Large language models (LLMs) have reached new levels of capability and accessibility, leading to the perhaps biggest boom in AI chatter since voice assistants first came onto the scene. (Remember how everyone in the family had an opinion on Siri in 2011?)
But what exactly are LLMs and how do they factor into AI? Here’s the quick context:
-
LLMs are a type of machine learning-trained model for language understanding.
-
The spotlight has been on generative AI: a special use case for LLMs that underpins recent AI advancements like the consumer-facing ChatGPT and other GPT models.
-
These generative AI tools have recently become available to the average person to test and use due to a significant increase in parameters available to the GPT model: from 175 billion to 1 trillion and growing.
-
-
(To put it into perspective, 1 billion seconds is about 31 years whereas 1 trillion seconds is roughly 31,000 years.)
-
For marketers, there’s a lot of buzzy conversations about everything that generative AI will unlock: hyper-personalization, AI-generated marketing personas guiding media strategy, GPT interfaces for data analytics and insights, and other efficiency-driving advancements.
But to take advantage of all that generative AI has to offer, marketers need to get their data ducks in a row. It is essential to have a solid data foundation together with a person-based identity solution before generative AI can play any kind of substantial role in bolstering campaign efficiency and effectiveness or driving deeper insights.
There is no replacement for quality data. You can apply the best, smartest, most sophisticated models to mediocre data—but without quality data, you will only ever get mediocre results. Capturing and cultivating your first-party data asset and then enriching it with high quality third-party data lays the right foundation for AI to work its magic.
In this blog, we address some common questions around AI, including where it’s headed and what you can do to get the most out of all kinds of AI possibilities for your marketing.
First, we’ll clarify some terminology.
Predictive or Generative AI: What’s the difference and where can they come together?
Predictive AI: Predictive AI helps marketers decide whom to reach, where to reach them, when to reach them and what to say to predict the most successful outcome.
-
Predictive AI uses statistical algorithms and historical data patterns to analyze data and forecast outcomes.
-
It is sometimes referred to as “predictive analytics”; however, the difference between predictive AI and predictive analytics is that predictive AI is autonomous, whereas predictive analytics relies on human intervention.
Generative AI: In contrast, generative AI helps marketers create: It generates content.
-
This includes written (like ChatGPT) or visual (like Dall-E) formats—across a wide array of needs, like writing an essay on the Civil War, generating 1,000 potential headlines for an article or converting text copy to an image.
-
Generative AI is unlikely to replace creative work, but rather will be a force multiplier that will accelerate creative ideation and generation. It will always be important to keep a human in the loop of the creative cycle for myriad reasons, like brand safety and standards, copyrights and privacy.
The consumer is the beneficiary of predictive AI decision-making, but they don’t necessarily know it because everything is happening behind the scenes. Predictive AI helps marketers connect with in-market consumers in a relevant, non-repetitive way that they appreciate. Predictive AI is used in marketing to define:
-
Whom to message
-
Where to reach them
-
When to talk to them
-
What to say
What generative AI can add to this process is the ability to create that message, ad or email dynamically and in response to the identified person’s preferences, wants and needs. It becomes the final step in the process, but you need the predictive foundation to provide the inputs for the generative AI side.
We see predictive and generative AI working together in the same way that art and science work together. Think of predictive AI as the science—it’s the methodology that enables deep insights and data-driven decisioning using machine learning techniques to produce better business outcomes. It’s powerful, but not terribly sexy. Generative AI is like the art—it can bring a more human understanding of data and insights through content generation that can evoke emotion. When you put them together—the art and science of AI—you can drive deeper insights and better decisions across platform and channel.
While generative AI can create content, predictive AI is needed to apply that content to the right person at the right time. We already have the science of marketing down with predictive AI; now we have started using generative AI to bring art to that equation.
How to build a strong predictive AI foundation
OK, so what does “strong” predictive AI look like? For one thing, time is of the essence.
AI learns and is trained over time. There’s no way to speed up the process. The best solutions should have a sizable amount of time training their predictive AI models already under their belt.
What does time do for predictive AI? Well, the benefit of more training time means that those solutions are faster and notably more accurate than solutions with less experience. This is because the model learns the data and improves over time.
Think back to the example of a marketer locating an in-market customer for a product using predictive AI. Over time as the model evolves and improves, it can inform technology like an ad server to find in-market customers faster and more accurately. It’s the difference between messaging someone on the verge of a purchase versus someone who has already purchased.
Marketers should look out for solutions with real-time model updates—not those that take weeks or months—because it allows for real-time customer connection (at scale).
Optimizing data for best AI functionality
AI needs the right data to train it, but your outputs are only as good as your inputs. Here are three ways to get AI-ready:
-
Data foundation
You need to enrich your first-party data to make it viable for AI applications in a privacy-safe way. Apply hygiene to it: cleanse, structure and reinforce. Think of your data as a stream, with AI requiring a continuous feedback loop to be effective.
-
Access to data at scale
Connect your data to a people-based identity. You need data about your customers everywhere they are, not just from your interactions. If you don’t have that robust identity resolution in place, connecting the dots just isn’t possible.
-
Readiness
Make your data asset accessible to AI methodologies. (This part is greatly aided by partnering with a marketing solution provider that can effectively apply AI to your data—and even get your data ready for AI in the first place, so no effort needed.)
Prioritizing consumer privacy and data ethics
Amid all the generative AI hype, it’s crucial marketers don’t neglect the importance of centering consumer privacy and data ethics.
Generative AI makes it possible to create an entire ad (from copy to visuals to call-to-action) in real time. Connecting predictive and generative AI could determine which creative will have the most emotional impact while also reducing waste and driving efficiency by only creating assets that are used. This hyper-responsiveness must be balanced against brand safety, content appropriateness and legality, ideally with a human in the loop.
To ensure data privacy and ethics compliance, a team of real people should monitor any generative AI outputs—you can’t just set it up and let it run. This team needs to ensure everything meets brand standards and does not infringe on copyright. Jodi Daniels, a data privacy consultant, said it best in a recent Forbes article: “A business could be put in a precarious position if it uses generative AI in a way that uses consumer data that runs counter to contractual obligations.”
Keep learning—just like good AI
Just like strong AI, it’s crucial that you never stop learning and stay up on the latest information.
As a starting point, we recommend checking out this Q&A with Steve Nowlan, our SVP of decision sciences analytics, on how organizations can truly harness the power of artificial intelligence. He answers some questions that we think just might generate some questions of your own to keep the conversation going.
And if you’d like to hear how Epsilon’s CORE AI makes real-time marketing decisions at the individual—not the audience—level, head to our webpage.
