Meta’s AI Team Reorganization Is The Structure Most Teams Will Follow

When businesses start with data, talent is scattered throughout the company. Analysts and even Data Scientists are in different business units. Data and Machine Learning Engineers are in IT or part of Software Development organizations. It’s chaos.

For maturity to happen, people need to be centralized into a single Data and Analytics organization. It is sometimes called a Center of Excellence model. With resources in one place, building out can happen with a single vision and direction.

Infrastructure is centralized and unified under a single architectural design. Tools and workflows are standardized. Data is centralized and managed under a single set of best practices. Standardization and continuous improvement bring order to chaos and optimize the Data and Analytics organization to build what the business and customers need.

Eventually, the organization builds products it thinks the business needs instead of what it actually needs. The Data and Analytics organization calls it innovation, but the rest of the company does not benefit much from initiatives.

New leadership gets brought in. The first round of leaders are mainly technical leaders. The second round are business and strategic leaders. If the old and new teams of leaders find ways to collaborate, the partnership brings the Data and Analytics organization back into the business. They start working with the business on innovative projects and delivering on more mundane but very high-value projects.

This all works until we reach the point where Meta was at the end of last year. The centralized organization has its own leadership, and they have strong ideas about where their people’s time should be allocated. This usually leans towards forward-looking innovation over immediate business needs. Obviously, that causes some friction across the business.

While innovation delivers high-value hits, the team will remain centralized and have a high degree of latitude over projects. When growth slows, like it is at many companies, the mix of innovation vs. immediate business needs comes under scrutiny. How committed is the C Suite to long-term investments?

At Meta, and many companies will follow their lead, immediate business needs won. The head of Meta’s AI group is leaving this month. That is the most common ending to the Center of Excellence or centralized Data and Analytics organization once the boom ends.

Each business unit needs greater leadership control over data and analytics resources for their projects to succeed. The centralized leadership creates friction between business units and the team. Leadership must coordinate with the Data and Analytics organization’s leaders. External organizations don’t have control of all the resources they need to deliver their work. If the Data and Analytics organizational leaders have different priorities, more innovation typically, external teams can miss their delivery deadlines.

The winner in these internal disagreements is always the group pulling in the most revenue. In my strategy course, I teach students how to get revenue attached to the Data and Analytics organization. The more revenue they can show, the more leverage they have to maintain control over their work. This scenario is why I emphasize this point so much.

Some AI Strategy KPIs are designed to show that data initiatives return more to the business than an alternative investment. This is another layer of defense for the Data and Analytics organization. The early attacks on the centralized organization’s leaders come during the budget planning process. While leadership can defend their budget based on a higher return per dollar spent than the alternatives, the Data and Analytics organization retains its autonomy.

The goal is to hold the centralized structure together long enough to mature and meet the business’s immediate and long-term needs. Break the organization up too soon, and progress slows. Data science capabilities plateau and the decentralized organization matures in multiple directions based on each business units’ needs. Again, the result is chaos.

At Meta, it looks like their leadership accomplished the mission. Their outgoing leader built the transformation plan from centralized to decentralized Data and Analytics capabilities. It is nearly complete, so they are stepping aside this month.

Meta’s team is being distributed across product lines and embedded into those organizations. Long term that is always the endgame. This structure puts domain experts and data experts as close as possible. The result is better products.

The next challenge is to maintain the innovative initiatives. Those can disappear overnight if a structure isn’t established to preserve them. In Meta’s case, each product organization must maintain a certain percentage of investment in innovation. I assume the leadership of each organization has goals tied to those initiatives. That is a successful framework to keep innovation alive while refocusing on mainline products.