NASA - night lights over the US

Analytics and artificial intelligence (AI) are two of the hottest topics in the technology industry today.

Unfortunately, they are also two of the most widely misunderstood and misapplied topics. 

From performing the job of radiologists to redefining the world of art, it's widely assumed AI in particular will be a game-changer for almost every endeavor in the years to come, if it isn’t already. The challenge is such predictions are largely aspirational and do not provide a path for the technology to be deployed, particularly in a modern business environment.

At their core, analytics and AI are means of mobilizing your organization’s data. They involve plugging the data sets that you have so meticulously collected and labelled into mathematical models — frequently simple ones that have been used in statistics since time immemorial — to derive insights. 

We often forget that none of the achievements we expect analytics and AI to deliver would be possible if we didn’t have large, reliable data sets with which to train the models.

Note: This post is the second installment of a two-part series. Read part one: “The First Steps of Data Mobilization: Data Audit and Data Consolidation

Deploying Analytics

Let’s take a simple example. Perhaps your company is experiencing a higher rate of customer churn than is common for your industry and you’re looking for a data-driven approach to counter the trend. Open-source tools like Python or R, or proprietary software like SAS or the Analysis ToolPak for Excel, offer a way to leverage statistical analysis to explore a wide variety of metrics (your independent variables) that may be correlated to customer churn (your dependent variable). Which products are customers using? Are they using all features within those products or just some? Which services are they using? How many support tickets has a customer filed in the past 12 months? On average, how long did it take for those problems to be resolved?

The answers to questions like those will form the building blocks of a customer health score that speaks to customer satisfaction and risk of churn. Running a series of regression analyses comparing the metrics of newer customers to long-standing and former customers will flesh out your customer health score by determining which metrics have a statistically significant correlation with churn. For example, you may find out customers who only use a few of your product’s features are more likely to flip to a competitor, whereas customers who attend one of your events are half as likely to churn when all other factors are held constant. The goal is to provide prescriptive recommendations to your executives regarding what can be done to better retain customers.

In the example we are discussing here, part of the solution will likely involve your customer success team (or whoever owns the customer relationship) doing more to actively educate customers about all of the features your product offers and encourage them to attend your events (you may even find it’s worth comping the ticket).

Similarly, you can apply this framework to assess the likelihood of closing new customer deals based on variables historically correlated with success. You might look at a prospect’s industry; whether the prospect has considered any of your competitors’ products — and if so, which ones; the customer’s level of online/offline engagement and perhaps even the sales rep managing the deal. Depending on your company’s goals and how strategic the deal is, you may pursue a more automated approach for deals that are likely to close and devote more resources to strategic accounts with lower probability profiles.

Analytics holds similar promise for better engaging with your “internal customers” — employees. The difference here is simply one of style, not of substance.

Consider the process of onboarding new hires. How do you know if they are ramping up successfully? You can certainly speak to their managers, but that’s not a scalable option and some managers may not have answers, particularly if they oversee large teams or don’t work directly with new employees on a regular basis.

Instead, you can assess how new hires may navigate the onboarding process using the same analytical approach you used to assess the likelihood that prospects would become customers. Do people with certain profiles (e.g., type of undergraduate degree) tend to struggle with the technical aspects of onboarding? Should the pace of training be modulated based on an employee’s function or location? Do lateral hires do a good job of handling certain responsibilities, such as regulatory requirements, but struggle when they have to acclimate to new sales processes and policies?

By answering those questions, you gain more insight into what individual new hires may need help with and how you can get them to ramp up efficiently and effectively.

Related Article: What Data Will You Feed Your Artificial Intelligence?

The Promise of AI

It can sometimes be hard to delineate where analytics ends and AI begins. The one is a precursor to the other and the terms are occasionally used interchangeably. However, the fundamental difference between the two boils down to how much work your analytics team has to do to reach prescriptive conclusions. Returning to the issue of customer churn, an advanced AI system would be able to test a series of variables for statistical correlation to customer churn without those regression models needing to be defined manually.

Of course, every customer is unique. Where AI can truly be a time-saver is in the ability of AI-driven systems to not only flag customer accounts at risk of churn and notify the account owner but also recommend specific courses of action tailored to individual customers. For example, the system could flag certain customers as being at risk because they have never taken training courses centered around the features most often used by similar customers, and then it could recommend that account managers send those customers specially chosen, relevant training materials for their use cases and industries.

A human employee could also do that, but the process would take much longer and it would be less scalable. The key difference between a human and an AI-driven system is the amount of data and relationships they can review autonomously.

In the same way, AI becomes more powerful as a tool for the human resources team to use in dealing with internal customers as employees develop track records within your company. For example, an AI system can automatically review manager feedback and other performance data to find areas where employees excel and areas where there’s room for improvement. Once weaknesses are identified (or potential challenges are anticipated after a promotion, for example), the system can automatically send employees relevant content designed to help them improve in specific areas and reach their full potential.

Related Article: Data Ingestion Best Practices

A Journey, Not a Destination

Analytics and AI hold enormous potential for helping organizations of all kinds make better use of their data. But that will only happen after organizations complete the painful work of collecting, tagging and cleansing their data.

That’s all part of keeping expectations realistic. Businesses have high hopes that AI in particular will improve productivity and generate business results, but the technology isn’t a magic bullet. No algorithm can work off bad or incomplete data, and organizations that don’t sow and nurture their data won’t be able to reap the rewards. Even then, analytics and AI are a journey, not a destination. Data will change, algorithms will need to be reworked, and humans will still be the ones who have to make the final decisions — albeit with the help of machine-derived recommendations.