Machine Learning techniques can help marketing departments to solve complex data problems – but marketers beware, you first need to ask yourself, is the problem you have complex enough to require a Machine Learning based solution? For example, a very common use case would be customer segmentation and classification. More complex classification may require Machine Learning, however most problems surrounding customer segmentation can be solved with CRM or marketing automation tools.
There are some cases, however, in which the application of Machine Learning is doubtless necessary:
- Product recommendation: the most widely used case within the ecommerce sector – think Amazon, Netflix and Spotify
- Voice and text recognition: or conversational commerce. Is there a Marketing Director out there who hasn’t already developed a roadmap for a chatbot or an Alexa skill?
- Image recognition: this includes both facial recognition and brand recognition – Blinkfire Analytics is a great example of this. Thanks for Machine Learning, it’s possible to measure brand impact, by the second. We can develop smarter sports sponsorships or create more intelligent video content with solutions such as Vilynx.
Blinkfire Analytics uses computer vision to measure media value accurately and in real-time
Let’s focus on one problem at a time. Imagine the scenario:
“I want to evaluate the ROI of sponsorship of a basketball team by measuring the number of purchases made by women aged between 18 and 35, who have selected one of a number of sports products recommended to them, weighted by the average number of seconds that my brand appears during a broadcasted game.”
It’s likely that the answer to this question could not be obtained through traditional measurement, but the time and effort required to obtain a true answer would be worth it in terms of profit.
Data, data, data!
Once it becomes clear to me, as a CMO, that a problem necessitates Machine Learning in order to solve it, I know that the first step is to evaluate, as an organisation, the degree of data maturity, that is:
- What data is currently available, and what data would it be useful to have?
- What data silos already exist within the organisation?
- Using the available data can we easily determine relevant KPIs in order to measure the results of marketing actions (campaigns, ads, etc.)
- And, most importantly, based on this data how much can we understand about the client or customer?
Data Driven Maturity Model
When it comes to data maturity, there are five stages of evolution:
- Data blind: nothing is visualised beyond the data that interests each department, either because it has not been properly collected, or because there is no real strategy behind its analysis.
- Data reactive: the data that has been collected aligns with business needs; it’s usually used by independent silos focus on reporting for decision making. Although this data is collected in aggregates, it is not analysed together.
- Data active: data is being used strategically; data lakes are built for the purpose of aggregating and analysing data together.
- Data proactive: before it is actually needed, the need for data is identified, and the team understands how it will be analyse and obtained, and what knowledge will be obtained from it
- Data driven: data is now an integral part of the value chain; the relationship between brand and client is automatically improved and built upon (i.e. if the data suggests the client is thinking of ending the relationship, retention actions are activated).
Your guide to a machine learning based project
Let’s examine Machine Learning within the context of a project at the intersection of business, technology and mathematics.
Machine Learning project execution framework
The key to successful marketing is teams who ‘speak’ the language of each of these three areas, and understand how to properly connect them by breaking down departmental silos. At Good Rebels, by combining our experiences in the execution of technology and data based projects, we have created a Machine Learning project journey that allows us to:
- Execute only the projects or project phases that really add value, without incurring unnecessary costs
- Maximise the success and usefulness of the project
Machine Learning project execution path
Tasks & Comments
Choose a question which stems from a real business problem – avoid ‘dummy’ projects for the purpose of experimentation. The resources needed to carry out these projects require a real use case which originates from a real business need. First analyse the potential impact of solving this need:
Once you’ve identified the business need, figure out:
Data Driven Maturity
Proof Of Concept
Go / No Go
Remember, the Continuous Evolution phase may involve the start of new Machine Learning projects. For example:
- Sometimes finding the answer to your problems, just generates more questions; the business need may increase in complexity
- The appearance of new data sources of new types of data may necessitate a change to the model or the start of a new project
- There is a constant increase in the number of Machine Learning tools on the market – each one opening up new possibilities in the realm of Machine Learning marketing
- These solutions make it possible to avoid the complexities of an in-house project, and help teams to improve their strategy with less investment
Keep it simple
Let’s return to the first question we asked ourselves: how can we use Machine Learning within the marketing department to achieve our goals? Ultimately, the key to a successful Machine Learning project is people. Machine Learning helps humans to solve complex marketing problems, using business and customer data to identify patterns and hidden knowledge; but it is people who identify those problems which need solving, who understand the context and can identify and place emphasis on relevant data. They’re the ones who’ll interpret and analyse the results based on their own specific knowledge area.
If you’re looking to develop a Machine Learning project within your marketing department, keep these three simple things in mind:
- Start with a good question
- Focus on a real business need
- Assess potential internal and external impact of resolving this problem
- There’s no need to reinvent the wheel
- There are already efficient and effective Machine Learning tools available
- A good understanding of these tools is fundamental
- People are the key
- If you don’t have any internal expert, collaborate with partners who can provide a more complete vision – business, technology, mathematics