By Guillaume Payen, Head of Data & GenAI Architecture - WENVISION, France
I’ve always been haunted by the number of Data Science projects that fail: 80%. This is enormous, and I’ve always wondered what could cause such an amount of failures. Data Science and AI projects are still globally IT projects, they still show a certain level of pragmatism, and IT projects are usually very structured, so what could explain that 80% projects fail?
Just Saying!
A couple of years ago, in a company where I had worked as a Data Engineer, a stakeholder shared with me the secret of success for Data Science and AI projects and transformations. He said: “You know, people believe these projects are about IT, technology, implementation and architecture, but they are wrong. What makes the success of these endeavors is the human factor. IT takes its part of course, but success is not predicated on how the program will be implemented, or where it will be deployed. It is about how people relate to it, how they understand the value it can bring, and most importantly, how the team is structured to have people coming from different backgrounds work together”. I was young, and I admit what he said didn’t really make any sense to me, because I was into engineering, and I couldn’t fathom a bond between IT transformation and people.
Transformations That Fail, Others That Succeed
I had the opportunity to work on many Data Science and AI projects, for many years. I saw AI projects led in different teams, in different companies to answer different use cases. I saw most of them fail, and after analyzing the post-mortems of these projects and transformations, I figured out that they all had one thing in common. In these transformations, the organizational and human dimension was always underestimated (if not discarded).
A great focus was put on technology, fine-tuning the models, and having the most reliable and robust infrastructure possible. However, in the end the projects failed because there was no one to use them, no one to provide with relevant feedback, no one to support them. Ultimately, when the project managers reached out to the stakeholders to ask for more budget to prolong the transformation plan, in order to launch a version 2 or to answer the misalignment with the business, the answer was irremediably the same: “we understand that this plan didn’t meet its market, maybe it would be better to put a stop to it, or maybe operate a reboot”.
The same way, I also had the chance to see AI projects and Data transformations that succeeded, and all of them also had one thing in common. In these endeavors, the teams were carefully managed, and a lot of effort was put into giving sense to people. The team members knew what their part was, what was expected from them, and a strong collective was created. Outside of the team, the way the plans unfolded was that they were oriented towards people, with a lot of communication, presentations, and feedbacks. Reviews with quizzes were organized at least twice a month.
After the projects were released, or the transformations delivered, very few of them were executed perfectly. There were a lot of optimizations to take into account, planned for a next version. However, the stakeholders and the potential users knew and understood what was done, how they could get value from what was deployed, and above all they were feeling like they were belonging to the project. These projects were successful, and there was no debate about asking the stakeholders for another budget to launch a version 2. Many times, the stakeholders reached out to the team and proposed their help (in terms of resources or finances) even before the transformation was done.
The Morale Of The Story
With the experience of the many projects I had the chance to work on, I now understand that the success of AI transformation depends less on what you do than how you do it. Of course it is important to consider your technical infrastructure, what you will deploy, what models you will choose and how you will orchestrate everything. However to guarantee the success of your AI transformation, these aspects are necessary but not sufficient. To have people adopting this transformation and living by those new processes, rituals and new ways of working, you must also focus on the people. Focusing on the people takes two aspects.
At the team level, an AI transformation is very challenging, engineers will encounter many obstacles. Therefore, they must be put in a position to succeed. To foster this, one must develop trust and communication within the team. This is crucial to develop a team spirit, and remember: your team is your engine, they will implement this transformation, they will maintain it, they will deploy it. This is why it is so important to dedicate a significant amount of time to give them purpose, to help them understand where you want to lead them, and why they will walk this path.
As a leader, you’ll create a dynamic within the team so that the team members will follow you, because they trust you, because they understand what their mission is, and how they will succeed. This task is necessary and will demand a lot of energy and emotional intelligence. The least thing you’ll want will be to:
Create frustration or competition within your own team. You want to avoid this at all cost, because while they will be busy messing around, they won’t deliver. Emphasize on how the team in charge of the AI transformation will work together, and be united to face adversities, rather than pushing them to deliver the most perfect and beautiful product ever (this will just feed your ego).
Underestimating the organizational component of AI transformation will be a mistake that could kill the entire project. There may be situations where AI transformation involves only a team or a department, but most of times, the impact of an AI transformation shines on all the company. There will inexorably be a moment where you’ll need feedback, validation, support or budget. No one will support a transformation they don’t understand, no one will give budget to a transformation they were not involved into. Thus, having the board with you will be crucial. Your product or your transformation actually doesn’t need to be perfect.
A lot of transformation managers fall into this trap because they believe that the success of their transformation will be evaluated on the perfection of what will be delivered, whereas the evaluation of the transformation will always be done on its adoption by the users. This is why it is also crucial to show great skills of communication, to allow people to appropriate your project, to relate to it. This will create envy and sympathy. People will be glad to test the first output of your AI transformation, even if your product or your new processes are not perfect. Stakeholders if onboarded can also propose support, in terms of visibility, or fundings. To make your transformation plan a success, people must believe in it.
Conclusion
AI transformation is a very subtle endeavor. It is usually led by people with a strong technical background. Therefore, naturally, the force is put on the technical side of the transformation. This makes sense, obviously there couldn’t be any AI transformation without a technical transformation. However, AI transformation is also about having people, users and stakeholders adopt everything that this transformation embodies.
The answer to this requirement is unfortunately not technical. Instead, solutions are found in team management, leadership, communication and inclusivity. In the end, AI transformation is an organizational transformation before it is a technical transformation. To make such an endeavor successful, the two sides of the coin have to be considered and handled with the same care. It is the people who will give you feedback, who will be the users of this transformation, and if your transformation is a success, they will candidly reach out to you and say “that was cool, now what’s next?”
