Introducing artificial intelligence (AI) models into production and business processes in a ‘raw’ form, merely to tick a box, does not magically improve efficiency. In most cases, AI solutions need to be tailored to specific tasks after deployment. Russian businesses are beginning to recognise this. However, not all companies are able to develop and refine AI technologies in-house; in practice, this is largely limited to large firms with greater financial resources than small and medium-sized companies, according to a new study by the Higher School of Economics (HSE).
Russian business has found itself at a digital crossroads. On the one hand, many large and mid-sized companies have set up dedicated AI competence centres, which have promised management that the new technology can solve virtually any problem: all that is required is to plug a generative model into business processes and reap the benefits.
In practice, however, such digital innovations, introduced unthinkingly in pursuit of trends or simply to tick a box, often end up diverting both financial and human resources, while the prospects for genuine efficiency gains become even more uncertain than before.
On the other hand, many Russian companies are beginning to realise that generative artificial intelligence is, to borrow the ironic terminology of IT specialists, far from a ‘silver bullet’: it is not a universal, one-size-fits-all solution. AI models need to be tailored to specific business tasks and integrated with other technologies.
This dilemma was discussed by representatives of the IT sector at a conference in Moscow organised by Russoft, the Russian association of software developers.
In particular, as Oleg Sazhin, adviser to the CEO of Content AI, noted, a ‘bare’ AI model on its own is not capable of managing processes: ‘It needs a supporting layer of modules that perform various operations.’ This means that platforms capable of effectively combining generative models with classical algorithms are likely to be in growing demand. IT experts now view the implementation of such integrated solutions as the most economically viable and beneficial approach for businesses, especially in conditions of constrained resources, from personnel to financing.
Sazhin explained that AI is an innovative and potentially effective ‘engine’, but, figuratively speaking, it has neither wheels, nor a steering wheel, nor seats: it is not a finished product, not a car.
For example, in document-processing tasks, generative AI models can be used to recognise documents, extract relevant information and classify it. However, for the process to work efficiently, such capabilities alone are not enough. Additional functions could include, for instance, autonomous control of scanners or the ability to access and open email archives.
However, as a new study by the HSE Institute for Statistical Studies and Economics of Knowledge shows, based on data from large and mid-sized companies, with small businesses outside the scope of the analysis, refining implemented AI solutions is not merely a strategy that companies need to grow into conceptually; it is, quite literally, a luxury not available to everyone.
As the HSE Institute for Statistical Studies and Economics of Knowledge reported on Wednesday, 18 March, around 5% of Russian organisations currently use AI solutions in their operations.
However, this figure varies significantly depending on the size of the business. The larger the company, the more likely it is to be using artificial intelligence, the analysis shows.
Use of specific artificial intelligence technologies by organisations, broken down by company size. Percentage of organisations of each size already using AI-based solutions. Source: HSE
Among large companies with a workforce of 500 or more, the share of companies using AI has nearly reached 15%. By contrast, among firms employing no more than 100 people, only 4% use AI, roughly four times fewer.
Across the sample as a whole, the most in-demand application for businesses is AI-based processing of visual data, including computer vision. However, preferences also vary depending on company size.
A comparison across companies of different sizes shows that large organisations (with more than 500 employees) lead in the adoption of AI for audio processing, including speech recognition and synthesis, used by 47% of companies, as well as in AI-based text processing, with a usage rate of 53%.
By contrast, companies employing no more than 100 people have emerged as clear leaders in the use of AI for visual data processing, with the share of users in this category reaching 71%.
In terms of integration into specific business processes, across the sample as a whole artificial intelligence is most widely used in marketing and sales, and least in logistics and transport
But once again, company size shapes these preferences. Firms with relatively small workforces (up to 100 employees) are more likely than others to prioritise the use of AI in human resources management, with 58% using such solutions.
Large companies, meanwhile, show a higher level of interest than other companies in using AI technologies for security (37% of such firms) and for logistics and transport-related tasks (23%).
Across the sample as a whole, the most common ways for companies to obtain the AI solutions they need are either commissioning development from external providers or purchasing ready-to-use commercial software. The smaller the company, the more likely it is to rely on these options.
Large organisations, meanwhile, lead among their peers in the share of companies that develop the AI solutions they need in-house (34%), as well as in the share that modify open-source software themselves (32%) or customise commercially acquired software on their own (23%).
The reason is straightforward. As HSE experts note, the financial capacity of large organisations allows them to develop and refine AI technologies in-house far more actively than smaller firms, enabling, with a well-designed approach to digital transformation, these solutions to be closely tailored to business needs
