Which analytics approaches to manage an Artificial Intelligence project? Quick guide for newbies.

Here we are! The new star AI project of your company is in the starting block. It is a big stake in your digital journey. Everyone is looking forward to seeing AI magic at work.

All the principles behind this technological field no longer hold any secrets for you. You know everything about how AI will transform your business. You understand what gradient descent is in deep learning models. You are ready for some amazing adventures. But when the first questions arise and time comes for strategic decisions, you are not so sure about your choices.

Keep calm and take a deep breath. This quick reading will help you to better understand what could happen regarding the decisions you will make. And how you can structure your approach to feel more confident on a daily basis.

First of all, you have to take time to find the right problem. It’s not as easy as you think. Some pointers you can use at this stage:

• What people say is not always what they need (Did you know the 5 whys method?)
• AI doesn’t do what you want, but what you ask (Did you ask for the right thing?)
• Process can be changed (Think outside the box!)

And if it’s not enough, Design Thinking can help to define exactly the topic you have to tackle with.

The problem is clear. What about the solution? We will bring your project through 4 analytics approaches:

• Rules based System
• Analytics
• Machine Learning & Deep Learning
• Multi Agent System & Reinforcement Learning

Manage your AI project with a rules based system? Data is on board! but… you have muted it.

Rules based system (or Expert System) is a determinist approach. The core system is built thanks to the knowledge of a business expert. He or she designs each step of the algorithm. It’s a decision tree in most of the cases. So the job of the Data Analyst is to translate the business rules in code. He doesn’t extract insights from data. He works as a “Data Developer” and uses only some IT basics like SQL.

It’s a very efficient quick-win and / or a good way to define a baseline model to evaluate more complex approaches later. It can be a good approach for so-called static topics that will not evolve over time. It’s also an opportunity to learn on the business side of the problem. But if the business case is complex or wide, you can reach the limit of the approach very quickly. The more you want to generalize, the more you need business experts and developers to support the rules. That can bring you to a point where new rules don’t improve your model due to the lack of control and overlap of these.

Manage your AI project with Analytics? Data is the co-driver in your project.

This approach is close to the first one. But you introduce some insights from Data in your model. The Business expert doesn’t decide on his own on the rules and the algorithm. And there is a peer-to-peer relationship between him and the data analyst. He brings his business knowledge and the data analyst comes with data insights and mathematical models. That helps to structure, to generalize the approach and to scale the solution.

It’s a good compromise between business and data. It allows to align both sides around common metrics and KPI. Your solution is still explainable. So it’s a way to build user confidence and start to change the mindset to more data-driven decisions. However the use of basic statistical models (correlation, decision tree, distribution analysis, linear regression, etc.) can be a limitation if you have to crunch lots of data or if the problem is more complex than a linear model.

Manage your AI project with Machine Learning & Deep Learning? You put some high-tech options in your project (like gps, speed controller, automatic lights on, etc.)

Revolution starts here! Paradigms shift. The algorithm is designed by a data-scientist. A third party enters in the development cycle: “the machine” autotunes the model parameters. But business experts don’t disappear at all. They feed the data-scientist with insights and feedbacks regarding the results of the model. Most people will use the solution not because they understand it, but because it just works.

Machine Learning and Deep Learning give the opportunity to automate more complex tasks with a high level of accuracy. However this quest is done at the expense of explainability. For the business part, it’s a black box. And to be sure it’s not the case for your data experts, you need strong mathematical skills to master the models. IT skills and a good infrastructure too is required to put it in production and monitor it correctly. At this step everything runs automatically, so you need to master all the data value chain: From the KPI and metrics definition to the data governance. As if that wasn’t enough, you need to think about ethics and change management because your algorithm does tasks that were done before by a human.

Manage your AI project with Multi-Agent System & Reinforcement Learning? You are in a self-driving solution!

We enter here in another dimension. Paradigms shift again. We can start talking about Artificial Intelligence systems. “The machine” doesn’t only auto tune the parameters. It makes decisions. It evaluates and it adjusts them. Business experts and data-scientists have to find the right balance between efficiency (the system uses what he knows to get the best results he already got in the past) and exploration (the system tries something new to improve his efficiency).

When Machine Learning and Deep Learning learn only from the past, Reinforcement Learning is able to create new options and work in a new environment. It is an autonomous system and it could be a real personal assistant. But are you ready to let a machine push the button for you ? In addition to the consequences of a potential bad decision, you have to deal with the ethical issue. Automation of job, responsibility between human and machine, etc. are now at the heart of your project management.

As every AI problem is unique, managing a project is not easy. You will meet lots of challenges. And I hope these frameworks can help you to have some landmarks, to choose the right approach and to lead your project properly.

It’s unlikely that your project will follow exactly one vertical line. Unfortunately. But it is what makes your job exciting. Your role as an AI project manager is to identify your strengths and weaknesses. You have to maintain the right balance between every part involved in your project. And your mission is to bring it at the right level that guarantees the objectives, controls your budget and makes your team proud to work on it. Artificial Intelligence is a trendy topic with high levels of expectation. Structuring steps and having clear interim objectives are a means to escape from the moon shoot effect : Big Project with big ambition but no usable results after many years.

Thanks for reading!

🙏🏼 If you enjoyed the article, please consider giving it a few 👏👏👏.

💌 Learn more about tech products opened to the community and subscribe to our newsletter on http://developers.decathlon.com

👩‍💻? 👨‍💻? Interested in joining Decathlon Tech Team? Check out https://developers.decathlon.com/careers