Learning Fast From Limited Data: The Hidden Measure Of Enterprise Intelligence

Originally published on Forbes Technology Council.

When people discuss intelligence, whether human or artificial, the conversation usually turns to raw power: memory, computing speed and data scale. But there's another and often more important measure: sample efficiency.

Sample efficiency is the ability to learn quickly from little data. Humans excel at it. It doesn’t take long to learn a new game or identify objects, even with limited visual information. Small children comprehend language without understanding vocabulary and grammar. Teenagers learn to drive in under 100 hours. Autonomous vehicles, by contrast, have already logged more than 145 million miles as of May 2025, and are still working toward full reliability in complex conditions like snowy weather and heavy traffic.

The gap between sample efficiency and inefficiency matters. Evolution optimized humans to make decisions from sparse signals, as survival often depended on it. In today’s volatile business environment, efficient processes can serve as a competitive advantage, allowing companies to better adapt to market changes, make improvements and scale their operations.

Why Sample Efficiency Matters

Traditional enterprise planning systems are built on history and perform best when fed years of data. But market conditions don't wait for historical certainty. Consumer preferences shift overnight. Disruptions strike without warning.

Enterprises that wait to accumulate and evaluate “enough” data before acting are already too late. Consider the following three planning moments:

• New Product Launches: Waiting six months for stable sales data means the company will miss its launch window. Intelligent planning systems can recalibrate forecasts after just a few weeks following the launch, combining early sales velocity with external signals such as social media buzz and competitor activity.
• Supplier Risk: If a business requires multiple late deliveries to recognize fragility, it will result in a loss of customer trust. Sample-efficient planning systems treat the first missed shipment as a signal, contextualizing it with strike alerts or logistics bottlenecks to rebalance supply.
• Promotions: If effectiveness is measured only after the promotional campaign ends, the company loses both money and momentum. Agile planning learns from the first weekend of sales and scales successful tactics in real time.

Each of these examples reveals a different aspect of sample efficiency. Together, they define how agile, resilient and competitive a business can be.

Lessons Learned From Experience

I’ve spent several years working with manufacturing leaders who are under constant pressure to make important decisions with far less data than they’d like. Sample efficiency has become a quiet revolution because an AI model can learn to perform well from a small number of examples instead of relying on massive datasets. Powerful applications are seen in demand forecasting, distribution planning and manufacturing optimization.

A retailer, for instance, wanted to improve store-level forecasting and replenishment but had limited data for its newer and smaller locations. Traditional models struggled to predict local demand patterns. A sample-efficient AI model was introduced and could learn from just a few weeks of data per store, while transferring insights from similar stores and external signals. Within months, the system could forecast demand with remarkable accuracy and automatically recommend replenishment quantities by SKU and location.

This improved shelf availability, lowered excess stock and reduced inventory carrying costs. Planners embraced the model because it didn’t replace their intuition; it amplified it by flagging where they would have the greatest impact.

Another example is a consumer-goods company entering a new region, using a sample-efficient forecasting model that “borrowed” insights from mature markets. As a result, forecast accuracy greatly improved, freeing up working capital tied up in excess stock.

Small datasets, if intelligently leveraged, can drive great outcomes. But they also demand discipline. When every data point counts, data quality becomes everything. One mislogged transaction can throw off an entire model. I’ve learned to double-check inputs, simplify model design and favor transparency over complexity.

It's also important to point out that many planners are understandably skeptical of a model trained on “too little” data. I make it a point to involve them early, showing how the model learns, validating results together and treating the AI as an assistant. I’ve seen planners evolve from spreadsheet operators to AI copilots once they get hands-on experience and upskilling through workshops, sandbox environments and embedding planners in real AI projects. Over time, trust grows when planners see the model predicting correctly or at least explaining its reasoning clearly.

One thing I’ve learned is that sample efficiency is both a technical and a human capability. The algorithms may learn faster, but real efficiency comes when people learn to work with them. When that happens, the line between data and decision disappears and planning becomes what it was always meant to be: fast, intelligent and deeply human.

The Winners Of Tomorrow

Some theorists argue that true intelligence is nothing more than sample efficiency: The ability to learn the most from the least. Regardless of whether that's fully accurate, it's a powerful lens for business leaders.

The winners of tomorrow won't necessarily be the companies with the largest datasets. They'll be the ones that can extract meaning from the smallest signals and act on them with confidence.

I believe this is the real definition and next frontier of enterprise intelligence.