The expert assessed the effectiveness of AI in industry and retail

Artificial intelligence (AI) already brings measurable economic benefits in industry and retail, but payback is achieved only when technologies are integrated into specific business processes and solve applied problems. Sergey Taran, Managing Director of LANSOFT, told Izvestia on May 20.

"AI pays off where it is embedded in a process with a clear economy, rather than being implemented as technology for technology's sake. Demand forecasting and inventory management remain one of the most stable scenarios in retail. Using algorithms allows you to take into account seasonal fluctuations, promotions and external factors, reducing the shortage of goods on shelves and reducing write—offs," the expert emphasized.

Taran called personalization and dynamic pricing another effective direction, which can increase conversion and average receipt. However, such solutions only work if high-quality and large-scale customer data is available.

According to the expert, predictive equipment maintenance systems and computer vision provide the most noticeable economic effect in industry. Predictive analytics allows you to switch from routine repairs to maintenance based on the actual condition of the equipment, reducing the risk of accidents and unplanned downtime. Computer vision, in turn, is actively used for quality control on production lines, especially in cases where defects can be clearly formalized.

Neural network business: the AI market in Russia is growing faster than the rest of the IT sector

However, domestic developers are still lagging behind Western ones — why is this happening

In addition, AI is used to optimize logistics and production planning, but here the result directly depends on the maturity of internal processes and the quality of data.

The expert identified three main reasons why companies lose money when implementing AI. The first is the lack of a clearly defined business objective and clear performance metrics. The second is underestimating the cost of data preparation, which often turns out to be the most expensive stage of implementation. The third is attempts to scale the solution without prior pilot testing.

According to Taran, the two- to three-month short pilot model remains the most effective in the industry, after which a scaling decision is made. A separate risk is associated with inflated business expectations. According to the expert, artificial intelligence shows high efficiency in tasks with repeatable scenarios and large amounts of data, but it does not work well in unique and poorly formalized processes. In such cases, AI does not replace the expert, but remains an auxiliary tool, the expert concluded.

On April 10, Russian President Vladimir Putin said that AI technologies are developing rapidly. Neural networks, along with digital platforms and autonomous systems, form a fundamentally different face of the economy, public relations, the social sphere, education, healthcare, logistics and industry, defense and security. Putin added that only if there are AI language models of their own, it is possible to guarantee the state's defense capability.