Generative Pretrained Transformer 3, aka GPT-3, is an artificial intelligence (AI) that can be trained to write original content. Developed by Open AI, the company co-owned by Ellon Musk. Yep, the dude who we have to thank, for those silent self-driving cars. Yes, ironically the person who was ‘publicly’ scared about what AI could do, went and co-founded an AI company. Go figure. In simple terms how GPT-3 works is that you feed it content on the rules of language and subject matter and it churns out content for you. This is not the only use case. You can even ask GPT-3 to code for you by describing the layout you want. There’s a real example of what it can do here. While exorbitantly expensive at the moment,
This article isn’t about what GPT-3 is. If you want a simple primer, head over to this amazing Forbes article. I think both the world’s leading brands and agencies are going to not just sit up and take notice but be in a race to develop a working evolution of this product that they can call their own. Here’s why.
Reason 1: Codifying the secret sauce in the ‘agency’
How do you distinguish one advertising agency from another? The obvious answer from advertising aficionados is their ‘creativity’. It is what defines their final product at the end of the day. Some advertising experts will whisper that they can spot an ‘Ogilvy ad’ a mile off because of its Art Direction and creativity. Yet, said creativity, is treated like a black box for clients of these agencies. Sure, agencies are never shy at flexing their creative muscle when it comes to sharing profiles and thus pedigree of their creative teams. Or talking about enumerable awards that they have raked up over the years as justification of their creative prowess. As someone who has worked in advertising all his life, there is a danger in defining creativity in such a manner.
Experience has taught me that creative output is all about the team that is assembled at a certain point in time and the resources that it has access too. You mix and match your team and you are going to get entirely different results for the same client brief. This has always been how it has been. So, I never gave it much thought of it being any different across agencies. Consistency is what brand guidelines and brand managers were for. This outlook changed a lifetime ago, on the sun kissed sands of Goa Fest.
The person that brought about this epiphany was a relative stranger who mentioned something about how a particular agency was renowned for their methodology and approach that they brought to solving creative briefs. This wasn’t something I’d ever read about directly in any of the numerous advertising books that I’ve picked up. It wasn’t how we worked at R K SWAMY BBDO. That a defined thinking process not only drives a particular flavour of creative output, but is proprietary, was something new. And frankly unbelievable. If it were true, then new talent would have to be indoctrinated in a style of thinking to execute their work flawlessly.
This is interesting, because agency creative talent often jump ship at the first opportunity, because they are often the least well-paid of the lot. Which means creative teams are constantly in flux. The techniques used that go into the creative output, migrate with talent. So, if that’s the case, then how does one retain the ingredients of the secret sauce in the correct ratio of new vs existing talent? Can internal training ensure a certain kind of creative output? Is there some universal agency formula for creative standardisation? I somehow doubt it. But I do believe an evolution of GPT-3 could be the answer to this very problem.
Imagine for a minute that you can feed the GPT-3 algorithm with your agency’s unique style of ideation. Show it every piece of work that an agency has produced since its inception. Literally codifying the secret sauce. Imagine, its like the ghost of David Ogilvy empowering AI to write his unique style of copy today. You then build a weightage system that incorporates your agency’s unique writing style and ideation and combine it with each individual brand’s heritage language syntax to churn out that illusive first draft of copywriting for your communication piece. You then task another AI function that sorts through previous brand visuals and taps into stock image sites and measure this against trending Instagram and social media posts to throw up an inspirational mood board of images that might go with the copy. AI enabled advertising!
Want to do something really different? I am sure you could build an interface that allows for this. Using this it would be easy for the algorithm to extrapolate this too, within the confines of the creative brief. The result is, even if the AI can’t craft the final ad, you have a starting point for a creative piece that will be on brand, and message while having your essential unique agency style. This cuts down the time for ideation and research and allows creative folk to fine tune their thinking and end products.
Reason 2: AI can be a style guide on steroids
If that is too farfetched, then here’s another take. Modify the GPT-3 algorithm not to create text but to edit it. Every established brand has a brand style guide. These tomes talk about everything from physical constraints such as the logo placement, colours and fonts to use, to the kind of visuals to use and the tone of voice needed. The problem is that these are brand artifacts. Books that lie dusty on shelves and are not referred to as much as they should be. AI tools are excellent to give that final sense check of whether a creative is on brand. Think of it like the Flesch-Kincaid score Microsoft Word produces for your writing. But instead of measuring readability, the AI analyses brand tonality and match.
I’m sure some of you are saying that brands switch agencies for an entirely different style of creative. So how does this work if its in-housed by brands? You’re right and each agency would have its own secret sauce coded into the algorithm. But if the brand were to own the algorithm, then it could probably tweak it to produce diametrically opposite results in terms of creativity. Helping to explore new avenue and levels of work constantly for the clients.
But it’s not just agencies on brands that can tap into this opportunity. Imagine a third-party service available on-demand. Brands upload their brand key (the brand’s algorithm that defines its style guide, tonality, values and such instructions) and you insert this into a system along with an agency key (the agency’s take on solving creative problems defined by set parameters.) The third-party algorithm then mixes and matches in the cloud and generates a creative output. Now for this to seamlessly work, the agency works with this third party to curate its secret key. Similarly, so do the advertising agencies. You then have an established third party platform through which there is a matchmaking of agency and brand AI. Who could build this? My bet is on Amazon or Google for its sheer access to computing power.
Reason 3: AI can bring more to every brief
GPT-3 could also be used to write briefs. Plug in the right amount of market data and customer profiling and it could churn out reams of insights in theory. Yes, this is going to take a lot of time and training. Yet this is what AI is best at. Sorting high volumes of data that is otherwise impossible by human beings, identifying patterns and making the necessary connections. Feed in data for seasonality, manufacturing data, sales data and numerous other factors that an individual advertising strategist might not otherwise consider in a typical advertising brief. The AI does its magic. Maybe you would need some kind of pre-cursor AI to analyse the initial data and GPT-3 to write out the briefs. Then let a human planner / strategist tweak and manipulate the data and cherry pick the best insight to create a brief around it. Yes, the same cues of how an AI could be used in the creative space is also applicable to advertising briefs. It could supply a ‘sense check’, to whether the brief is on brand and is relevant. AI could be the yardstick that ensures strategy in advertising isn’t superfluous. Existing as packaging to sell creativity. We aren’t too far away from this. We already have analytics and charting tools to sort market research data in multiple ways.
AI can write copy already!
There is no real debate here. AI can partially tasked to write a lot of cyclic copy. For example, say you have a festive post that is connected to a tactical offer that you always run every year. This is typical of e-commerce. Think Amazon sales. Or Black Friday Sales. Nothing really changes when it comes to the essence of the copy, just the details. An AI is perfect for writing this kind of copy. Or e-commerce product descriptions. In fact algorithms are already being used by publishing platforms to test multiple headlines for effectiveness.
Every writer knows that every blank page comes with its own degree of trepidation. GPT-3 is just another tool that takes away some of this sting. There are already commercial AI solutions that already offer services like this. Two examples that I have heard of are Jarvis and Copymatic. If their sites are anything to be believed, there are already numerous companies that are jumping on board the AI copy generated bandwagon.
I’m sure these tools will be used by individual writers too. Those who don’t will struggle to keep pace with writers who do have a writing edge using these tools. Deadlines will also be recalibrated with clients expecting work churned out faster. But these tools as they exist today aren’t evolved enough to generate the kind of advertising copy that we need.
So what’s the holdup?
The first stumbling block is obviously cost and computing power. Let’s assume large agencies and brands manage to get over this hurdle, seeing it as a long-term investment. There’s also how well you’ve been documenting your agency’s journey across accounts. Historic data is essential for success. Everything from briefs written, to campaigns created, all need to be well documented and archived online to have the matter you need to feed and train the custom GPT-3 algorithm.
How do you quantify your agency’s secret sauce when there really is none? If you think about it, an agency signature style shouldn’t exist. Only brand styles. As a consumer you see a Mercedes-Benz ad. Not a BBDO ad for Mercedes-Benz. You wouldn’t and shouldn’t be able to distinguish a BBDO vs a Publicis advertisement for the Mercedes-Benz C-Class for example. There should be just brand ads right? So perhaps it makes sense for brands to own their own AI and use them for all their tactical campaigns, turning to agencies for the bigger creative ones.
Who will lead the charge?
For a very long-time agencies have focused on their creativity. The technical stuff is usually silently farmed to vendors outside the agency. So, it is more probable that we will have a third-party solution come to fruition than seeing a digital agency build this in-house. It would take just too long. The best bet would be a technically inclined agency. Publicis has already got an AI solution called Marcel. It’s easy to bet on them. Just as I was writing this article Wunderman released news about its in-house AI tool. That said, I think agencies like AKQA and Isobar could strike up the right partnerships to pull it off.
Truth be told, it could be any of the digital service providers, advertising giants or new kids on the block who achieve this form of automation.
One last thought
What if we are already there? What if agencies already use off the shelf AI to write that initial copy without telling their clients? Do they have an obligation to do so? Or are the ads we see in the public domain the ultimate Turin test.
I also wouldn’t be surprised if there were agencies who would proclaim that they had such AI tools but actually use old-fashioned human muscle to do the creative work. I already suspect one new service spotted online is an example of this.
For the clients, will AI enabled advertising make creative work cheaper? Or do the computing costs make it unprofitable in the long run. Time will tell. The future is written in the bits and bytes. For now, this article was not written by an Artificial Intelligence.
