AI Marketing Insights Data-driven Advertising

3 essential ground rules for marketing analytics in the age of AI



Open Mic article

This content is produced by a publishing partner of Open Mic.

Open Mic is the self-publishing platform for the marketing industry, allowing members to publish news, opinion and insights on

Find out more

June 21, 2023 | 7 min read

It’s a pivotal time in marketing analytics, as ML and AI applications are increasingly used to drive the best possible customer experiences

Kevin Lyons, general manager of analytics services at Acxiom, explains the ground rules brands must consider to ensure their marketing analytics projects succeed.

From where I sit, the hype around generative artificial intelligence (AI) is deserved. We recently witnessed ChatGPT take half an hour to deliver the same conclusion it took my team weeks of research and testing to reach, which was both exciting and daunting.

In this particular case, we could validate the AI’s solution because we already knew the answer to the question. We’ve also seen cutting-edge generative AI spit out nonsense (also known as 'hallucinating'), so it’s still early days, and you can’t and shouldn’t totally trust it. Validation is key, and not all use cases are as easy to validate.

There’s also still a divide between solutions that learn from your input and seem to generate better results and solutions that are more secure but also limited in the data they have access to. As Acxiom CEO Chad Engelgau recently shared, if we want a positive AI revolution, there’s no room for bad data.

While there are many issues with generative AI at this early stage, there are still incredible developments taking place in the application of machine learning (ML) and AI for analytics and business decision-making. Previously used as tactical tools, generally to make very quick and specific decisions, they are now moving further up the strategy chain.

Handing over decision making

In theory, with perfect and trusted data, it could be possible to reach a scenario where all business decisions are made by machines doing math.

If you can access and gather all the necessary data from all sources across the business (no small feat) and set a specific outcome – for example, a corporate revenue target – AI could make recommendations to optimize the whole business to achieve that outcome. It could decide what products to develop and which ones to sunset. It could determine which suppliers and vendors to use and to avoid.

In reality, we’re not there yet. There is much work to be done across technical, business, and ethical fronts to get it right.

On the technical side, we’re still in the early stages of building fully performing applications. On the business side, we’re experimenting and assessing risks. Perhaps, most critical is the ethical side, with many calling for a pause on the training of AI systems more powerful than GPT-4. Society still needs to make many social and ethical decisions regarding AI uses that go beyond a few business metrics. But smart, AI-informed business critical decisions will become possible, and the gap between businesses that are on this data-driven path and those that aren’t is getting wider by the day.

Marketing decision-making is all about CX

When we look at marketing business decisions in particular, it becomes about delivering the best possible customer experience for each individual. Determining that ideal experience requires many small interacting decisions – at a speed and volume humans could never manage alone.

What offer to make? What channel to use? What time to engage? What creative to choose? And how do these decisions impact one another? AI can work through the possible outcomes and combine them to deliver one overarching solution – the best possible experience for that specific customer. The experience that will also drive the desired business outcome.

Given the more limited scope of this problem and the general availability of marketing data, we’re far closer to that reality than allowing AI to make wider business decisions. But even within marketing, caution is required, as there’s no simple solution, and many of these initiatives will fail.

The ground rules

All models come with a reasonable risk of making bad decisions which can adversely impact customer experience. However, there are three ground rules to give you the best chance of success and help avoid some of the common pain points in machine-driven marketing analytics projects.

1. Insight is useless without action

Far too often, analytics projects are set up to generate insights. To boost knowledge. But insights themselves should never be the end goal.

Data should be used to gain insights. Insights should be used to inform decision-making. And those decisions should drive action. The term ‘decisioning’ refers to a combination of decision-making and taking action, but too many analytics teams stop at the first without implementing the second.

2. Models must be able to be operationalized

It’s important to remember AI needs to be operationalized within the environment available. It’s easy to get carried away with the latest developments and create solutions without thinking about what is practically possible to implement.

There are various steps within an analytics project, including understanding the problem, gathering and preparing the data, doing the analysis (the math in the middle), implementing and operationalizing the model, and then fine-tuning it. Analytics teams are generally getting better at most of these – especially the data part – but it’s still scarily common to get past the analysis stage and realize there simply aren’t the capabilities, the infrastructure, or the resources to set the results to work.

It’s far better to think realistically about what’s possible and limit your solution to something that fits with what you can act on today than to have an overly sophisticated model sitting on a shelf that never gets used. Focus on generating results right away, and you can always up your game when the capabilities or resources are available to expand.

Further, operationalizing analytics on a large scale requires varied skill sets, so it’s pertinent to have a strong mix within analytics teams including data scientists, applied scientists, and ML engineers.

3. Guardrails must be put in place

As we’ve discussed, AI is capable of making the ‘best’ decision to achieve a particular outcome or coming up with the most effective answer to a problem. But that doesn’t mean you’ll like the decisions it makes. It might, for instance, shut down a marketing channel that’s not performing. But if that’s a channel where you expect significant growth over the coming years, you’ll probably want to overrule its decision so you can lay the foundations for the future.

You’ll always need to put strict parameters in place to ensure the decisions made and the actions taken fall within certain boundaries. This will inevitably include ensuring any decisions and actions are legal, ethical, and socially responsible. We’ve all seen examples where ML makes recommendations that are morally questionable. AI isn’t a magic box; it needs rules and human oversight. Putting guardrails around the math helps ensure it doesn't do things you don’t want.

Keep the ground rules in mind

The use of AI for marketing analytics is developing rapidly and progress in generative AI will only accelerate it. But the above principles will still apply. No matter how advanced the math, you’ll always need to make sure analytics drives action (not just insights), any solution is able to be operationalized, and there are effective guardrails in place. This gives your analytics projects the greatest chance of success and allows the math to make the decisions that will deliver the best possible customer experiences.

AI Marketing Insights Data-driven Advertising


Industry insights

View all
Add your own content +