Machine learning and AI research are making strides daily, with interesting implications for business. While it feels like a lot of the cool innovations are happening only in the scientific fields, progress can be seen in the marketing discipline too. Below are some areas for marketers to explore to improve their processes.
Increased Accessibility to MLOp Processes
Open source machine learning frameworks have given rise to machine learning ops (MLOps), a deep learning-influenced variation of devops that involves data modeling.
The concepts may sound overly technical for some marketers, but the growth has generated plenty of tools and solutions that make it easier for non-techie marketers to assist with the model workflow. Marketers unfamiliar with GitHub, for example, can still help with capturing model preparation steps in GitHub Projects, a project manager feature in GitHub repositories.
What This Means: Marketers should seek opportunities to apply their business acumen without taxing their skills or being drawn into programming decisions that should reside with data engineers or managers.
Improved Capabilities in Integrated Analytics Solutions
Not every business relies on open-source solutions for analysis. Packaged solutions such as Microsoft Power BI, Tableau and Google Data Studio are also frequently used to help drive data-related decisions. These usually have a user-friendly UIs that make adding plugins and dependencies easy.
Microsoft, for example, offers a premium option in Power BI to access Azure Cognitive Services, which introduces different algorithms such as sentiment analysis, language detection and image tagging to your workflows.
Dependency options can enrich existing data with data from different sources, which can help when planning machine learning models for training. The end result is a more sophisticated workflow for data preparation, allowing faster preparation for specialized data models.
What This Means: Marketers should examine how their tools blend data and simplify the data workflow. Seek out solutions that provide the best integration for repeated processes. Expect more options of workflow integration as well, as some solutions add documentation or project management integration feature to ease collaboration among teams.
Related Article: How Pipelines Can Help Marketers Gain a Better Understanding of Machine Learning
Know the Fine Points of the Great Data Debates
Marketers should be aware of ongoing discussions and debates regarding the kinds of modeling data preparation techniques used in machine learning models. Many different frameworks and techniques have found new applications, so much so that even experienced data scientists are starting to join in debates.
Last summer well-known Stanford University professor for machine learning, Andrew Ng, wrote an interesting post about the evolving roll of data and code in model development. Overall, he felt professionals should reemphasize data over code to develop properly performing machine learning models, noting that for years code was emphasized over data for modeling changes.
What This Means: Marketers should include news about machine learning practices and concepts in their daily resources. These debates may feel strictly academic, but emerging approaches can have downstream consequences on business ML models. One example of such a debate includes a broader awareness of the implications and use of facial recognition on people of color (which I wrote about here).
Learn How to Apply Old Techniques for New Uses With Machine Learning
To further Professor Ng's point, data scientists are discovering how to apply older research concepts to new datasets associated with real-world concerns, be it in business or in the life sciences. Data has increasingly incorporated real-world features, such as geolocation. Moreover, cloud solutions have opened up new capabilities for what can be done with data, from opening data access to coordinating shared projects across teams in various locations.
In the process, data modeler have learned how to apply concepts that were previous strictly academic into real-world business applications. Feature engineering started to include more data based on product features. A machine learning performance concept, the Shapley value, has existed since the 1950s. Developers have found a new use for it through the volume of data involved in customer approval processes, such as home buyers being approved for a mortgage.
What This Means: Marketers should look for opportunities where context is applied to data. Old techniques can lead to creative ideas of how to apply machine learning solutions to business problems. Look for a variety of model concepts that focuses on output refinement, such as General Adversarial Networks (GAN) which produce verifications for how a model differentiates between different kinds of images.
Related Article: Cutting Through AI Marketing Hype: It's About Machine Learning
Keep the Supply Chain in Mind
The holiday retail season this year has been overshadowed with supply chain shipping delays. A variety of products ranging from consumer goods to cars have been impacted. During 2020 businesses quickly slapped together their best machine learning processes to predict supply chain flow. 2022 will likely see a refinement of those processes and features available for predictive modeling.
What This Means: Marketers should stay informed on the impact of supply chain management on product and service delivery. Use the knowledge to decide timing on marketing campaigns and messaging.
Increased Exploration of Price Optimization
In part due to the supply chain issues, pricing has shot up on a number of different products and goods as well. Expect to see interest in pricing optimization to therefore rise in 2022 to better match customer demand with potential revenue. The latest machine learning frameworks can be applied to pricing data to discover correlations and potential optimization opportunities. This is particularly helpful when the pricing of the product is extremely volatile. It is also helpful for businesses that operate in industries with high customer churn rates. Base risk models on optimization choices meant to avert customer churn.
What This Means: Marketers should explore how to develop the best optimizations in their industry, relying on the techniques emerging from open source machine learning frameworks.
Machine Learning in Digital Marketing Media
A plethora of analytics and digital ad solutions now incorporate machine learning algorithms as a feature of the platform. In 2022, marketers will continue to rely on platforms where some degree of predictive analytics is needed for bidding optimization.
What This Means: Marketers should be ready to decide when and where to introduce human intervention in programmatic ad campaigns. Some concerns have been raised in this area, but new features will keep marketers questioning when messages overreach in their digital marketing campaigns.