We have known personalisation to be a winning tactic for years. But what could we do if we go beyond the possibilities of yesterday, and create new content and experiences on the fly? With personalisation as the starting point, brief AI to generate new content that is totally personalised? This is what Generative Personalisation is all about.
And as this is a new concept and application of technology, there was no term for it. So yes, you can thank me for introducing a new term to marketing vocabulary. With that said, I welcome you to the world of generative personalisation.
What is Generative Personalisation? (definition)
Generative Personalisation is a form of marketing that uses generative AI models to create new, personalised content. Unlike traditional personalisation, that use past behavior and data to make recommendations or adapt existing content, generative personalisation can create entirely new content that is tailored to each individual.
We can look at the difference in types of personalisation like this:
Level of Personalisation
- Unpersonalised One size fits all. Sometimes it does, but you are using one version, one message of your content to communicate.
- Direct Personalisation: This is using data and showing that data directly in the content. Like dear “first name”, thank you for buying “product name”. Or using very simple and basic if/else statements to do the same.
- Segmentation & dynamic content: Segmentation is creating multiple versions and showing them to specific groups of people. Or to the same effect, dynamically loading pre-created content.
- Generative Content: By using Generative AI to create content quicker, or even fully automated, your potential library of content for segments and dynamic content increases. This means scaling up, 4x, 10x or even 100s of versions – say in a case you would create texts for any of your product SKU’s. This can lite the fire to your Dynamic content and segmentation efforts.
- Generative personalisation: Create unique content based on personalised AI prompts. Here we get to the new advent. Effectively briefing the AI to create content new, personalised content for a specific person.
The light blue part of the graphic stands for the introduction of content generated by AI. Going from manual creation of content to a Generative content model. As you will understand, this is not on a linear scale, as going into the realm of generated content the number of versions quickly jumps to… countless.
How does Generative Personalisation work?
The model behind Generative Personalisation is quite easy to understand.
On the one hand we have customer data and on the other we have a prompt template. By combining the data and template, we get a Personalised AI prompt. Which functions as the briefing for our AI.
Running the Personalised prompt instructs the AI to generate output, which can be stored or used directly. The output is not limited to any content form or type. Text, Images, Audio, Print, Video, Products, Emails, Social media, Code or a combination of multiple are possible.
As we automate the process of calling the right prompt and data, Generative Personalisation becomes scalable. Some marketers may be lifting an eyebrow, have a feeling of loss of control, and wondering about the quality of output. Which is understandable, but in practice the generated content can be as limited, innocent or free and more impactful as the Prompt Template instructs.
So how is Generative Personalisation different?
Reality is that we could tailor existing content to preferences, data, and desires already. Not always easy as it sound, but possible. But now the advent of limitless creation of new content opens up a whole new range of possibilities. – and yes I am aware that has been promised us for ages by tech companies and marketing agencies. Giving a bit of fire to sceptics – but as you will soon see this is something else entirely.
With the help of advanced AI systems like current version of GPT, MidJourney, DALL-E and many more, Generative personalisation allows marketers, but basically everyone, to create never-before-seen content that is unique per individual. But more important than being unique, content that is useful and leaves marketers and content creators with more options and flexibility.
The practical application of these Generative AI tools have the potential to redefine in part how marketers do their job – and bring new ways to engage, surprise and delight clients. (and entirely new ways to mess up, obviously).
An example:
A DIY home improvement seller would love to know what the next job you are doing. Remodel the kitchen? Work on the garden? Laying a new floor? A few multiple choice questions certainly lead to better matching content and offers though segmentation. Yet there would be no way to write an email that also takes into account the interior decorating style you like, products you bought, and your specific house properties. Combining those would require a unique text for each specific situation.
So instead of writing 1000s of versions, we would create a few prompts, and let the AI generate specifc personalized versions. The template and data can generate more relevant instructions and highlight product features that fit with reason you bought the products.
The role of prompt craft and prompt engineering
When creating content with AI it is to the marketer to design effective prompts. Called prompt craft or prompt engineering, a person writes a prompt to make use of the strength of AI in particular context. It takes a skill and often many refinements to get it right. Often enough the first try, called “zero shot” doesn’t satisfy completely.
With generative personalisation, an extra layer of skill is involved. Because the marketer should make decisions on how free or restricted the AI can be, which formats to use, what data to use and how broad or narrow the personalisation is.
Generating Images that are more relevant
Why would anyone want to search for hours to find a stock image that “sort of was what you were looking for”, but not quite? With Generative AI imagery you can just type in what you want to see, and press “generate”.
That is pretty great, even in todays state of image generation. Because it allows you to create images that are in the style and topic that you want. But also can be way more relevant for your subscribers and the topic of your newsletter. Let me show you what is possible. Because Generative Personalisation works to create new personaly relevant images too.
Here is the start query. Meant to look like a stockphoto, I generated the image: “man looking at laptop, over the shoulder, statistics”.
To make it more personal we could use any of the data available.
Think about the possibilities – and personalisation options we get “for free”.
- The screen display (Statistics)
- Location (Netherlands)
- Make sure he has a coffee in the picture! (versus tea, or other habits, products in the shot)
- Setting, time, tone (night time)
- Combination of any
To show some of power of the AI, the bottomright image in there of another agegroup and at the Jaarbeurs – a specific location. Which the image generator (MidJourney) also knows.
See how this could work? If can make a “stock image” multiple time more relevant. Working with generated content you can create your own stockimage library. But that is just the beginning. Again with a great prompt, we can go into Generative Personalisation mode. Just as easy as generating one image for my newsletter, I could generate an image for each individual email subscriber – based on their data. (Generative Personalisation).
Flexibility and refinement of the Generative Personalisation model
The base model of Generative Personalisation is meant to be adaptable and extendable. We can further adopt the model in practice to:
- use external content, data and variables. Like for instance the weather, features of purchased products, relevant images, text from blogs read, topics that are in a newsletter etc. This is variable data and content, next to customer data. To personalise the prompt in context. Other, fixed data may be part of the prompt, plugin, or even the AI training or fine-tuning set.
- add external logic, to make decisions about what the prompt template should look like and what and where to use it. External logic can be placed before the AI processing or after.
- add a control mechanism: a quality check on the inputs and outputs.
- add tasks and CTA’s: Besides content output, the AI can offer the options to the reader to engage in tasks (CTA’s). And it can even prepare and do tasks for you in advance. Current state of AI (march 2023) shows GPT-4 Plugins and browse enablement.
Future Outlook of AI for marketing personalisation
How does the future look? I get this question quite often, and nobody can say with absolute certainty how fast AI and Generative Personalisation will catch on. But a few of the following are already gleaming at the horizon, so why not open that little door.
+ Preform more intricate and free form tasks. A Toolformer-like AI mechanism would allow it to be trained to use tools to perform tasks by feeding it examples.
+ future data and prompting mechanisms: It is very likely that imagery is about to start playing a bigger role in customer data, as is open (unstructured) text. These can just as well be used in prompts… exciting times!
Looking to get inspired or start using Email + AI?
Enthusiastic about AI? Want your team to be updated? Or even work on strategy, a pilot or experiment. Shoot me over a message, or reach out on Linkedin to chat.
