Artificial intelligence and machine learning can deliver unprecedented value to the business. Unfortunately, fantastic findings often get lost in translation. To avoid expensive blank stares and stakeholder frustration, data science practitioners also need to master the art of interpreting and explaining results in simple, plain terms business people can understand.
What? So what? Now what?
In my work with numerous organizations implementing cutting-edge machine learning technology over the past two and half years, I have seen one common recurring problem. Countless data science experts assume that they are communicating results successfully when they are not. From expressing metrics in unfamiliar terminology to presenting odd visualizations, today I see a massive gap when it comes to storytelling for AI. My analytics peers that participate in #MakeoverMondays would be horrified.
Accurately interpreting and explaining findings from advanced analytics is becoming a crucial skill to bridge language barriers. In February, Gartner shared new research titled “Fostering Data Literacy and Information as a Second Language: A Gartner Trend Insight Report”. They opened with a call to action for artificial intelligence creators and consumers to "speak data" a common language. McKinsey and other industry leaders are expressing similar concerns.
Machine learning and advanced analytics can be difficult to understand and explain. Describing the problem, the model, the relationships among variables and the findings are often subtle, surprising and technically complex. Effectively translating quantitative insights and telling a compelling story requires planning, compelling design, and visualization choices. Successful analytical communicators don’t wait until the end of the analysis but rather use the entire process as a vehicle to communicate with stakeholders.
To help decision makers get the most value from AI and machine learning projects, don’t overlook your existing business analytics talent. In-house subject matter experts that build executive dashboards and reporting solutions using Excel, Power BI, Tableau, Qlik, and TIBCO Spotfire will most likely know best practice visualization and powerful data storytelling techniques. They are also skilled in translating data into relevant business domain language.
Don’t let egos or pride get in the way. Urge data scientists, data analysts and BI professionals to collaborate and work together on machine learning projects to improve outcomes. All these groups can learn something from one another as you upskill your teams.
Machine learning translation
Interpreting machine learning entails decoding both algorithms and the processes machines use. Until recently, many types of predictive algorithms have been infamous black boxes. Aside from basic regression techniques, machine learning and deep learning processes were far beyond feasible human comprehension. Thus, decision-maker trust and regulatory agency compliance prevented adoption.
When relying on a machine to make high-stakes decisions, it’s crucial to know how a decision is made. You should be able to document and audit machine learning processes. In some industries, such as banking, insurance, and healthcare, it is a regulated requirement to do so.
For explaining or defending a machine’s good decisions and fixing the bad ones, you’ll want to be able to see scored machine learning output for each record combined with the top variable value details in plain business language that most influenced the predicted outcome. For example: Record ID 232333 was predicted to be a high value customer because of size greater than 10,000 employees, monthly spend between $1M and $1.5M, and so on for relevant decision influencing input variables.
To start earning stakeholder trust early on in your machine learning projects, share intermediate reports such as top outcome influencers that can be invaluable to the line of business. Machine learning can rapidly narrow down the scope of potential variables that matter most when faced with hundreds or thousands of variables to analyze.
As you create models, visually share progress and insights on where your model is accurate and where it makes mistakes using scatter plots, combination charts and interactive data visualization tools. This joint exercise with the business can help continuously improve your model input and results.
Another one of my trust building tips is to deploy new machine learning models along with a model accuracy monitoring report side-by-side with existing reports. This will allow the business to contrast and compare any differences in outcomes. You’ll also be able to detect model performance degradation.
Last but not least, I recommend avoiding black box machine learning tools. If you need to explain machine learning results, look for solutions that provide highly human-interpretable model process visualizations. Robust, modern machine learning solutions should be able to provide diagrams of preprocessing steps that each model uses to arrive at its outcomes.
To realize the true potential of AI and machine learning, you will need to involve the business every step of the way. Data science teams should not operate in isolation. Collaborate with the business and with business analytics teams to best translate machine learning results into familiar business metrics. To reduce skepticism and resistance to change, consider delivering machine learning output within existing decision-making apps and reporting tools the business already knows and trusts. By following these suggestions, you should be able to enjoy far more tangible benefits from your machine learning initiatives.