The Bots Are Taking Over!
Well, kinda. The bots are trying to make heads or tails of your data. There are mature capabilities such as image, voice, and text recognition, but that’s only the fractional edge of what’s possible. Those types of technologies replace a rote human process. That’s valuable, but the real magic happens when the tech starts to tell you how to improve your overall business, or even does it for you, which is the true power of AI/ML. The potential applications are endlessly exciting, but this month, let’s focus on understanding the journey from learning to intelligence in layman’s terms.
Let’s back into this. The abbreviation “AI/ML” has always seemed reversed to me, as machine learning necessarily precedes artificial intelligence. Humans can’t determine a safe driving speed without intelligence, and that’s impossible until they have learned concepts around traffic density, stopping distance, weather impacts to road surfaces, etc. Artificial intelligence works the exact same way. Teach it the concepts; it can then start putting them together.
First, what business critical metric do you want to impact? Is it NPS? Do you need to improve reserve allocations? Next, what do you suspect drives variance in that metric? Multiple upstream processes inform those outcomes, so think broad, and then talk with your data teams. How are those variable upstream outcomes captured, and where to they live? You’re forming hypotheses that certain processes impact a specific business outcome. The complexity of your project will depend on the number of those that you’d like to test.
A bot’s purpose is to process data repeatedly at lightning speed. They can’t do that job unless they are being fed data they can interpret. So, step one in training them is to ensure your data for those upstream variables you thought of above is well-structured and, if needed, sanitized for any sensitive information it may contain such as PII. After you have workable data, you’ll go through various rounds to ensure that machine is “gets it.” The quality of insights machine learning provides will highlight the quality of your data, and don’t be surprised if data limitations prompt a completely different approach. Once the bots are thoroughly trained, you’ll switch from training data to real production data, and now they can optimize or replace a formerly human-led process.
Your machine has now learned one thing and knows that thing very well, but it still can’t provide business-impacting insights until you correlate what it knows with an overall business driver that’s meaningful to your business, for example, NPS, reserves, etc. So, another workstream with the same steps as above trains different bots to understand this new data. Now, you can start making associations and testing those hypotheses. If I change this variable, what happens to that business-critical metric? In a well-designed program, you’ll get additional correlations that you didn’t predict. In an exceptional program, you’ll be able to trust the AI to make adjustments real-time on your behalf to optimize outcomes.
Business leaders need to understand the basics of this journey now more than ever. The interplay between knowing the data and knowing the business requires IT and business to work together, and you need to be armed with a basic understanding of the process when engaging with potential AI/ML partners. The process is dependent on the quality of your data more than their algorithm, so take the time to invest in internal discovery with your IT partners before you begin the journey.
