We are all used to algorithms, they pervade our everyday lives
This from Paresh Patel, consulting services director at Qbase…
Whether it’s part of Google’s website rankings or the recent A’level results debacle, the formulas make decisions about us and our environs all the time, and many to great success. Machine learning is being used in everything from creating music and art to providing farmers with predictions on crop success via their tractors; and of course algorithms were employed widely when governments around the world were making
Making sense makes sense
Data modelling using machine learning has always been employed in the marketing department to great success. And with the vast increase in the volumes and sources available to marketers or customer journey owners, utilising algorithms/machine learning to make sense of that data, makes, well, sense.
But how do you start using your (and others) data to deliver more effective marketing or customer journey campaigns? Here’s what you need to know.
Supervised and unsupervised
There are lots of different ways that machine learning works for data modelling purposes. Typically, these fall into two categories: supervised and unsupervised. Supervised sees a data scientist working with data to make decisions based on its outcomes. The recent A’level standardised model was a supervised approach where it simply applied a set of rules to the data.
Unsupervised data modelling is where algorithms learn by themselves, based on patterns found in the data sets. While unsupervised often delivers insights the businesses hasn’t thought about, often the approach fails to take into account broader business knowledge or external factors that could impact the accuracy of the predictions it makes.
Once you’ve decided on the approach, there are a number of different types of data modelling using machine learning that can be used for improving marketing campaign success. The best approach really depends on what you’re trying to achieve – whether it’s improvements to segment your customers, customer journey understanding, reducing churn, next best action, and so on. Typically there are three models that marketers may use:
Descriptive models – this allows customers to be grouped together according to shared demographics, behaviours and other factors.
Predictive models – look at the probability of an outcome. For example, whether the customer will place a second order, or the likelihood of churn.
Prescriptive models – not only looks at the probability of an outcome, but also presents the business user with specific options and determines which are best based on certain criteria. For example, users can integrate the model to establish the best decision according to
predefined criteria such as maximizing profitability and throughput.
All three model types are really useful to marketers and customer owners. Organisations can use them to identify an individuals’ affinity towards a product and then assign engagement scores. Or can help you understand when to sell the product to the right customer at the right time, or even use predictions to place a value on the customer across their buying lifetime – ensuring that high value customers are treated differently.
We’ve seen data modelling work incredibly effectively for charities who are looking to optimise their marketing spend and ensure that they communicate to supporters who are more likely to engage with their products and services.
Businesses are now combining a variety of models into what is known as model factories where multiple outcomes are considered at different levels to meet overall objectives. For example you could have a model that predicts the likelihood of a customer placing a second order, then another model that looks at what channel they are likely to order from and then a sub-model to identify what product to promote to support the conversion to second order.
However, being able to refine and deliver effective campaigns depends on more than just your models. Here’s what else you need to think about:
Data modelling predictions are only as accurate as the data itself. While 70% accuracy or more is the goal, accuracy also requires good quality data that is relevant to the outcome you are trying to predict or describe. Ensure that you have governance processes in place to manage data end to end and clean and standardised where you can. Having more volume, data breadth and variety, and better data quality all help to support accuracy.
This is so often where businesses go wrong. The challenge of machine learning is knowing what you’re trying to achieve before you start so it’s important to have a user case to support this so you and everyone else who will use its output are aware of what you are going to achieve. Machine learning can be really powerful but if you don’t know where you’re trying to get to, your decision-making next step isn’t likely to deliver.
Interpretation and understanding
From the outset you need to know what you are modelling, what assumptions you are making, data labelling, and so on. This ensures that you’re giving data meaning. While you might discover customers who respond to email then buy online, what does that mean and how does that impact the campaign in the future?
Testing and validation
Make sure you try out your data model. Test, validate, split the model, re-validate the model. This is important to the accuracy of your outcomes because data changes, systems change and even campaigns change so you need to make sure that you’re incorporating these.
A final word of warning – make sure that you manage expectations all the way through. Often you’re not going to solve all of your challenges from data modelling using machine learning. For example, perhaps during the lockdown, you created a model that shows how to increase sales by 20%, but at the same time you’ve lost 50% of your customers – the outcome is going to be different than you first thought.
Good continuous communication with your colleagues who will use the results of this model to support business outcome is essential, together with reports that provide visibility in a way that they understand would be recommended.
Effective marketing is data driven
Machine learning has lots of applications for marketers. We know that the most effective marketing is data driven, and with so many tools available to deliver customer information captured at every stage of the buying process, there’s no reason for mis-targeted campaigns or ones that deliver low ROI. If you understand your data and really know how to use it, the results can be immediate.