10 ways AI can Improve Marketing Outcomes (Part 2)
GenAI is all the rage nowadays. That’s no bad thing and Large Language Models (LLMs) are already disrupting marketing and eCommerce. But there are other ways AI can boost marketing outcomes that are quicker, easier and most importantly cheaper.
In Part 1, we explored five AI techniques to improve marketing outcomes and here in Part 2 we’ll explore the remaining five.
The 10 AI Techniques to improve your Marketing Outcomes
Segmentation
As the saying goes, to treat people equally, you have to treat them differently. Some customers might be highly engaged and welcome additional communication, others might not be. Some customers might spend regularly, some might not. And so on...
Segmentation is a way to group customers by these kinds of factors to enable you to monitor performance and more importantly personalise your content.
- RFM: RFM stands for Recency, Frequency & Monetary Value and it’s a highly effective way to split your customer base to answer questions such as “Who are my most important customers?” or “Which customers am I in danger of losing?” and perhaps more importantly “Which customers should I target?”. If you don’t have an RFM model, you should probably stop reading this and go make one.
- Behavioural: The next biggest set of Segmentation models are behavioural. These use a variety of different data points to group customers into buckets based on similarity. Now, if you have a few data points (like our RFM model above) you can write rules for it. If you have a few hundred data points, you’ll need an AI clustering model such a K-Means which groups customers into buckets based on shared characteristics.
- Demographic: My personal favourite, Demographic models, unsurprisingly, use customer demographics to split your customer base. However, most companies only have a few demographic data points to use and that’s where Open Data comes in. If you have an idea of your customer’s location, you can use sources such as the UK Census to supplement your data. The UK census in particular is a treasure trove of detailed & granular data encompassing demographic, household and economic data points. The downside is that it requires a lot of work to engineer it into something usable but that’s a blog for another day! 😉
Most companies are already doing some form of segmentation on their customer base, but there’s definitely levels to this game and it’s one of those things, the more effort you put in the more you get out with analysis, unexpected insights and new strategies.
It’s also a tried-and-tested winner, with Segmentation providing a healthy boost to almost all key Marketing KPIs.
Churn
You know that RFM model that you have? Well, keep the data from that because you’ll need it for your Churn model. Churn models predict the likelihood of a customer having left the orbit of your company
Now, there’s no such thing as a magic money tree, but a well developed and implemented Churn model is the next best thing. Acquiring new customers is between 5 and 25 more times expensive than retaining existing customers and increasing your retention rate by 5% results in a 25% to 95% increase in profit.
Most companies approach this challenge incorrectly, treating it as a binary yes or no answer (we call that a “classification problem” in the AI trade). As with anything, the truth is neither black nor white but in shades of grey and that’s why it’s much better to take a probability-based approach with a Bayesian Churn Model.
This uses previous customer behaviour to predict Churn Risk as a percentage rather than a simple yes or no.
Not only is this approach more quicker and more accurate, but it enables earlier warning, different intervention strategies for different levels of churn and if you’ve segmented your customer base with something like RFM, you can prioritise your intervention actions effectively too.
Price Optimisation
Whenever I think about pricing, I’m reminded of the excellent Starbucks Economics article by Tim Harford and buying a coffee has never been the same for me since.
Setting the right price is both an art and a science. AI-based Price Optimisation, is a game-changer allowing businesses to maximise their profits while maintaining customer satisfaction.
Price Optimisation models use both historical and real-time data and insights to predict optimal pricing strategies. For instance, they can recommend when to offer discounts, how much to adjust prices during peak seasons, or when to increase prices and by how much without losing customers.
Recommendation
Unless you’ve been living under a rock, you’ll have heard of Amazon and Netflix, and these two tech behemoths have made recommendation engines famous.
These AI-powered systems analyse customer data, preferences, and behaviour to suggest products, services, or content that users are more likely to like, appreciate and most importantly part with their hard-earned cash for.
There are three main types of recommender system:
- Content-Based: Identifies previously purchased or engaged-with items and recommends similar items (e.g. you bought a pair of gloves! Want some more gloves?)
- Collaborative Filtering: Identifies items that similar users have purchased or engaged with (e.g. people like you bought gloves! Want some gloves?)
- Hybrid: As you might expect, a mix of Content-Based and Collaborative Filtering.
Which is best? As always that depends, but whilst Collaborative engines are a little more work, they’re more nuanced and generally a safer bet for increased ROI. A human can only have so many gloves.
In terms of impact? Well the fact that Amazon are the fifth biggest company in the world and built their whole business on recommenders, and Netflix estimates that their recommendation system generates then $1bn a year tells you all you need to know.
Forecasting
We can’t predict the future, but accurate forecasting is the next best thing.
Forecasting in marketing, allows businesses to stay ahead in a competitive landscape. It enables precise predictions of market trends and customer behaviour and early warning of potential rough times or choppy seas leading to better decision-making and informed intervention strategies.
But how do you do it, I hear you ask? Well as always, there are options:
- Time Series Forecasting: Time series algorithms have been around for as long as time itself. Nearly. They’re still a good bet if you want to do something quickly or either don’t have or don’t want to include any data over and above the time-series itself and can produce accurate results at-scale.
- Machine Learning Forecasting: If you do have some additional data to go along with your time-series, the Machine Learning might be a good bet. You can also include time series Forecasts and their components in your model as well
- Deep Learning Forecasting: Deep Learning Forecasting is a newer field, accomplished through the use of something called a Long Short Term Memory (LSTM) network. Reports from the field indicate that it performs slightly better than Machine Learning forecasting to the tune of around 10% at the expense of additional complexity and processing power.
About
Tom Ewing is Head of AI and Data Engineering at Station10 and has developed a churn model to predict when he’s going to get fired.
Check out Station10's Awesome-Marketing-Machine-Learning list repository on Github which contains a list of open-source AI tools with applications in marketing.
Station10 believe in open source and sharing knowledge and our Medium content is never paywalled, so please leave a clap if you’ve found this useful. This helps motivate us to write more content and gives us a better idea of what to write more about.
