By Senior Strategy Director Emma Woodward-Church
AutoML promises a revolutionary path to sophisticated machine learning and predictive data models – made more widely accessible to much larger array of business stakeholders.
‘Plug in, automate and go’ algorithms that uncover the hidden gems in your data such as propensity to buy, at risk of churning/lapsing and customer lifetime value, to name but a few.
Let’s start with a quick refresh, to clearly and correctly differentiate artificial intelligence and machine learning. Henrik Nordmark, Head of Data Science at Profusion explains this nicely: “Machine learning is a subset of AI – it is the subset that follows a bottom-up approach in which the algorithm learns from data. This is in contrast to Symbolic AI, which follows a more top-down approach in which the algorithm is in principle supposed to already have all the intelligence it needs already built into it from the very beginning and thus without the need to learn from data”.
AutoML is the next step to making AI and machine learning accessible to everyone.
However, the data scientists’ role within AutoML remains critical.
You may assume that automation means an end to expensive data scientists, but this is not the case at all, what AutoML enables you to do is get the best value from your data scientists. By removing some of the time consuming, tedious and repetitive work, we allow the experts in this field to focus on what they do best, which will ultimately increase retention and reduce the costs of churn.
As a Strategy Director in the realm of marketing, this big explosion of smart and advanced data models, with less cost and resource implications, means my clients suddenly have ‘the promise land’ in terms of big data opportunities, regardless of job role or industry.
My job is to ensure AI is a fundamental part of any marketing strategy across industries and I’m fortunate to have ‘on tap’ access to an incredibly talented team of data scientists who have been developing a variety of AI and machine learning models with superb accuracy levels.
In this blog, I’m going to explain the what, why and how of AutoML so let’s get started with the WHAT…
1. What is AutoML?
Let’s see what Wikipedia says including my interpretation:
· Automated machine learning (AutoML) is the process of automating the process of applying machine learning to real-world problems.
· My take: AutoML represents a fundamental shift in the way organisations of all sizes approach machine learning and data science in the future
· AutoML aspires to cover the complete pipeline from the raw dataset to the deployable machine learning model.
· My Take: Applying traditional machine learning methods to real-world business problems is time-consuming, resource-intensive, and challenging. AutoML speeds up the Data Science process.
· The high degree of automation in AutoML also aspires to allow non-experts to make use of machine learning models and techniques without requiring becoming an expert in this field first.
· My take: plug in prediction models that provide clear efficiencies for savvy self-service business users
· Automating the process of applying machine learning end-to-end additionally offers the advantages of producing a faster and more scalable creation of those solutions.
· My take: cost efficiencies at scale!
2. Why AutoML?
You have lots of 1st party data. You want to access and uncover data you have not been able to before. Make your data strategy smarter and monetise your data more effectively. You want the ability to predict things before they happen and gain advantage on your competitors.
Then Covid-19 happened and sadly continues, so things look different. You want to adapt to a new normal and AutoML can help with that.
It promises automation which equals cost effectiveness and efficiencies within the business. It aims to be platform agnostic and plug into anything that supports an API! That includes your CRM, DMP, CDP, email platform – the lot!
Data Science solutions are suddenly within reach of any department, any role within any industry. There is no faster way to see the impact of that than in marketing.
However, there are caveats and it’s important to understand these. To explain these further I’ve asked, once again, Henrik to outline the restrictions to provide a full picture.
“The promise of AutoML is to streamline the data scientists project delivery process. This does not mean that the role of a data scientist will disappear, it means that a data scientist will be able to do more with less of their time. If we think about that process, it starts with translating a business problem into a mathematical problem that requires a certain amount of imagination. We get generic requests like, ‘I want to increase sales’ and a data scientist’s role is to tackle this and consider what prediction would make this a reality. AutoML can’t do that bit because a machine still doesn’t have the capacity to translate a business pain point or requirement. Once the business requirement is clearly defined in a mathematical context, AutoML can be applied to do its magic to help us quickly identify some good predictive models.”
So, we understand that AutoML can provide access to cool predictive models and accelerate the time and reduce the cost but it’s still crucial the models are sense checked, for this we still need an expert. Now on to the ‘how’…
3. How does AutoML work? Here’s the high-level process.
a) Data cleaning – getting data into a place where it can be effectively analysed can be a very lengthy process. Some AutoML platforms offer this but the reality is that this part often happens outside of the AutoML platform because it’s vital the data is in a good place in order to be analysed.
b) Feature engineering – this is the process of creating new variables out of existing variables. Again, this part of the process often happens outside of an AutoML because human imagination and expertise are required to establish which variables might be more predictive than others.
c) Model selection – is where AutoML can really support the whole process and run the same data through several algorithms to determine which algorithm can learn best on your data.
d) Feature selection – is the process where you automatically (AutoML) or manually select data features which are deemed most important to a prediction variable or output.
So, we have completed a mini exploration of AutoML. We have briefly covered the benefits and the challenges.
Next week we will explore how Profusion has responded to AutoML.
A combination of the best of the ‘auto’ part whilst still leaning on the experts to design a series of invaluably accurate predictive models to super charge your marketing efforts.
It’s been especially developed for digital marketers to hide away all the complexity of the Data Science piece in order to make it an accessible part of your marketing strategy.
Our solution is called Ai Marketer and here’s a sneak peek at the models that plug right into your environment whatever that may be.
Over the coming weeks we will explore each model in more detail and highlight the benefits for you…
Introducing Ai Marketer – an AutoML platform that helps democratise AI specifically for digital marketers.
|Ai Marketer Models||Outcomes for your business|
|Customer churn prediction||Catch them before they are stop buying|
|Customer lifetime value||Predict your next best customers|
|Propensity to buy||Find customers who are ready to purchase|
|Email engagement prediction||Engage them before they opt out|
|Send time optimiser||Deliver at the right time for each individual|
|Inbox Accelerator||Protect and improve your sender reputation|
|Frequency capping||Stop emailing before you reach their limit|
I would like to finish by highlighting the award we won for Ai Marketer, The University of Essex Best Knowledge Transfer Partnership Award. Ai Marketer was originally designed and launched to support Charites with their customer engagement and continued donations.
Read part 2 here.
Until next week! Or get in touch: email@example.com