Same projects. Different combination. Greater results.
You can achieve higher returns with your existing projects.
We calculate the optimum scenario - before you decide.
Free of charge. Without obligation. Based on your existing projects.
StratePlan calculates the optimal portfolio where traditional tools reach their limits.
Instead of evaluating projects in isolation, we analyze all possible combinations - and identify the best solution.
The global optimum is not an assumption - it can be calculated.
Select business area:
Blog main article:
Optimizing local optima with AI: Why better decisions can still be wrong
In many organizations, optimization is a sign of professionalism. Processes are improved, projects are fine-tuned, budgets are fine-tuned. KPIs increase, risks decrease, processes appear controlled. And yet the results often fall short of expectations.
The reason is paradoxical - but systematic: Organizations optimize successfully, but in the wrong place.
They optimize local optima. And this is precisely the core of the problem.
This article shows why "optimizing local optima" with classic management logic inevitably fails, why AI works here not as automation but as a decision-making tool - and how StratePlan enables the leap from local to global optima.
What are local optima - and why are they so tempting?
A local optimum is a solution that cannot be further improved within a limited decision area. Every small change seems to worsen the result. From the perspective of the team, department or program, you are "at the finish line".
Local optima therefore feel right:
- They are the result of intensive analysis.
- They are supported by figures and KPIs.
- They are politically compatible.
- They appear efficient.
This is precisely why local optima are so stable. And that is precisely why they are so dangerous.
The structural error: optimization without an understanding of space
Optimization presupposes that the space in which optimization takes place is known. In practice, however, this space is severely restricted: Departments only see their projects, programs only see their initiatives, committees only see the variants under discussion.
What is missing is a view of the entire decision-making space.
As soon as several projects compete simultaneously - for budget, resources, time and attention - the result is no longer a linear decision-making process, but a combinatorial space. This space grows exponentially.
With N projects, there are not N options, but:
2N possible project combinations
This means that with just 50 projects, we are talking about more than 1.125 quadrillion possible portfolios. Each locally optimized state is only one point in this space - not necessarily a good one.
Why more optimization often exacerbates the problem
A common misconception is that if the result is not good enough, we simply need to optimize even better. More analysis, finer KPIs, more detailed control.
In reality, the opposite often happens:
- Local optima are further cemented.
- Dependencies between projects become more entrenched.
- Resource bottlenecks shift, but are not resolved.
- The system becomes more sluggish overall.
This is not a management problem, but a mathematical one. Local optimization without a global understanding of space reinforces suboptimality.
The error in thinking of classic control systems
Classic control systems are designed for stability. They measure, compare and correct. This works excellently in linear systems. In highly complex, networked decision spaces, however, it leads to a systematic bias:
What is improved is what is visible - not what would be effective.
Local optima lie within the field of vision. The global optimum is almost always outside.
Why AI does not "automate" here, but opens up the space
When we talk about "optimizing local optima with AI", we are not talking about making existing processes faster or cheaper. It is about a categorical change:
- From discussion to calculation
- From variants to combinations
- From individual projects to portfolios
AI is not used here as a forecasting machine, but as a decision space technology. It models projects, restrictions, dependencies and goals - and searches the entire space for better combinations.
This reveals what was previously hidden: how many local optima exist - and where the global optimum actually lies.
A comparison of sizes:
our Milky Way and a corporate decision space with "only" 50 projects
of 1.125 quadrillion possible project combinations
Why local optima are particularly expensive in portfolios
In portfolios, local optima often arise in the wrong places:
- A project is perfected but blocks critical resources.
- A program is completed even though it delays others.
- A budget is used optimally - but in the wrong year.
Each of these decisions can be "right" on its own. Together, however, they create friction, delays and opportunity costs.
The paradox: the better individual units optimize, the worse the overall result can be.
StratePlan: Making local optima visible - and leaving them behind
This is precisely where StratePlan comes in. The approach is not to ignore local optima, but to make them transparent. Only when it is clear where local maxima lie can a conscious decision be made to abandon them.
All projects are modeled together. Restrictions become explicit. Dependencies become calculable. This results in a decision space that is searched algorithmically.
StratePlan calculates the entire decision spaceand finds from it:
The one project combination that generates the maximum overall benefit.
The real added value: deliberate deviations
A global optimum does not mean that it always has to be implemented. The decisive advantage is a different one: deviations become conscious.
If a committee wants to retain a locally optimized project for political, regulatory or strategic reasons, StratePlan shows what this decision costs - and which alternatives are close to the global optimum.
This turns implicit suboptimality into an explicit decision.
Executive takeaway
Local optima are not a sign of poor leadership. They are a sign of limited vision.
Those who continue to optimize local optima stabilize the system - but do not improve it. Those who calculate the decision space gain the freedom to leave local maxima.
AI thus becomes an instrument of strategic clarity rather than an automator of decisions.
StratePlan makes exactly that possible.
Dr. Igor Kadoshchuk is a computer scientist, algorithm architect, and one of the leading minds behind mAInthink's optimization and decision-making algorithms. As scientific director of the StratePlan™ and DeepAnT platforms, he combines in-depth mathematical research with practical applications in project portfolio optimization, business, finance, and public administration.
He holds a PhD in computer science from the renowned Moscow Institute of Physics and Technology (MIPT), where he also taught as a professor of computer engineering and mathematics. He has decades of experience developing highly complex mathematical models for project portfolio optimization and financial systems, investment planning, and strategic decision-making. His professional career includes leading positions such as Head of IT at Gazprombank and Director of Project Management at TransTeleCom.
Dr. Kadoshchuk writes on the mAInthink AI Blog. Kadoshchuk on:
- Algorithmic strategy optimization
- New methods for calculating ROI and impact
- Project portfolio optimization beyond traditional tools
- The limits of human decision-making – and how AI overcomes them
His aim: to calculate strategy, not estimate it.
His contributions combine scientific precision with clear, understandable language – always with the goal of making complex decision-making spaces transparent, manageable, and measurable.