The business of data science
The 3 Steps to Create a WORLD CLASS Artificial Intelligence Center of Excellence
Breaking down the steps to create your organizations AI CoE
If you’re reading this, it’s likely that you hold one of the following roles (or very similar)
- Head of Data Science
- Head of Artificial Intelligence
- Chief Data Officer
- Chief Information Officer
And you’ve been tasked with creating your company’s Artificial Intelligence “Center of Excellence.” Luckily, I’ve worked on this problem before with two clients, and I’d love to share my learnings with you. I hope you’ll find them valuable and thought-provoking as you engage in this journey.
The first thing I’d like to cover, is the idea of “systems-thinking.” We hear this phrase a lot, yet rarely apply it to our day-to-day operations.
The principles of systems thinking
There are (5) main principles to consider when designing an organization or management system and I find that they hold true.
Source: Richard A. Johnson, Fremont E. Kast, and James E. Rosenzweig, The Theory and Management of Systems, 3rd ed. (New York: McGraw-Hill, 1973), pp. 144–46.
- Simplicity
- Flexibility
- Economy
- Efficiency
- Acceptability
Briefly, let’s discuss each
- Simplicity
We should apply the principles of parsimony, that is, don’t add complexity for complexities sake. Your organization should look and run simply. Any additional complexity adds bureaucracy, thus increasing the cost of administration. Keep it simple. Keep it nimble
2. Flexibility
By flexible, I mean, your organization should be able to change and grow or, in cases, reduce in size, as the needs of the larger company change over time. You shouldn’t have to keep “recreating the wheel” for each evolution of the company. The organization should scale and downscale as necessary
3. Reliability
Your system should work as intended, without adding pain-points to the main task: getting good work done. When someone refers to your organization, or center of excellence, it should be in high regard because of the quality and reliability you’ve naturally designed into the process
4. Economy
Efficiency is key. No duplication of efforts, no redundancy. This will allow you to minimize the cost of running your center of excellence
5. Acceptability
You can’t upturn the entire company to create this. It has to fit somewhat into your company’s modus operandi. If people can’t accept it, they won’t, and your hard efforts will go no where.
OK. That covers the main principles. Keep these in mind at all times as you design the organization that will likely change your company and drive it’s future potential.
The (3) steps to creating your AI Center of Excellence
I wish I could say each of these would be an easy task, but more often than not, they are difficult. It may make sense to hire outside consultants to outsource some of these activities if you’re still managing day-to-day operations, but they need to be done.
In order, they are:
- Assess the current state
- Design your operating model
- Implement
Assessing the current state
In the best case scenario, you’re building this from scratch without an existing data organization. But, given the likelihood of that is low, you’re likely inheriting some legacy processes, people and systems. You need to get a complete and cohesive picture of everything “data” in your organization. This consists of interviews, meetings, charts, flows, etc.
I find it extremely valuable to meet with each business unit leader, then go one (or sometimes two) steps down to the manager in the front line. You’ll often find that the picture painted by the leader is often wishful thinking, and the actual state of things is much different. This is fine, remember, you’re going to create the organization that solves these problems.
At a minimum, you need to get a picture of
- The systems and data being collected
- Any “data-science” or business intelligence activities that are being used to support the business
- The people embedded in the business unit that are responsible for delivering on item (2)
If you’re dealing with a large corporation, this may include understanding your data warehousing organization, and each individual operating unit. If you’re small, this probably includes interviewing a handful of business analysts. You’ll need to understand everything data going on in the business. This is the key to designing a CoE that serves everyone and ultimately, your success.
Once you have a detailed picture of the current state, you can move on to design
Design your operating model
I’m going to introduce two different models, and explain the pros and cons of each. You will need to decide which is appropriate for your organization, given the existing culture and status quo. Remember, the principle of Acceptability — you will go far if people can accept and work with your Center of Excellence.
The “federated” or “hub-and-spoke” model
This model works best in very large organizations, where each business unit has existing data scientists and data specialists. You’ll find that their standards and operating procedures are disjointed. This is an opportunity for your new CoE to create standards and procedures that *enhance* the day-to-day work of the people who work in these specialized groups.
What this means:
You will create a smaller (relative to the number of people in your data organization) group of experts. These are seasoned veterans, machine learning experts and data warehousing gurus. These individuals will provide support, mentoring and standards of excellence to the people who are working in the business. They will serve as the source of new information ,techniques and ideas for how to use their data more effectively.
They will work alongside the embedded data specialists in the business unit to deliver on interesting new use-cases. In some cases, you may want to create a matrix reporting structure, where business analysts and embedded data scientists also report to a member of the CoE
They will provide governance and acceptance criteria for data projects.
Essentially, you’re creating a team of experts in Artificial Intelligence and Data. You are not hiring entry level data scientists, but seasoned veterans who are capable of demonstrating and generating “authority” in the field. If your embedded data specialists have more expertise in Artificial Intelligence techniques, you’ve already lost. You should be prepared to hire experts who can also evoke organizational change through leadership. This means the salary requirements are significantly higher than your average AI specialist.
The day-to-day operations largely remain the same unless some egregious behavior occurs (data duplication, duplication of efforts, etc). This will come out of your current state assessment as areas to consider.
Pros:
- Retain the specialized model of embedded data scientists
- Smaller cost and burden on the organization
Cons
- Smaller size may lead to difficulty affecting change in procedures and business units
- May be more difficult to provide governance and oversight
The Centralized Model
This works with smaller organizations or organizations that have little to no AI capability today.
You will create a “soup-to-nuts” data organization that handles everything AI. This could include, a centralized data warehousing and management function, a team of data scientists that may serve different areas of your organization and a team of data engineers/BI specialists that work to massage data and create reports and visualizations.
If you’re restructuring a larger organization, this may mean re-organizing people to new operating units and unfortunately, letting redundant analysts go or repurposing their roles.
The centralized CoE delivers everything and anything data. They help the business understand what data they’re collecting, and come up with the use cases to better leverage their data with AI.
I find that the size of the underlying organization dictates the most effective model. The centralized model breaks down as the organization gets very large, and likely becomes a bottleneck or bureaucracy.
Pros:
- Easier to manage oversight and governance
- Bigger budget may allow for more experimentation
Cons
- May become a bureaucratic burden to launching AI projects throughout the organization
- May not have enough resources to act on every opportunity to use AI
What your Center of Excellence MUST do
There are a number of things that you want or envision this center of excellence doing, but at a minimum it must accomplish (3) things, preferably soon after creation
- Become the central source of AI thought leadership
This group must be ready and willing to deliver thought-provoking and interesting use cases and ideation regarding AI to the rest of the organization. It must be ready and willing to answer anyone who has questions such as: “How can I benefit from AI?” As a plus, this organization may engage in external seminars and content creation that establishes your company as an AI thought leader
2. Centralize AI vendor management
This group must evaluate and inventory AI tools and provide a means for analysts and data scientists to gain access to these tools.
3. Provide a roadmap for AI for your executive team
Your newly established group must provide a roadmap of future AI initiatives for the company to embark on. This is the main purpose of your AI Center of Excellence
There are many more considerations to make, such as how to hire and retain the best talent, what standards should be implemented, what vendors and tools can enhance AI “uptake” at your organization, etc. If you’re interested in finding out more, please subscribe!
Thanks for reading!
