The potential of generative AI to revolutionize marketing has been widely reported and discussed this year. McKinsey has estimated that gen AI could increase the productivity of the marketing function by a value between 5 and 15 percent of total marketing spend, while a Bloomberg Intelligence report suggested gen AI-driven ad spend is set to grow by 125% to $192b over the next 10 years.
There has been a plethora of eye-catching announcements, including the partnership between WPP and Nvidia – which promised to produce advertising creative “more efficiently and at scale” – and Coca-Cola’s “Create Real Magic” campaign, which enabled people to generate creatives using gen AI.
But gen AI’s role within marketing effectiveness has been less widely discussed. One of the reasons for this is that we are very much at the beginning of a journey so it’s too early, for example, to measure the impact that gen AI-based media campaigns are having on an organization’s bottom line.
Nevertheless, we are excited about where this journey will take us. Conversations we’re having with our clients show that marketers agree. As you read about the role that gen AI could play in the three areas we focus on this article, it’s useful to think about whether the technology can make what you already do more efficient, or whether it can help you do what you can’t do currently.
With that in mind, let’s look at the three areas:
1. Can gen AI help to improve my data assets?
Access to the best possible data is crucial to a successful marketing effectiveness program. Without it, the ability to uncover insights about your marketing investments and drive growth is compromised.
Exploratory data analysis (EDA) – the essential step before raw data is inputted into a statistical model, such as a marketing mix model – is one existing process that could benefit from gen AI. EDA helps to ascertain things such as what kind of raw data has been supplied and whether it is in the right format, is consistent, or has any gaps. AI and machine learning are already being used to increase the speed and efficiency of EDA, but gen AI – specifically large language models (LLMs) – has the potential to enable us to go further thanks to the launch of dedicated enterprise versions of well-known gen AI tools.
In August, the poster child of gen AI, OpenAI’s ChatGPT, unveiled a version of its LLM for business. As well as promising enterprise-grade security and privacy, ChatGPT Enterprise includes unlimited access to an advanced data analysis capability that combines the language skills of ChatGPT with Python-based coding. Crucially, it also lets users upload and download files. A month later, the company unveiled voice and image capabilities for ChatGPT that enable users to take a picture of a chart, document, or graph. Such tools mean the efficiency of EDA could be increased by speeding up the way in which data is collected, processed, and readied to generate insights.
The second area worth investigating is synthetic data, which can be defined as data that has been artificially generated using some of the same algorithms that power gen AI. Brands have been using synthetic data for years – Amazon has used it to improve the accuracy of its contactless payment service, for example – largely thanks to the work of Harvard statistics professor Donald Rubin and his seminal 1993 paper on the subject.
But recent advances mean it could fall into the “help you do what you can’t do currently” category. A 2022 report by The Royal Society and The Alan Turing Institute in the UK noted that synthetic data has “the potential to address issues of fairness, representation and bias through data augmentation”. Meanwhile, this Marketing Week article discussed research that showed how synthetic data was used to create perceptual maps in near time.
At a time when marketers are all too aware of the challenges involved in getting hold of the right data due to data privacy regulation, digital walled gardens, the fragmentation of audiences across a growing number of channels, and the increasing cost, such advances hold promise.
However, marketers should carefully consider what use cases may be appropriate and what budget and resources they have before going ahead with a synthetic data project. As The Royal Society report noted, synthetic data “should not be considered a replacement for ‘real-world’ data” due to issues including accuracy and privacy.
2. Can gen AI help to generate more valuable insights?
Understandably, this question is one that many of our clients are starting to ask us given the current buzz around the technology. Before we answer it, it’s important to note that, similar to how we discussed the application of LLMs to EDA, more sophisticated, enterprise-grade techniques and tools are needed to generate meaningful insights from your data.
If you’ve used one of the main LLM-based products, such as Bard, Claude, or ChatGPT, then you’ll be aware of the value of prompting – using queries or instructions to elicit the output you want.
However, to get the best possible insights prompting should not rely solely on an LLM’s representation of information. This is where another AI framework called Retrieval-augmented generation (RAG) comes in useful. As this IBM article explains, RAG supplements the LLM with external sources of knowledge, ensuring it has “access to the most current, reliable facts, and that users have access to the model’s sources, ensuring that its claims can be checked for accuracy and ultimately trusted.”
It’s important to be aware that using RAG can require substantial computational resources and model training. As with any gen AI-related tool, you will also need to mitigate any bias and inaccuracies that may occur.
Fine tuning can also help with insight generation. This technique trains an LLM on many more examples than is possible with prompting, meaning you can get more efficient and consistent insights. Fine tuning needs high-quality data to work effectively, which requires budget, time, and skill to create, but ongoing updates to ChatGPT are making the technique easier to do. Businesses can now create their own customized versions of ChatGPT for specific use cases, for example.
How you apply RAG or fine tuning will depend on a wide range of factors, but identifying opportunities and trends in unstructured data is one use case that should be universal. Analyzing customer feedback, news stories, or social media posts could yield insights that inform new sentiment analysis, new audience segments, or new market adjacencies, for example. It’s hard to know how much of an opportunity this represents but this article claims that 90% of unstructured data is never analyzed.
Overall, it’s very probable that a marketer using techniques such as RAG or fine tuning will be able to generate a larger quantity of the same value of insights compared to a marketer who is not using such techniques. However, there is no guarantee at this stage that these insights will be more valuable. As such, they should be viewed as a complement to, rather than a replacement for, insights generated using existing methods.
3. Will gen AI augment or replace humans working in marketing effectiveness?
When asked to name the biggest challenges that could arise from the use of gen AI, 48% of marketers in a survey published in the Wall Street Journal predicted that their own teams would shrink.
While such views are understandable, humans are essential to the successful implementation of gen AI. This is particularly true for those who work in marketing effectiveness. While having the ability to unearth insights is a key building block of best-in-class marketing effectiveness, unless they are validated and actioned then brands will not see the value. Talented humans are best placed to make this happen.
To ensure success, it’s crucial that marketers align with other stakeholders on how the business is going to use gen AI and its likely impacts. According to this Marketing AI Institute report, 72% of organizations do not have policies in place to guide the use of gen AI. Transparency is particularly important – you can’t have a situation where gen AI is working on something when the business thinks it is a human, for example. The 12 principles for the use of gen AI in advertising developed by British industry bodies ISBA and The IPA are a useful starting point for marketers looking into this area.
There’s another way to answer our initial question. We know there is a talent shortage – the UK Chartered Institute of Marketing’s most recent report that benchmarks digital marketing skills, for example, found such skills are stagnating or declining. Gen AI could be used to upskill marketing effectiveness teams with the data, analytics, and insight capabilities they need to succeed. It could be particularly useful given when, as this Econsultancy report notes, digital skills need updating “at least monthly”.
The start of an exciting journey
We are still at the beginning of understanding what gen AI can do for marketing effectiveness. In the majority of examples discussed here, gen AI can help marketers to do what they already do more efficiently. But there are new capabilities and use cases appearing all the time that means its ability to help marketers do what they can’t do currently is likely to improve. It is also likely that gen AI will create new areas of work that don’t exist yet.
What we can say with a degree of certainty is that today gen AI is a productivity tool rather than a tool that will replace marketing effectiveness staff. Given gen AI is a complex area that is evolving rapidly, marketers should engage with experts in the space to ensure they are heading in the right direction – but be wary of those who say it’s easy or have off-the-shelf solutions. What is more certain is that if you’re not using gen AI in the near future then you will be at a significant disadvantage.
Contact Gain Theory to discuss the role that gen AI could play in marketing effectiveness in your organization.
A version of this article first appeared on ana.net.