AI for Business: A (non-technical) introduction

these days it seems like everyone is talking about AI with new Innovations seemingly coming out every single week however if you're a professional an entrepreneur or business operator you might be thinking to yourself what is this AI thing and how can I use it to drive value in my business so in this video I'm going to share a non-technical introduction to Ai and machine learning and share how it can fit into how we do business and if you're new here welcome I'm sha I'm a data scientist turned entrepreneur and if you enjoy this content please

consider subscribing as a great noost way you can support me in all the videos that I make so in this video I'm going to be talking about two main things in part one I'm going to be answering the question what is AI and along the way defining some key terms such as artificial intelligence which is what AI stands for we'll be talking about models and why those are important for AI and then finally we'll be talking about machine learning so once we have a good understanding of what all these terms mean we can turn to

part two and figure figure out how we can actually use these Technologies I'll give a concrete example of what AI might look like in practice then I'll share five rules of thumb that I like to use when thinking about how and when to use AI in practice starting from the top what is AI so when you hear the term AI you might think chat GPT or AI generated art or you might think of the Terminator or something similar but if we just take a step back AI stands for artificial intelligence so we've got two words

here and one of them is a bit more problematic than the other the first word artificial is not the issue artificial simply means something that is made by humans however the second term intelligence isn't so well defined and even today there's not really a consensus of what this word actually means however a definition that I like to use and one that I think is relevant in a business context is intelligence is the ability to solve problems and make decisions based on this operational definition AI or artificial intelligence is simply a computer's ability to solve problems

and make decisions and so to get a better sense of what we mean by intelligence let's see it in action so let's say we wake up Saturday morning and we're trying to figure out what we want to do today and we look out the window and see this so if we see this and we're trying to decide between a pool day or a Netflix day I think most people would pick the Netflix day because the dark clouds in the sky is probably a good indication that the weather's not going to be so great another example

is if we see this sales data with this peak in November and someone asks us what caused the peak we might reasonably say that oh was probably because of Black Friday which is one of the biggest retail days of the entire year and then finally if we see this text exchange of someone saying fine do what you want the other person responding are you okay and then that original person saying yeah whatever and if we ask are they really fine most people probably say no know even though the person is saying that they're fine their

choice of words like fine do what you want and their use of whatever is probably indicating that they are actually not fine each of these scenarios is a situation where we used our intelligence to make a decision or to solve a problem so even though each of these examples is very different there's a Common Thread that runs through each of them which is intelligence requires knowing how the world works but of course the world is a mass place and it's very complicated so the way we make sense of this huge and complicated world is through

models and a model is simply a approximation of a real world thing that can fit into our heads and more specifically models allow us to make predictions for example when we saw the dark cloudy sky that information went into our mental model of the world and allowed us to make the prediction that it's probably going to rain later however models aren't only restricted to the ones that we have in our heads we can also have computer models and in fact essentially all weather forecasts are done by computer models instead of your weatherman standing outside for

5 minutes and making a forast for the day so models be they mental models or computer models are an essential part of intelligence but a natural question here is where do these models come from so there are two types of models that I'm going to talk about the first I'll call principle driven models which are based on a set of rules these are things you might read in a book or learn from your grandma the other kind of model is based on past examples a principle driven model would say that if we see dark clouds

in the sky then it's probably going to rain later while a datadriven model would say the sky is similar to other times when it rained and so each of these models comes to the same conclusion that it's going to rain but they are built on top of a different different Foundation but of course each of these models isn't restricted to something we hold up in our heads but these are things we can program into computers so principle driven models we can explicitly program computers to execute using standard programming techniques but more recently we've seen the

rise of datadriven techniques to derive models the most popular of which is called machine learning machine learning is potentially another one of those buzzwords you may have heard around but it's a really simple concept machine learning is just a computer's ability to learn by example so the way this works is we have a set of training data which consists of predictors and targets where targets are the things that we're trying to predict like if it's going to rain or not and predictors are all the information that we're going to use in order to estimate the

target the key Point here is instead of explicitly telling the computer how to take predictor to estimate the target machine learning allows the computer to figure out the relationship between predictors and targets simply by seeing many different examples of the two so the way that works is we pass this training data into a machine learning algorithm and out poops our machine learning model with this machine learning model in hand what we can do is get new data pass it into the model and obtain a prediction which is exactly what we did when we saw the

dark cloudy Sky we looked out the window we received some information and we were able to make a prediction that it's going to rain later and a machine learning model Works in exactly the same way so up until this point we've talked about three things we talked about artificial intelligence which we Define simply as a computer's ability to solve problems and make decisions we also talked about models which were a essential part of intelligence because they allow us to understand how the world works works and then finally here we talked about machine learning which is

a way a computer can generate a model based on past examples with these three terms defined we can move on to the second part of the talk which is how do we use these things how do we use these Technologies like Ai and machine learning to drive value in our businesses so I'll start with a concrete example and talk about credit decisioning which is something I have some real world experience with so I can talk about it some somewhat intelligibly so when we're talking about credit decisioning what we're talking about is people applying for a

loan and financial service providers evaluating that application and making a decision of whether to approve the loan or deny the loan so the way that works is someone submits an application for a loan and the financial services company makes a decision of whether to approve deny in the terms of the agreement so the traditional way of doing this is that the application goes to an underwriter which is a person who makes the decision and defines the terms of the contract however now that we've learned about Ai and machine learning we might think oh we can

just replace the human underwriter with an AI underwriter right and the answer to this question is yes and no while it might be easy to imagine replacing a human job with an AI the reality is a bit more complicated so what this looks like in practice is something like this with all these steps within the blackbox being our AI underwriter and really what it is is not just a machine learning model but rather a large number of business rules data and it processes all working together to take the application and finally make a credit granting

decision so although machine learning is a critical part of this AI underwriter here it is only a component in a much broader solution and so this is often the reality of what AI looks like in practice although from an outside view it might look as simple as we have an AI underwriter in reality what's going on under the hood is a bit more complicated which is an important point to keep in mind when trying to implement AI Solutions in your business so while this might be an illustrative example it may not give us a good

idea of how and when to use AI to solve business problems so for that I'm going to talk about five rules of thumb that I like to use when thinking about how and when to use AI in practice so the first one is to focus on problems not Technologies next is to apply AI to problems that you solve repeatedly next we have look for problems in which a 70% solution is good enough to generate value the fourth is pick situations in which one failure of your AI solution doesn't erase nine successes and then finally start

simple fast and easy and build sophistication as needed so I'm going to talk through each of these rules of THB one by one and share concrete examples of each first focus on problems not Technologies so this brings up what in data science we might call the hammer problem which is when you have a really nice Hammer everything looks like a nail so let's say you have some problem in your business something is broken if you take a technology first approach you might grab your hammer and say I got this which obviously is not going to

solve the problem and is probably going to make things a lot worse so an example of this from my personal experience is something I saw over and over again which was last year I had a lot of clients reaching out to me asking for help in building a custom chatbot or fine-tuning a chatbot for a particular use case and this was a classic example of the hammer problem because often what had happened was the client had seen the power of chat GPT and saw all the incredible innovations that have been happening in the space of

natural language processing and large language models and was probably thinking something like I need one of these for my business however the Trap that you fall into with the technology first approach is that you can spend a lot of time and money building a solution for a problem that isn't very critical to your business and essentially this time and money is wasted however let's flip things around instead of starting with the technology what was starting with the problem look like so let's say we have a problem where our customer support line is overwhelmed well from

here instead of jumping into building something you jump into problem solving because when you have a tool your instinct is to build but when you have a problem your instinct is to solve the problem so you might ask why are people calling if people are calling for some specific piece of information you can update your FAQs and if that doesn't cut it you can improve call routing to make sure that callers are getting sent to the right person and there isn't time wasted where customer support representatives are on the phone with someone just to transfer

them to someone else and then maybe after exploring a few Solutions then you start thinking about building a chatot for your website but the key Point here is that when you start with a problem you don't jump to building a solution you jump to finding the root cause of the problem so you can find the best solution and ultimately When comparing these two approaches the technology first approach approach on the left and the problem first approach on the right you almost always want to go with the problem first approach because that is almost guaranteed to

generate value in your business while the technology first approach might be intellectually stimulating and exciting is often something that doesn't drive any real value the next rule of thumb is to apply AI to problems you solve repeatedly and the reasoning behind this is that AI is just the continuation of the story of Technology since the beginning of time it's simply a tool to help make our lives easier so the problems that you're solving over and over again are great candidates to apply AI to for a few reason one if you can automate it with AI

you no longer have to spend a lot of your time solving that problem or even if you reduce the amount of effort it takes you to solve that problem by some marginal amount it can still translate to some big gains other reasons are if you're solving a problem repeatedly you likely have a deep understanding of that problem which puts you in a good position to build good solutions to solve it and finally if you're already solving a problem that means you have an existing solution which is a fantastic starting place for building an AI solution

so an example of this is something that I use in my own work which is a literature review assistant so I read a lot of papers about Ai and machine learning for both my content and my Consulting business and often when reading research articles I discover gaps in my my understanding and so this is a problem that I face over and over again I'm reading the paper and then I stumble across a sentence that seems obvious to the authors but is completely not obvious to me so for that I will turn to chat GPT I'll

upload the PDF of the paper ask chat gbt what the paper is about then ask chat gbt specific questions until I have a clear understanding of what's going on so using Chad gbt in this way has significantly sped up how quickly I can read articles because now instead of spending a 30 minute tangent on Google trying to figure out what a particular term means or putting an idea in a larger context Chach PD does a pretty good job of explaining things and adding additional context where needed the next rule of thumb is find situations where

the 70% solution is good enough where this is coming from is a model is simply an approximation of the real world thing no model is ever going to be perfect and there's a famous quote from statis George box which goes all models are wrong some are useful so the key thing is to accept that your model is not going to be perfect but pick the ones that are actually useful to you in whatever problem you're trying to solve and so a good example of this is Spam filtering the way that works is you have a

bunch of spam emails flooding your inbox in this situation even an imperfect model is very valuable because even a 70% reduction in spam emails is very helpful that'll give a thumbs up for from any user another important rule of thumb is ensure that one failure doesn't erase nine successes and essentially what we're talking about here is find the low stakes or low exposure situations an example of this might be using chat gbt as a writing assistant is pretty low stakes if it gives nine good recommendations followed by one bad recommendation for writing it's no big

deal you can just ignore that recommendation and move on with your life however if you're using using Chachi BT to make cancer diagnosis it doesn't matter if it's right nine out of 10 times that one time when it's wrong can have a tremendous negative impact so that is a situation where you probably don't want to use Ai and if you do you have to be very thoughtful about how it's implemented and then the final rule of thumb is to start simple fast and easy and each of these words simple fast and easy is important so

starting simple is important because sophisticated Solutions are fragile and costly they'll cost you a lot of money to build and they have a high likelihood of failure because they are well complicated next you want to build fast because to build good Solutions you'll need to iterate so that means you'll need to try out a lot of different things and if it takes you a long time to do one iteration it's going to take you a long time before you implement a good solution and then finally you want to make it easy so you want the

solution to kind of be on the way and not something way out of the the way for people because if it's hard to access no one's going to use it including you even if you're the one that's implementing the solution so let's look at a specific example let's say we're trying to implement a sales email sequence in our business the way this start simple fast and easy approach would play out is you'll start by writing all these emails by hand so what that looks like in my business is I'll have someone book a discovery call

with me and I'll send them a follow-up email asking them a couple of follow-up questions based on their specific use case that's me doing the process by hand however after doing that for a bit I've naturally developed email templates for responding to someone booking a discovery call and like a post Discovery call email and then maybe another template for following up with people after 40 days or following up with people after 90 days and so on and so forth so over time instead of just writing emails by hand you start to develop templates and then

over time those templates can get loaded into a CRM so you use a CRM tool to automatically fill in these emails with some bit of personalization like including people's names and maybe some other information that they provide but then you can take this one step further and use some kind of large language model or NLP solution to make the emails a bit more personalized so instead of just using a template and just filling in a name you can make the email sound more like a person so all that to say it's good to start here

you know start by just doing things by hand and build toward that sophisticated solution often times when you're a small business you'll find that just doing it by hand or having some templates are more than suitable so for me I have a small Consulting business so I'm spending most of my time here I don't have a CRM but let's say you have like a 10p person business then you might want to be looking at a CRM then let's say you have a larger Enterprise and let's say you're working with hundreds or thousands of clients then

maybe building the AI solution makes sense but the value in taking this simple fast and easy approach is that you don't artific officially just jump to the end you take it step by step and you only move on to the next level of sophistication if the value is there if it makes sense for your business we talked about a lot of different things so just to recap a few key terms we talked about Ai and how it's a computer's ability to solve problems and make decisions we also talked about models and how they help us

make predictions and that they're a necessary part of any AI system and then finally we talked about machine learning which is a datadriven way computers can generate models from past examples as opposed to being programmed explicitly and then we talked about how we can use AI through five different rules of thumb focusing on problems not Technologies applying AI to problems you solve repeatedly seeking problems where the 70% solution is good enough identifying problems where one failure doesn't erase nine successes and taking this simple fast and easy approach to iteratively develop AI Solutions so while this

was a pretty highlevel introduction I hope it gave you some clarity about what AI is and how you can start to use it in your business this is the first video in a larger series on how to use AI in business in future videos I'm going to dive into more the project management side of machine learning and model development so if you have any specific questions or anything specific you'd like to see in future videos of this series please drop those in the comment section below and as always thank you so much for your time

and thanks for watching