As a data analyst at Mintigo, my job primarily consists of figuring out the best way to draw insights from the massive pile of data that our prospective clients have in their CRM and MAP systems using our AI platform. Another part of my job is educating prospects on how this technology works so they can be successful when implementing it in their own businesses.
Most of the clients we work with are familiar with the concept of AI, along with other related buzzwords like machine learning, regression analysis, and clustering. While most people recognize the promise of these tools, many folks are skeptical that they are the “silver bullet” that AI worshipers may have you believe – and for good reason.
Of course, nothing is that simple! It’s true that the fundamental logic behind many tools under the umbrella of AI are sound; the pillars of AI have been applied to business problems since the 60s with the advent of statistical computing.
The game changer in the past decade has been the emergence of big data that is readily accessible in the business tools that so many companies are already using in their day-to-day activities. We generate far more data today than ever before and that means we have significantly more to feed into our AI-driven models. This also means that we can be more specific and nuanced about the types of questions we can answer with data.
The flip side of this is that there are many more decisions to be made around exactly which data is optimal for a given business goal. When talking with clients this is the point where we do the most education around how to select and pre-process the data from their CRM or MAP to train their predictive models. As it turns out, the hardest part of any statistical analysis lies in selecting the right data so that the analysis is calibrated to the type of success you want to see.
In the past, marketing analysts might look at two to three, or up to a few dozen, types of indicators when they were building models to identify their best prospects. At this scale, it is straightforward to identify the data points that correlate with success. It’s also pretty easy for humans to think about using a handful of data points to make a prediction. However, as the quantity of data increases it becomes incredibly difficult to grasp which of the thousands of attributes about potentially millions of prospects are the most important.
Machines, however, do a great job of dealing with this massive amount of data. The machines simply need humans to tell them what to optimize for, and the way to do this is to give the machines a clear picture of 1) what a successful outcome looks like and 2) the total set of prospects you want to convert to that successful outcome.
For example, if your goal is to help your sales team better focus on the open opportunities that are most likely to close this quarter, you need to build your predictive model in a way such that your successful outcome is a closed-won opportunity and your target list is all of your recently created opportunities.
However, if you are trying to help your demand gen team serve up better leads, ones that are more likely to be marketing qualified or sales accepted, you need to create your model with the successful outcome set as a conversion to MQL or pipeline and your target list should be comprised of all recently created leads.
Both questions are very straightforward to answer simply by pulling reports from your MAP or CRM. But whether the model is going to give you the results that you are looking for is almost totally determined by that very first step of telling your modeling platform exactly what you want to do!
Of course, determining what you want to do within the limitations of the data that is available is easier said than done. In my experience, it often requires that stakeholders from different parts of your organization – marketers, sellers, account managers – align on what their shared goals are. This is often the biggest hurdle to getting off the ground, but once that alignment is there, artificial intelligence driven by machine learning does the heavy lifting.