Practical lessons when implementing machine learning in marketing projects
Machine learning is a better prediction technology, and with better predictions you can more easily optimize business outcomes. However, for many business users, the idea of implementing this feels unattainable. At G2M Insights we know that leveraging machine learning in marketing is possible for every company, no matter the size or maturity of the business.
We know, because for the past several years the G2M team has been living this day-in and day-out, supporting our clients with building, training, and implementing unique-to-their-business analytic models in areas like propensity modeling and clustering. We’ve learned a few things along the way, and wanted to share some of those lessons with you.
We’ve distilled our learnings down to four key tenets, that together provide the foundation for our approach to machine learning analytics.
Tenet #1: Predictive modeling should be tailored to your business needs
One of the most common problems in any business is that the employees who own the data often don’t understand the business needs of the employees who leverage this data to do their job. Getting to a meaningful result can look like the children’s game of telephone where the output doesn’t satisfy the initial request, or even make sense from a business perspective. While data scientists and technology tools can provide the heavy lifting, end users need to be involved in variable selection and model build process to ensure the model output makes sense against the objective.
Our take: Be deliberate in selecting the outcomes you are predicting; just because a system predicts something doesn’t make it useful or actionable.
Tenet #2: Models should be transparent and accessible to the end user
This may seem straight forward, but many teams who indicate they are using artificial intelligence are in fact relying on black box models that exist within technology platforms. These platforms ingest whatever data you have, regardless of veracity or cleanliness or bias, and are formulated for a single dependent variable, whether that is relevant to the business outcome you are looking to achieve or not. These models are often touted as proprietary and therefore, the variable selection process, predictive nature of these variables, and the algorithms used are not exposed or explained to the end user.
Our take: Avoid pre-built models whenever possible. If you can’t clearly explain to an executive why a model is giving you a particular answer, you shouldn’t be using that model to make important business decisions.
Tenet #3: Empower users to focus on variables and outcomes, not technology
For machine learning to be useful, the output needs to be available in the tools that teams use today. Building a model in a Jupyter notebook is all well and good, but how does this information help when customer-facing teams don’t have access? This is where reverse ETL, the process of sending data back to native systems, like CRM or Automation, becomes very powerful. If your Sales, Marketing and Customer teams are using Salesforce, then the data needs to be in Salesforce for them to access (Salesforce is just an example — insert any other CRM, automation, visualization tool, etc. here).
Our take: Don’t force teams to access new tools or go hunting for data. Make machine learning outcomes available where business is being done today.
Tenet #4: Modern analytics requires a modern tech stack
Machine learning is 95% infrastructure and 5% ML code (for more on that see this post). The data tools that enable teams to put models into production are still nascent for most companies, and even more so for any company more than ten years old. If you are reading this thinking, “my IT department is way behind the times…we are doomed,” know that not all hope is lost; a new category of tools is emerging that handles the engineering infrastructure to bring customized machine learning models to business end users. Stay tuned for more on this…
Our take: Advances in technology are fast moving; business adoption is not. Expect to see companies move in to fill the infrastructure gap that plagues so many businesses around building business-centric, production-grade machine learning models.
Machine learning provides better predictions to support companies in optimizing business outcomes, but relying on pre-built models has many pitfalls, and putting in place the right technology resources to scale production and expose the data can be a real challenge. Don’t let these speed bumps hold you back. G2M enables businesses just like yours to implement customized, business-centric machine learning.
How can we help?
Do you need better predictive analytics? Want to learn more? Feel free to check us out at https://analyzr.ai or contact us below!