GOODSFORECAST: When Forecasting Becomes Strategy

Sergey Kotik, Business Development Director of GoodsForecast, explains how Zhuravlev’s mathematical school became the basis for a modern IT product, why manufacturing companies need to forecast demand, and why planning automation is now not just support, but a competitive edge.

 

What does GoodsForecast do? What kind of company is it?

Our company develops and implements business planning solutions. We primarily work with manufacturing companies, but we also carry out projects in heavy industry and retail.

We are a highly “focused” company in the sense that we don’t try to spread ourselves thin. We truly focus only on planning-related solutions: from demand forecasting—when a manufacturer or retailer needs to know which products and in what quantities will be in demand—to the procurement or production of the required amounts.

Then comes planning the production process itself, including scheduling—down to the exact hour, and sometimes even minute: which products, on which lines, and in what sequence to manufacture to achieve maximum efficiency.

Our clients are companies with annual revenues starting from 5 billion rubles.

 

Tell us about the creation of GoodsForecast. How did it all start?

Our story is unusual for a startup. In 2000, a company called Forexis was established at the Computing Center of the Russian Academy of Sciences. It was founded by an MIPT graduate and his professors. In addition to MIPT, they also have and still run the Department of Mathematical Forecasting Methods at Moscow State University’s Faculty of Computational Mathematics and Cybernetics.

The company was created to commercialize scientific developments from Yuri Ivanovich Zhuravlev’s school—a renowned mathematician and member of the Russian Academy of Sciences. Back in the 1950s and 60s in the Soviet Union, he began research in the field that today is called artificial intelligence and machine learning.

From the very beginning, Forexis worked on custom projects for government agencies and businesses, involving data analytics. At that time, common terms like “artificial intelligence” and “machine learning” didn’t exist yet, but many of the tasks were already close to what we understand them to be now.

The company developed various business lines and eventually became a sort of mini-incubator: successful lines would "grow up" and be spun off into separate companies. Forexis still operates in this format today.

In 2004, we had the first project related to demand forecasting, with the supermarket chain Perekrestok (which wasn’t yet part of X5 at the time). That's when the demand forecasting and inventory management business line started within Forexis. A little later, production planning was added to this.

This direction developed step by step, and in 2012 we decided to spin it off as a separate company, GoodsForecast. We joined Skolkovo, and since 2013, the company has existed as an independent legal entity. Nevertheless, we maintain close ties with the scientific community.

Originally, GoodsForecast was a subsidiary of Forexis, but later the shares were distributed amongst individuals. In 2022, KAMA FLOW came on board as an investor.

Basically, we didn’t follow the classic startup path—we’re more like a spin-off. We spun off from Forexis, then redistributed ownership. Before KAMA FLOW, we didn’t have external investments; we lived off what the company earned. At times, shareholders helped out, and we also received grants.

The team from KAMA FLOW reached out to Forexis themselves—they were interested in several areas. They eventually focused on GoodsForecast. We prepared the deal for over a year and finalized everything in May 2022. KAMA FLOW is actively involved in the company at the board of directors level. Their help is real: for example, in setting up proper financial and management reporting. KAMA FLOW also has plenty of market contacts—thanks to them, we’ve been able to reach new clients and partners.

 

So how do your technologies help a specific company? Can you give an example? Suppose I have a confectionery factory, and I keep having a ton of unsold candies left over. Is this exactly the kind of problem you solve?

Yes, exactly. With correct, more accurate demand forecasting, you can build effective sales plans. Of course, it’s not just a single number, like “How much will we sell this year, and what will the revenue be?” Large companies usually have geographical breakdowns—say, if they operate in several federal districts. There’s also a product hierarchy—say, barberry candies, toffees, hard candies, and so on.

There’s also sales channel breakdown: in some places, the product is sold directly to major chains, while in others—especially if the product has a long shelf life—a distributor buys and resells it to small retail stores in different regions.

Across all these areas—geography, product hierarchy, channels—companies need to understand exactly which product to produce, in what quantity, and when. So you don’t end up with the infamous ton of unsold candies, and so you don’t have a situation where demand for barberry is high, but you didn’t produce enough.

That’s exactly where our software helps optimize planning. It builds in forecasting models, optimization models, machine learning, and artificial intelligence.

 

But a factory might already have a team or department doing this: making regular reports, quarterly or yearly. They might notice that barberry candy volume should increase, while, say, production of salted caramel should be stopped. Previously, all this was done by hand, right? Does your product make the process faster, or add a totally new level of analysis?

If you only have 2-3 items—hard candies, barberry, whatever—and you’re selling in 10 stores in one city, one staff member with Excel can manage demand analysis, build a sales plan, manage inventory, and even plan production. But usually things are much more complicated: we’re talking about tens, hundreds, or even thousands of SKUs, across different regions, and different sales channels—and that’s where manual methods just aren’t enough.

Here you need process automation and the ability, using configured mathematical models, to

* calculate more accurate forecasts,

* properly determine required production volumes,

* plan production line schedules

—basically, enable companies to grow further.

Because processing this volume of information by hand just keeps getting harder. Sure, you can stick with Excel, but at some point the amount of data needed for effective planning just outgrows what a spreadsheet can handle. That’s when a company is really ready to consider specialized systems.

 

When a company comes to you and gives you all their data, what are the typical weak spots? How do you approach those? Do you program for certain risks?

These are long projects, delivered by a joint team—client-side and our side as the software provider. It’s not like they come, pay, and a month later everything is ready and working—definitely not. It’s always a joint effort.

One of the main challenges is the quality of the client’s data. In every project, we come across this problem in one form or another. We solve it together, step by step, working closely with the client's team.

Integrating with the client's IT solutions and embedding our product into their digital landscape requires high-quality data at the input. Only then can the system operate correctly, calculating accurate forecasts and forming realistic production plans.

Business processes are equally important. Properly built planning processes are another key to success. Often, we help clients at this stage, because as the saying goes: you can automate a mess, but it won’t get you the results you need.

Automation is effective only when the processes are already well-designed and logical. It enables companies to build truly functional planning scenarios. In recent years, there’s a new term—integrated planning —which is a process where all plans within a company are interconnected into a single system. Not where the sales department comes up with one plan, and production something completely different, and then it turns out their numbers don’t “talk” to each other at all.

 

But couldn’t you just take the data and have it analyzed by something like GPT? Or is this all much more complex?

We don’t make the forecasts ourselves—we provide the **tools** for building and adjusting them. Let’s say you’re a planner at a factory. You have a given production volume to meet, set from above. But there are many production lines, each capable of manufacturing various products in different combinations.

You come to work, open our system, and the schedule is already automatically calculated: on line one, from 10 to 12 we make barberry, then one hour for changeover, then from 13:00 to 18:00 we make toffee. On the second line, you’re running hard candy all the time, because of a big order, so resources are shifted there.

There are, of course, many nuances. But the planner can see everything in a visually convenient interface, and if need be, make corrections to the schedule—drag-and-drop blocks, edit the sequence, all with a mouse.

Frequently, data doesn’t fully match reality—for example, a line’s capacity may have dropped, or something unexpected has come up. So a human can quickly tweak what the model has generated. Once the plan is approved, it goes to production and the relevant amount of barberry, toffee, candies, etc. is manufactured. The automated models take into account holidays, seasonality, special demand patterns on weekends—a huge number of factors. And these forecasts get more accurate all the time.

But when we’re forecasting for long horizons, there’s always expert knowledge the model just can’t factor in. That’s why, within the process, experts need to supplement what the model produces, making it more accurate. Proper processes enable companies to plan optimally, cut costs, make additional sales, and increase profits.

 

And in figures—how do metrics change for companies after implementing your technology?

We can only speak in relative numbers, as we’re usually not allowed to disclose absolutes. For example, in one project, forecasting accuracy improved by 15–20%. Revenue increased by several percent, thanks to better planning. Profitability in manufacturing companies rose by an average of 5% due to precise volume and schedule planning. This brings a direct economic benefit.

Such projects have clear ROI, which can often be calculated right at the start. Thanks to these solutions, companies greatly increase their efficiency.

 

What are your plans for the next few years?

According to current estimates—including those from the planning solutions market—we became the market leader by the end of 2024. Our plan is to keep developing the product's features, because there is ongoing demand and growth potential. We also want to do as many implementations as possible, both independently and with partners. Our goal is to help Russian companies build effective processes.

GoodsForecast continues to grow—both in revenue and client numbers. We’re maintaining positive momentum and hope this trend will only grow stronger.