Team architecture/structure options

In my mind, most concepts exist on a continuum. Building a successful advanced analytics and AI team is typified by two approaches that inhabit the two poles of the relevant continuum – Artisanal or Factory:

Figure 2.1: Artisanal and Factory team structure comparison

The Artisanal team architecture/structure

Let's start with the Artisanal approach.

The Artisanal approach is where the data scientists are the owners, managers, experts, and driving force behind their projects.

The data scientists design and execute every step of the process. The data scientists engage with the project sponsors, the subject matter experts, internal and external consultants, syndicated data providers, and any other inpidual or group that has a role to play in the project.

Data scientists capable of executing and managing the artisanal approach possess exemplary communication skills, are open to listening to a wide range of stakeholders, can work with internal and external parties, are knowledgeable about a wide range of analytical techniques, are probably experts in one or two analytical approaches, have the ability to understand the immediate next steps as well as the long-term project goals and objectives, and can translate the technical objectives of the project into business goals and financial returns.

Data scientists who can execute this model successfully are the most skilled data scientists that you can find and hire.

When does it make sense to take the Artisanal approach to building your analytics team?

1. Cost: While inpidual data scientists will cost more (you will need to pay top market rates for these staff members), the aggregate cost of the entire staff will be less than in the Factory model.
2. Focus: We are still assuming that you are just committing to building a team and launching the corporate analytics initiative. The organization is, in all likelihood, not ready to undertake more than a handful of advanced analytics and AI projects. And the projects that you will undertake are probably high profile and very important. Hence, you want to start with a small number of highly qualified projects that line up with the technical expertise of the data scientists that you will be hiring.
3. Hiring: It will be easier for you to find 3 to 5 highly qualified and talented data scientists rather than trying to hire 15 to 30 people who have narrow and specific skills that fit into the data science process outlined above.
4. Team cohesion and dynamics: Hiring a few rock stars and treating them right and molding them into a team is easier and more fun than hiring a large number of people focused on more mechanical work.
5. Executive buy-in: When highly qualified data scientists complete their work and present the projects and related business impact to project sponsors and senior executives, you will have greater executive buy-in and an easier time obtaining new and additional funding for team expansion.
6. Word of mouth: When the initial project succeeds, the subject matter experts will talk to subject matter experts in other areas of the company. Those subject matter experts will talk to their managers who in turn will seek you out to discuss how they can sponsor projects with your team.
7. Team marketing: After the success of the initial projects, you will have stories to tell of business improvement and transformation. Nothing increases your chances of future success like past success.
8. Personal reputation: Your personal effectiveness and the organizational awareness of the impacts of your team and your inpidual efforts will raise your profile and provide you with access to new opportunities.

The Factory team architecture/structure

Now let's move to the opposite end of the continuum or spectrum and discuss the Factory approach to building an advanced analytics and AI team.

The Factory approach draws its name from the mass production approach to a process-oriented endeavor. Envision a factory where there are stations and personnel who occupy those stations performing a singular task. Imagine an automotive factory where one team puts in the engine and transmission assembly and another person puts on a tire. The Factory approach to data science is similar to that of the automotive assembly process.

In the Factory model, the work process is broken down into single-focus, repeatable steps, and the process is staffed with people appropriate to execute those steps as part of the larger data science process. You may structure the team in several ways.

One possibility would be to structure the work process in the following manner:

1. Data acquisition
2. Data cleaning and integration
3. Feature engineering
4. Model building
5. Model validation
6. Project management
7. Stakeholder (sponsor and subject matter expert) management
8. Project-to-production management
9. Ongoing model and application upgrades and maintenance

The Factory process relies on a larger number of people who are executing smaller, specialized steps in the overall process.

When does it make sense to take the Factory approach to building your analytics team?

1. Cost: While the inpidual staff members will cost much less, the aggregate cost of the entire staff will be more than in the Artisanal model, but to be fair, the team will be much larger and have a substantially larger aggregate team capability and throughput. The cost in this approach is greater in aggregate, but the Factory approach enables a model and foundation for scaling the corporate capability and functionality.
2. Focus: We are still assuming that you are just committing to building a team and launching the corporate analytics initiative. If you are fortunate enough, executive management may have made a significant, public commitment to going full throttle into analytics. In that case, the organization has already committed to undertaking substantial and well-funded advanced analytics and AI projects. The projects will be high profile and very important. You want to take direction from executive and senior management as to where to start from a project perspective, and that will determine who you hire first.
3. Hiring: Hiring is easier in that the team members do not need to be multifaceted data science or analytics professionals, but you do need to hire more people than in the Artisanal approach. A suggested approach is to design your data science approach and look to hire one person for each station or step of the process to start. Optimally, it would be best if you could hire leaders for each station or step in the process, but in some cases, you will need to start with entry-level talent to have a functioning end-to-end process.
4. Executive buy-in: As the leader of the advanced analytics and AI team and effort, you will be presenting the projects and related business impact to project sponsors and senior executives. Hiring will take longer than in the Artisanal approach and therefore you will need to manage executive expectations of when the initial projects will deliver operational results.
5. Word of mouth: When the initial project succeeds, the subject matter experts will talk to subject matter experts in other areas of the company. Those subject matter experts will talk to their managers who in turn will seek you out to discuss how they can sponsor projects with your team.
6. Team marketing: After the success of the initial projects, you will have stories to tell of business improvement and transformation. Nothing increases your chances of future success like past success.
7. Personal reputation: Your personal effectiveness and the organizational awareness of the impacts of your team and your inpidual efforts will raise your profile and provide you with access to new opportunities.

The Hybrid team architecture/structures

We have outlined the two ends of the spectrum for hiring an advanced analytics and AI team. I recommend that you select one and execute toward building a functional, high-performing team. The objective of reaching a high-performing team can be achieved from either starting point with the same probability of success.

What happens now? A few things can happen. You can maintain the Artisanal or Factory model that was the starting point of this journey, but more than likely, the team will begin to evolve. You can let the team morph and change on its own, but it is smarter and better for the team if you manage the evolution.

The midpoint of the continuum that we have discussed is the Hybrid model. Of course, your team can skew toward the Artisanal or the Factory model, and in both cases, that is fine, and you will have a functioning team and operation. You should determine how the team evolves.

Do you want the Factory model to move toward the other end of the spectrum where you augment the existing Factory type operation with Artisanal data scientists? This can work very well. You can have the Factory operation undertake the work to support the data scientists – work such as data acquisition, data integration, structuring visualizations, and building infrastructure and security elements required to support complete applications.

Conversely, you can move in the reverse or opposite direction and build a Factory operation in addition to your existing Artisanal team.

Typically, the Artisanal team will be working on high-value applications that will have a substantial impact on business operations. These high-value applications can be considered proprietary intellectual property and you may want to engage the legal team to protect them via patent applications.

In addition to these high-value applications, there will be applications that you should build and implement that ensure that your firm and operations maintain competitiveness and cost-effectiveness. These analytical applications vary based on the industry and market that the firm operates and competes in, but some examples that are relevant to multiple industries and markets are supply chain simulation, pricing optimization, dynamic delivery route determination, and many others. These applications can be built and deployed by the Factory elements of your operation and team or outsourced to third-party consultants. The consulting effort would be managed by the Factory operation of the team.

With a Hybrid model, you can achieve the creation and deployment of innovative intellectual property that can change the competitive profile of the company and you can build applications that keep your firm at the leading edge of operational effectiveness and efficiency. The Artisanal group can drive innovation, and change, through the deployment of highly valued patentable applications that put and keep the firm on the leading edge of any industry. The Factory group can build, deploy, and maintain analytical applications that achieve and maintain operational efficiency across global operations at scale.

I have built and evolved analytics teams starting at each end of the spectrum. My preferred method is to start with an Artisanal team and evolve toward a Hybrid model. Starting with the Artisanal model has a number of benefits to you, the analytics team, and the company as a whole. The first benefit is that starting with an Artisanal team enables you and the analytics team to connect with C-Level and senior executives on mission-critical problems. The second benefit is that you and the analytics team can act like a SWAT team, moving quickly to understand the problem, moving through modeling the problem, and offering solutions and insights in a matter of weeks or days rather than months or years. The early success can be followed up by taking on larger, more complex problems, which will require a larger team, which marks the start of the evolution toward the Hybrid model. This approach works very well, and I recommend it highly.