Seth Dobrin, vice president and chief data scientist, IBM has a lot of experience in leading and executing artificial intelligence and data science projects; he has been doing this for the last seven years. Furthermore, he has been leading the IBM Data Science Elite Team of 60 people since 2017.
The IBM Data Science Elite Team is a group of data science and artificial intelligence experts who work with companies to help them kick-start or build an AI minimum viable product.
With this wealth of experience, Dobrin can easily recognise what are the key features of a successful AI project. These are the use of an agile methodology, a hub and spoke configuration, a focus on teamwork as well as a focus on the end-user.
“Nothing should be in an ivory tower.”
The benefit of agile working is that allows the teams to have the flexibility and the ability to choose their own tools however there does need to be oversight from the executive. “Some of these tools can get you into trouble so there need to be guard rails that are set by the central organisation in conjunction with the business so nothing is in an ivory tower,” he said.
In addition, one of the benefits of agile working is peer learning. “A side effect of agile is you can bring people into this team who may not have a depth of understanding but through the activity of executing these projects, they get up to speed.”
“They’ll sit side by side with them and get skilled up.”
In fact, he said this is how knowledge transfer takes places when members of his team are embedded in other organisations. He said: “We want to skill up in the space. They’ll join the team, they’ll sit side by side with them, they’ll do the work with them and they’ll get skilled up.”
Dobrin said that it is important to bear in mind the need to choose the most appropriate form of agile working that makes the most sense for different phases of the project.
He said that when you look at the successful implementation of AI projects in large enterprises, it is typically in a hub and spoke model, with a centralised team and other teams in the business unit.
Dobrin said there are two reasons for this. The first is that those who are in the business units retain
a sense of urgency and understand the need for the outcome of the project.
Secondly, the centralised team is able to ensure consistency across the organisation. “There should be a lot of reuse of the features, of the data that the model grabs on to as well as some of the models that are built from one business unit to the other. You also want consistency of toolsets and guardrails for the team,” he explained.
“Data science is a team sport.”
Dobrin conceded that a hub and spoke model would easier for some companies to adopt if they are further along their data maturity journey. He advised companies that have no AI capability to start with a centre of excellence and later move towards a hub and spoke model.
Other companies that have AI capabilities spread across the organisation should think about how they could coalesce those teams, get them working together and then move towards the hub and spoke model.
"You need to focus on outcome and implementation."
Teamwork is closely aligned with the practice of peer learning, which Dobrin said is imperative for a successful AI project as for him, data science is a team sport. “You will not be successful if there is not a relationship and they cannot work effectively together on a team,” he said referring to machine learning experts, decision optimisation experts and data engineers who have an understanding of those technologies, and would all fall under the catch-all term of data scientists. He also added that AI and machine learning engineers are critical to the successful deployment of AI projects.
In order to realise the value of the AI project it is important that the resulting model is used, and therefore according to Dobrin, it is really important to focus on the group that is going to consume the models and how they will do so. “You need to focus on not just the ease of delivering the outcome but the implementation because implementation is actually harder. It’s a cultural change. You’ve got to get buy-in,” he said.