Note: As a standard practice, I will gradually include a plagarism detection link in all the articles on my blog site. You can use the online tool below to evaluate any article published on the web for plagarized content. All you have to do is to paste the url of the article there and it will go through every sentence in the article and flag any plagarism. I believe as a follower of “writer’s integrity,” I should include this tool in all my articles:
Just like some people binge watch their favorite TV series, I binge read (whenever I get a chance). This weekend I got a chance to binge read “Prediction Machines”.
With all the hype around AI, this book bridges the gap between the technical aspects and the business aspects of Artificial Intelligence-so it is focused primarily on economics and strategic aspects of Artificial Intelligence algorithms and technology. I recommend this book to managers and strategists who want to delve deeper into these aspects of the AI revolution, will be leading AI related initiatives and want to start/plan their initiative better. This may also be useful for rookie Data Scientists who are interested in learning more about the impact their work makes on businesses and the world.
Though the entire book is full of useful insights, one aspect that I thought is worth sharing on this blog is the concept of AI Canvas.
What exactly is an AI Canvas?
Every organization wants a piece of the so called data science revolution cake. However, only a few of them are equipped to actually successfully implement useful AI capabilities. The word “useful” here is extremely important. You need a solution that delivers, that actually helps you business. Not a tool that checks a box but does not deliver any value. Don’t do Data Science just because everyone else is doing it.
Careful planning needs to be done prior to starting the AI and Data Sciences journey. However, this post is not about those steps that need to be taken or the infrastructure/architecture that you need to build. We will keep this post focused on one of the tools that can help in the planning stage to layout the strategic framework of an AI solution-The AI Canvas.
The AI Canvas
Before we jump into designing an AI solution/Predicting machine, we need to understand the aspects of the tool like (not exhaustive):
- What is the objective?
- What kind of inputs are required?
- What are the metrics to test the outcome?
I think the AI Canvas can be a good framework to use in the planning, building and assessing process of an AI tool. It can be a very useful tool in the solution design phase of an AI algorithm and can help you structure your development and execution phase better. This tool can also serve as a “backup” slide in your business case pitch to help senior leadership understand the value the AI initiative will deliver.
The key components of the canvas has been explained below with examples for each block of the canvas, from a Supply Chain perspective.
Prediction: This is the prediction objective. Example-predict whether the shipment, which has certain traits in terms of origin, destination, product type, mode, day/month of shipment etc., will be delivered in time.
Input: Based on the prediction objective, what data to you need to run the predictive algorithm. As indicated in the example above, the inputs for predicting on time delivery would be origin, destination, product type, mode etc. It is important to get a handle on the inputs, as this will enable you to plan your data collection/availability accordingly.
Judgement: How will you value different outcomes and errors. You need to determine the value of correct prediction verses false negatives verses false positives etc.
Action: What is the step/decision that you are trying to take that needs the output from the tool? Essentially, what is the action you will take based on the tool? In this example, it could be switching carriers on a particular mode.
Outcome: What is the business objective you are looking to achieve through this analysis? In this case, say the problem is customers are complaining about late deliveries and you are trying to figure out which lanes or modes are your problem so that you can work on that. The outcome desired is improved customer satisfaction.
Training: What data do you need to train the algorithm ? In this example, you want to use the planned outbound shipments data to train the algorithm.
Feedback: How can you use the outcome from the algorithm to make your algorithm perform better? This goes hand in hand with judgement
There are multiple examples of leveraging AI Canvas in the book which will help you understand the tool better. A little bit of practice and you should be able to take a business problem (that a prediction algorithm can help mitigate), and translate the problem into an AI solution using the canvas. This will ensure that you don’t realize midway through the project that you are on the wrong path.