Can reasoning agents predict as well as machine learning with less data?

Organizations have wanted to predict the future better for decades. Whether it concerns revenue projections, demand forecasting, churn analysis, or financial planning: better predictions lead to better decisions. Traditionally, this challenge was addressed with predictive models and machine learning. These techniques are powerful but also have clear limitations. With the rise of generative AI and so-called reasoning agents, a new approach is emerging that could potentially have a major impact on how organizations make predictions.

The power and complexity of traditional predictive models

Predictive models based on machine learning have become enormously popular in recent years. They can recognize patterns in historical data and predict future outcomes based on those patterns.

However, these models have several clear characteristics:

1. Large amounts of data are required
To make reliable predictions, machine learning models often need extensive datasets. Not only over a long period but also with many different features. Examples include customer behavior, economic indicators, marketing activities, or seasonal effects.

2. High development costs
Building a good predictive model is labor-intensive. Data scientists must collect data, clean it, select features, train models, and optimize them. This process can take weeks or even months.

3. Customization per use case
Many models are specifically designed for one problem. For instance, a churn model does not automatically work for revenue forecasting. Consequently, organizations often have to develop and maintain multiple models.

All this makes traditional predictive AI powerful, but also costly and complex.

The rise of reasoning agents

With the breakthrough of generative AI and large language models, we are seeing the emergence of a new category of AI systems: reasoning agents.

These systems are not primarily built to learn statistical patterns from massive datasets. Instead, they are trained to reason logically, calculate scenarios, and establish connections based on available information.

Thanks to the enormous progress in model architectures and computing power, these agents can:

• analyze complex problems
• make assumptions explicit
• calculate scenarios
• build reasoning step-by-step

This raises an interesting question: can reasoning agents still make reliable predictions with relatively little data?

An experiment: machine learning vs. reasoning

To investigate this, a practical case study was conducted at an organization that wanted to create financial forecasts. In this case, two approaches were compared side-by-side:

1. A traditional machine learning model
2. A reasoning agent based on generative AI

Both systems were given the same business context and comparable input data.

The result was surprising.

The forecasts from the reasoning agent matched the machine learning model's predictions by approximately 90%.

In other words: despite significantly less data preparation and model training, the reasoning agent was able to generate nearly the same outcomes.

This suggests that logical reasoning combined with limited data can be surprisingly powerful in many situations.

The advantages of reasoning agents

This development has several important implications.

1. Less dependence on large datasets

Because reasoning agents rely more heavily on logic and context, they can often work with less historical data.

This is particularly interesting for organizations that:

• have limited datasets
• are launching new products
• want to generate forecasts quickly

2. Faster implementation

A traditional machine learning trajectory can take months. Reasoning agents can often be deployed within days or weeks.

3. More broadly accessible

Because less specialized model development is required, predictive AI becomes accessible to a larger group of organizations.

But there is also a downside

While reasoning agents have significant potential, there are also important caveats.

1. Token and compute costs

In machine learning, the largest investment is often in the development phase: collecting data, training, and optimizing models.

With reasoning agents, that cost structure shifts.

The model itself is already trained, but every time you use it, it must think, analyze, and calculate. This happens via tokens and compute, which can generate significant costs with intensive use.

2. Consistency

A well-trained machine learning model often produces very stable results.

Reasoning agents can be more variable because they perform each analysis anew.

3. Governance and control

With statistical models, it is often clear which variables influence the result. With reasoning agents, it can be more difficult to fully control that influence.

A new balance in predictive AI

We are at an interesting tipping point.

Traditional machine learning models remain extremely valuable, especially in situations where:

• massive datasets are available
• high accuracy is crucial
• predictions are used continuously

But reasoning agents offer a faster, more flexible, and more accessible route to predictive insights.

In many cases, organizations will likely opt for a hybrid approach:

• machine learning for structural, large-scale predictions
• reasoning agents for rapid analysis, scenarios, and new use cases

Conclusion

The development of reasoning agents shows that predictive AI is no longer exclusively dependent on complex machine learning pipelines and enormous datasets.

Experiments show that this new generation of AI systems can achieve up to 90% comparable results in forecasting, with significantly lower data requirements and development time.

That makes predictive AI more accessible than ever.

At the same time, this new approach requires a different perspective on costs, governance, and implementation.

What is clear: reasoning agents are moving closer and closer to the power of traditional predictive models and that could fundamentally change the way organizations make decisions.

Getting started with AI yourself? View our e-learning AI Agents or AI workshops for teams.

// About the author

Job van den Berg

Mede-oprichter, AI Keynote Spreker & Techondernemer

Tech-ondernemer (1989) met een achtergrond als socioloog (Research Master (MSc) in statistiek en sociologie) en een van de meest gevraagde keynote sprekers over AI en data in Nederland. Als mede-oprichter van Ai.nl, The Automation Group en Proxies leidt hij engineers die agentic AI van prototype naar productie brengen binnen enterprises. Op het podium vertaalt Job die hands-on praktijk naar concrete strategieën. Eerder was Job Chief Data bij o.a. DPG Media en Kantar. Hij is co-auteur van 5 boeken over AI waaronder 'AI Agents' en 'Handboek AI Strategie' en een veelgevraagd expert in de landelijke media.

LinkedIn