of search engine optimisation datasets and Generative textual AI to help create content that wins that all-important SERP space. For anyone who has toiled to include a fractious list of keywords, naturally, within a piece of long-form content – Generative AI represents a real boon (though to what extent readers agree is still open to debate).
Image synthesis: creating new images based on patterns learned from a set of training data. You will undoubtedly already be familiar with another of OpenAI’s products – Dall-E – and its simple interface. Type in a prompt (the more detailed the better), wait a moment and enjoy the results. The avocado armchair was a popular initial showcase of the technology (as mentioned in articles dating as far back as this one from MIT Technology Review in 2021). Here’s an image set we produced earlier using the prompt “piece of user-generated content advertising yellow and pink men’s tracksuit”. Uncanny, perhaps, but also quite snazzy.
Style transfer: a build on image synthesis, this process involves generating images or videos that combine the content of one image with the style of another. Again, Dall-E will immediately spring to mind. Here we’ve asked for that same yellow and pink tracksuit, only now in the style of 20th-century surrealist painter, Salvador Dali (the clue was in the name, right?).
Music generation: Generative AI is also increasingly adept at creating new music compositions, again based on patterns learned from a set of training data. Platforms like Boomy allow the user to set a range of parameters on a song (genre, tempo, instrumentation etc.), and even allow the user to import and remix their own audio files. An initial, very hastily pulled-together attempt to create a theme tune for StoryStream shows some promise.
Video generation: creating new video content, including animations, based on patterns learned from a set of training data. In-market examples include Synthesia – which uses plain text to create videos featuring talking avatars, principally for marketing and training videos – and slightly more moonshot-in-nature projects such as Meta AI’s makeavideo.studio.
As exciting – or diverting – as each of those applications may be, there are both practical and ethical concerns surrounding Generative AI too. Perhaps most immediately: the ramifications for intellectual property law. Since Generative AI relies on training data that is likely the result, in the first instance, of human endeavour – Generative AI can produce content that may infringe on an original creator’s intellectual property rights. (It’s worth noting that behind that conversation sits a broader, more philosophical question too: whether machines can create content that can be considered art or whether it is simply a replication of existing art. This writer eagerly awaits the slew of post-doctoral theses concerning the utility of Walter Benjamin’s arguments in “The Work of Art in the Age of Mechanical Reproduction” for the era of Generative AI.)
In a similar vein, concerns that inherent biases in the training data used to train the AI may mean that output generated by the model perpetuates or furthered those biases are increasingly prominent. “New AI, Same Old Biases” ran a recent headline in Bloomberg.
Further concerns exist around intention; not of the Generative AI in itself (yet!), but instead, the intent of the human actor scripting a prompt. While rules and “guardrails” designed to safeguard a user from being returned racist, sexist, disturbing or libellous outputs are programmed into Generative AI platforms, the technology still has the potential to generate false or misleading content that could further the spread of misinformation. In addition, there’s the potential for malicious actors to use Generative AI to create fake identities, forge signatures, and create other fraudulent content too.
In mitigation of some of these threats, Gartner has predicted that “[by] 2027, 80% of enterprise marketers will establish a dedicated content authenticity function to combat misinformation and fake material.” Whether the productivity gains unlocked through Generative AI ultimately exceed the productivity strains placed on marketing teams will have to remain to be seen…
Keep Calm And Carry On?
With the upcoming launch of ChatGPT and Whisper APIs, services can now begin embedding the power of OpenAI’s Generative AI into their own applications. So even if you’ve yet to spin up a portrait on DALL-E or converse with ChatGPT, chances are that you’ll be interacting with the technology soon enough.
Away from Generative AI, though, we’re excited about the potential for AI and another of marketing’s most prominent trends: video commerce. The need for more (and more authentic) video is a pressing concern for brands; yet producing video is famously resource intensive, while sourcing it can be a real timesink.
It’s at this juncture that we think a renewed focus needs to now be placed. At StoryStream we’re asking how it is that AI can be utilised to help brands advance their video commerce strategy; and how it is we can extend the time-and-resource benefits of our own AI, Aura, beyond imagery and into video?
Excitingly, you won’t need to wait too long to find out.