Analysing use cases is the most important part of AI adoption (and the driver behind TAU)

As I have prepared for the launch of my new venture TAU, I have been thinking and writing a lot about A.I. in Marketing. Many of the commenters thanked me for my work, but asked for details on use cases. This is absolutely correct because it is in the identification of use cases and how they are challenged that truly separates the wheat from the chaff in A.I. adoption. 

For that reason, use-cases and utilisation was the starting point which has shaped the whole TAU model. As we worked through these we’ve broken them down into six big categories and around 40 sub-categories. (Expect a lot more on this on the TAU website over the next few weeks https://taums.ai/).

However, ask any expert how to do anything and the answer is always “it depends.” It does: there’s no cookie cutter solution, so instead, I wanted to talk through a couple of common examples which help explain our rationale for the TAU model. 

Let’s look at a typical use-case in Creative Production. You could just ask an A.I. to produce the creative for you in one short prompt: “Build me an ad to sell a mobile phone,” you might say. More advanced prompts will lead to better results but, at this stage of A.I., this approach will lead to pretty poor results. Today, image generators for example, can’t really handle generating text very well, but A.I. can still be highly effective if used another way. 

Chunk up the elephant

The key is to “chunk up the elephant” (a phrase many of you will know I am very fond of).

At the stage A.I. is at today it’s particularly pertinent: breaking down the tasks into their components leads to the best results. Even when it's possible to skip these steps, it's still best to understand them and build them into a process because this is how you take control of quality and progress.

Creative Production.

Consider that most simple static creative is a collection of images, a background, a theme and groups of text all assembled together. So the first step is to use an A.I. to work on each of these component pieces and use a different A.I. or tool for each.

In the past, just gathering a library of images could take a lot of time and multiple photo shoots. Today, much of this can now be replaced by A.I. image generation to a high level so the first A.I. component can be that one.

• A second A.I. can, separately, help the creative author produce copy that can be used, edited or tested.
• A third A.I. might apply or alter a theme.
• A fourth A.I. might apply different combinations to produce different sizes or formats.

• A further A.I. might then be used to check the resulting creative against brand guidelines.

It is through variations on the above process that, today, some creative teams are working far more effectively on content creation, without compromising on quality. 

Let’s take a look at another use case like Media Buying. This area seems to have been less  considered, but, I believe, it may be the area where A.I. can be used particularly effectively. Media planning is a very broad area and it's not possible here to go through it fully, but let's explore some A.I. techniques that can have huge value in media buying planning (all of which are best done to support a human media planner.)

• A traditional LLM with access to the web can, with good prompts, produce a good initial media plan and suggest data points. Try this for yourself.
• A custom LLM can be trained with an advertiser's first party data and also data from selected partners (the likes of Experian, Nielsen, TGI, YouGov etc.) This will allow the A.I. to pull insights and planning data from a broad and amazing dataset.
• A synthetic data model can be also created that will simulate people’s reactions to a range of messages and concepts. (Look out for a whole article on this concept soon.)
• All of the above can be calibrated by an A.I. against previous performance data and related metrics (attention, views). This can assist in media planning crossing over to scenario planning and forecasting as the sophistication develops.

• Crucially the outputs of the completed planning can be coded by A.I. to be able to be used directly by a range of channels and systems (different channels will need different data points. (Again more on this to follow in our Updates on the TAU site).

There’s no one size fits all solution here: different LLMs may provide better outputs for component tasks, different diffusion models may produce better image techniques, etc. Machine Learning is a powerful tool that needs to be integrated. AI agents can stitch tasks together to create a powerful whole. On top of that are traditional API, algorithmic, visual or other tools. In short, there’s a lot of variables to consider.

What both of these typical use-cases highlight is the need for an understanding of 3 core things:

• Utility -  the marketing context in which A.I. will be applied (the job to be done).
• A.I. expertise - the capabilities of A.I., the different types of A.I. and how they can be deployed
• Technology expertise - in order to be able to integrate the technology appropriately to deliver the best results.

Deconstructing to reconstruct

Essentially, anything you want to do is invariably a collection of different sub-events which need to work in harmony. The critical skill set is knowing how to break up the components, work on improvements and then reassemble them.

This requires broad marketing knowledge to know what good looks like for Marketing, but also the ability to envisage how that should be delivered technically. You also need to be able to explain the technical tasks and requirements to the relevant stakeholders internally.

An end to A.I. missing the mark for Marketers

Having a few, but not all of these aspects, is why, I believe, other attempts at helping brands develop A.I. have missed the mark. It was this frustration with the gap between the potential of A.I. for Marketing and the challenges in delivering on that which became the driving force behind TAU (Technology, A.I., Utility) 

We believe the greatest value for organisations will come from A.I.-empowered internal Marketing teams. So, the goal of TAU is to bring in the relevant combination of experts to support Marketers (having to juggle getting on with the day-to-day and making this transition). That led us to set up TAU as a collective of experts spanning Marketing, Technology and A.I. expertise, who could collaborate in different ways depending on the specific needs of clients. 

The A-Team meets Mary Poppins for Marketing

Designed to be like the A-Team or Mary Poppins for Marketing, the TAU team come in and support internal Marketing teams by working with their agencies, their CTOs, Procurement, CISOs, CDOs and all the other partners they need to engage with. Then, when the Marketing team is set up for success and no longer needs us, we move on.

I’ll be sharing more details on who we’ve got in the collective soon, but one of the reasons I’m so optimistic about the team’s ability to take on this challenge is the overlap and the fluidity between their areas of expertise. None of them are just Marketing experts, but Marketing experts with a strong understanding of Technology; our Technology experts are experienced in applying solutions to marketing; and our A.I. experts have been working on how to apply A.I. to Marketing.

In summary, in answer to the comments, yes we agree, understanding the use-cases is the starting point, but knowing how to chunk up the elephant and design the technology is the difference between mediocre improvements and game-changers – and that’s the premise on which TAU is built.