We all know there are a LOT more than 3 steps to implementing artificial intelligence (AI) or machine learning (ML) in an effective way. This blog post is the first installment in a series focused on AI implementation, and we suspect it will be the least technical of the bunch.
To give a little background, the team here at PrecisionLender wants to provide some information about our journey that will help as you put AI and ML into production within your application/environment. This blog series is one of those outlets of information. If you are looking for “What tools did you use?” or “How did you address the problem of recursive prediction/action loops?” those will come soon! What you will read here, and in future installments, are the real-world lessons we have learned/are learning in the development of our AI Persona.
It is worth clarifying that when you see the term AI here, we're referring to a personification of applied intelligence with a learning and feedback loop. That applied intelligence may be a simple heuristic or a finely tuned neural network, but the AI in this context is the personified manifestation interacting with a human. If you want an idea of what our AI (her name is Andi) looks like now, you can see a little bit of what she does at https://precisionlender.com/platform/.
To successfully implement an AI, there is some necessary up-front work. In this post, we will focus on three things we found helpful to do before writing code or building models. These are tasks you can do ahead of time that will make life easier as you move forward. We initially wanted to write “the most important of those is …” but, in fact, these have all proven to be highly valuable to us, and there is no stack ranking to the presentation order.
1: Understand the job your AI is intended to perform.
We are fond of the “Jobs to be Done” framework, pioneered by Harvard Business School professor Clayton Christensen. You can hear more about that here. The key to the framework is to understand the value you intend for your AI.
You must start by asking the question: What is the customer hiring the AI to do and why?
Knowing the answer to this question helps prioritize the work and makes the AI more consistent. With any effort in AI, one of the issues you will face is trust. We believe that having a consistent AI with a defined job helps people trust it.
2: Personify your AI ASAP.
There is power in personifying your AI, but the personification should not only rest in client communication. It should be a part of internal discussions, as well, and there are a lot of internal dialogues that will need to happen for your AI to get into production. Being able to say, “Andi needs to be able to monitor promises and the impact of broken promises” is simple and easy to understand.
Personification also has the added effect of stirring the imagination. “What will Andi learn next?” is a much more exciting question (to most people) than “what functionality will we build next?” and it can spark the creative interest of customers, as well as developers. Combined with understanding the job, it is easy to ask, “What is the next most important thing for Andi to learn to improve as a pricing analyst?”
3: Find an exemplar(s).
In our experience, it has been very valuable to have a person in mind who exemplifies the goals we currently have for our AI. Someone who is so good at the job to be done that you find yourself asking “what would Joe do in this situation?” There is no substitute for experience and subject matter knowledge. Having a practical and experienced mentor for your AI is a great way to solve a lot of the questions you will encounter and has the added benefit of keeping your development team tied to the customer benefit.
Once you have done the three things above, at the very least you will understand how your AI is going to earn its paycheck and whom the AI will model its immediate development after. From there, you should be able to start addressing some more technical items. More on a few of those next time.
If you have specific questions on this topic, you can email George Neal directly at email@example.com.
About the AuthorFollow on Linkedin More Content by George Neal