We sat down with Gurgen Melkonyan, Jume Founder & CEO, to speak about the practical benefits of machine learning (ML)-based forecasting beyond the FMCG space. Below are key insights for business leaders—from the role of human involvement in forecasting processes to predicting crop emergence.

---

— ML forecasting is most often associated with CPG, retail, or e-commerce. Why did these industries become early adopters of such technologies?

There are several reasons. Global CPG corporations have always been at the forefront of technological innovation. They cultivated a culture of constant business transformation—driven largely by fierce competition: evolve or fall behind. For decades, these companies optimized business processes, implemented advanced ERP, MDM, CRM, and BI systems, structured their data, and built teams capable of executing complex projects. Naturally, they were among the first to pilot AI-based solutions. In our experience, CPG businesses consistently showed the strongest demand for ML forecasting.

E-commerce is a different case. These companies were technology-native from the start. For them, forecasting demand and traffic is mission-critical. However, they often built custom in-house IT systems and had strong internal development teams, which made collaboration with external vendors more difficult.

---

— So is ML forecasting relevant mainly to these industries?

Not at all. ML forecasting delivers value across every sector. Over the past five years, we’ve worked with clients in transport, agriculture, restaurant chains and cafés, furniture manufacturers, classified ad platforms, and fresh produce suppliers—and that’s far from a complete list. Each use case was tailored to a specific need: sales forecasting, order predictions, incident tracking, even forecasting vegetable crop growth.

Sometimes, we worked with multiple data sources; in other projects, a single high-quality dataset was enough to achieve better forecast accuracy.

We’ve tested virtually every ML algorithm available for time series forecasting. We conducted real-world validation on the datasets of over 30 major companies. That experience helped us build what we believe is one of the most advanced ML-powered forecasting tool on the market.

We’ve created a kind of “digital planner” that handles all the routine and over 80% of the technical work done by forecasting specialists: from data collection and historical cleaning to forecast generation, factor analysis, cannibalization calculations, and prescriptive analytics for exception-based management.

And yet the role of human experts remains critical: they guide the technology and make real-time adjustments in response to unpredictable events—whether it’s an urgent sales boost, supply shortage, or special client request.

The platform’s universality means it can be adapted to virtually any company or planning module—from demand forecasting to distribution and supply chain management. Thanks to our pre-configured ML engine, we’ve shortened pilot timelines to 1–2 weeks and full implementations to just 1–3 months.

---

— Can you share examples of unconventional forecasting use cases?

Sure. For one of major regional airport operators, we built a pilot model to address three forecasting challenges:

• passenger flow on current routes, broken down into regular and extra flights,
  
• forecasts for emerging destinations,
  
• seasonal traffic trends.
  

Each of these used unique datasets and modeling approaches—from historical passenger numbers to socio-demographic data. The forecast results helped the operator negotiate with airlines to increase flight frequencies and launch new routes.

In agriculture, we worked with a large vegetable producer to build a model that:

• forecasts daily harvest volumes—helping plan workforce shifts, logistics, and storage, and
  
• predicts crop emergence timing, optimizing production cycles.
  

---

— How do companies measure the effectiveness of ML forecasting?

That’s one of my favorite questions because there’s always a clear, practical answer. Unlike large-scale automation projects, whose benefits are sometimes hard to quantify, ML forecasting provides concrete metrics: forecast accuracy, forecast bias, and the time specialists spend on each forecasting cycle.

Every company we’ve piloted with has its own models for evaluating how forecast accuracy affects core business indicators—service levels, raw material waste, working capital needs. As we often say, forecast is the king, because forecast quality impacts everything: from raw material delivery schedules and production planning to logistics costs, safety stock levels, and profitability.

So, if a major company nowadays successfully completes an ML pilot but decides not to implement it further, it’s usually a sign that operational efficiency is simply not a business priority.

— Are you saying that ML provides better forecasting quality than human experts?

Not exactly. It’s more accurate to say that ML, when used in tandem with domain experts, can work wonders. Today’s economy is defined by constant change — dynamic supply chain restructuring, scenario planning, and re-planning happening literally on a daily or even hourly basis.

In this environment, traditional approaches based solely on human input become a bottleneck. Competition is increasing across all sectors, and the winners are those who can respond quickly to changes and adjust plans in real time, based on real data.

For instance, it takes a person a significant amount of time to recalculate a forecast when new variables emerge. The risk of errors increases accordingly. ML algorithms, on the other hand, can process incoming data within seconds, detect correlations, and quantify the impact of each new factor.

— So the algorithms handle the routine data checks and processing, and the expert verifies the forecast’s quality?

Exactly. ML minimizes the impact of the human factor and subjective judgments — which are often influenced by lack of experience, internal pressure, or simply the physical limits of how much information one person can process.

Instead of spending time on routine checks, business teams can focus on analyzing key deviations and forecast positions. This speeds up decision-making, improves accuracy, and gives companies a competitive edge in an increasingly fast-paced environment.

— What’s the main takeaway from your experience?

ML-based forecasting is not a luxury reserved for high-tech or CPG companies. When configured and used properly, it can bring tangible value to virtually any industry. The key is to overcome outdated mindsets and embrace innovation.

Just look at how quickly tools like ChatGPT have become part of daily workflows — it wasn’t long ago that they seemed experimental, and now we use them for everything. I believe ML will follow the same path, becoming indispensable for companies looking to boost efficiency in today’s competitive landscape.