What if the biggest challenge in enterprise AI were not making better decisions, but getting all decisions to work together?
Why aligning goals, information and local decisions matters if AI is to improve outcomes for the whole organisation.
A few days ago, I published a reflection on a question I hear constantly in the world of Workforce Management:
“Can we use artificial intelligence to generate an optimal schedule automatically?”
My answer was that the challenge is not simply finding the best solution, but defining what optimal really means.
Optimising costs is not the same as optimising productivity, service quality, employee wellbeing or team stability. Improving one of these dimensions can come at the expense of another.
But looking more closely at this question leads to another:
Who decides what is optimal within a company?
Senior leaders set strategic goals: profitability, growth, productivity or cost reduction.
Middle managers have to translate those goals into operational reality: people, shifts, absences, constraints and service needs that change every day.
Both perspectives are necessary. Both are legitimate. And they do not always align.
Imagine a company with twenty workplaces. Each manager must match staffing to demand while meeting a labour-cost target.
At one site, activity drops on Thursday. Its manager reduces coverage to avoid paying for hours that, according to their forecast, are not needed. At a nearby site, demand surges on the same day and an unexpected absence occurs. That manager turns to overtime to maintain service.
Each decision may seem reasonable in isolation. Yet a coordinated plan might have anticipated that an employee at the first site—with the right training and within working-time and travel constraints—could cover part of the need at the second.
Why did that not happen?
Perhaps each manager could only see demand at their own site. Perhaps the schedules were finalised at different times. Or perhaps lending staff would hurt the lending manager’s performance metric, even if it benefited the company.
The problem was not just calculating better shifts. It was coordinating information, goals and responsibilities.
An algorithm can optimise each site separately and reproduce the same problem—only faster.
The sum of individually optimal decisions does not necessarily produce the best collective outcome.
This question connects two areas to which I have devoted much of my professional life: business management and intelligent-systems research.
It also connects with some of the great ideas in management.
Peter Drucker focused on management by objectives and managerial responsibility. Herbert Simon explained the limits of our rationality and of the information available to us when we make decisions. Henry Mintzberg studied the full complexity of managerial work. And Peter Senge highlighted the importance of understanding organisations as interconnected systems.
Artificial intelligence gives us new tools to explore these questions.
Mathematical optimisation can help us plan resources around multiple goals and constraints. Game theory lets us analyse how incentives shape decisions. Multi-agent systems provide models for coordinating autonomous entities. Reinforcement learning lets us explore policies that adapt based on their outcomes. And digital twins can help us simulate the consequences of different decisions before applying them.
But none of these tools can decide what the company should prioritise.
We still need to define which goals we are pursuing, who has the authority to decide, and how conflicts between responsibilities should be resolved.
For the example of the twenty sites, that means answering very practical questions:
Who can authorise a staff transfer? What information do managers need to share? How should we recognise a site that helps another? When should a local need take priority, and when should the overall result prevail?
Over the coming weeks, I want to share a series of ten articles on the relationship between business management, artificial intelligence and decision-making.
We will begin by examining why optimising each department does not mean optimising the company. Then we will look at the role of middle managers, management by objectives, incentives, and conflicts between strategic and operational decisions.
From there, we will explore how hierarchical optimisation, multi-agent systems, Decision Intelligence, digital twins and reinforcement learning can contribute.
I will use examples from people management and Workforce Management, although many of these ideas apply to other parts of a business.
I am not trying to offer definitive answers. I want to connect business experience with scientific research and explore how we can design better decision-making systems.
Perhaps the future of enterprise AI is not about building one grand algorithm that makes decisions for everyone.
Perhaps it is about helping people with different responsibilities, information and goals make better decisions together.
That brings us to the question that will run through this series:
Which decisions should a company centralise, and where can artificial intelligence expand managers’ autonomy without losing coordination?
I believe it is worth investigating.