Research on how corporations adopt AI in 2025, what blocks scaling, and where the technology delivers business impact.
> "AI could be humanity's greatest achievement, but also its last mistake if we do not learn to manage it responsibly" - Stephen Hawking
In 2025, AI, or artificial intelligence, has become an integral part of the digital agenda for most large companies. But behind the headlines and marketing often lies a different reality: implementation difficulties, organizational barriers, unpredictable results, and limited understanding of where AI actually works. At KT.Team, we launched a study to understand how AI is used in practice in CIS and international corporations, which solutions truly deliver results, and which remain at the experimental stage.
The project was launched after working with one of the major logistics companies. Analysis of data from 9 IT services and daily operational analytics revealed common issues: lack of automation, inefficient use of data, difficulties adopting AI tools, and team training. To test our hypotheses and broaden the picture, we interviewed representatives of large companies.
The study is based on interviews with representatives of large corporations in logistics, retail, energy, and other industries. Participants included digital transformation specialists, CIOs, and CDTOs. The interview questions covered: - The current level of AI adoption - Successful and failed cases - Barriers and internal resistance - Strategic plans for 2025 and beyond
The participants included national brands, international holding companies, and technology companies with advanced digital infrastructure. All participants shared not only successes but also challenges - honestly, without embellishment. All companies mentioned are anonymized - this was agreed with the study participants. We respect the trust they placed in us when sharing their experience and follow a confidentiality principle so the focus stays on practices, not brands.
When and why AI became a priority For most study participants, AI emerged on the agenda in 2021-2023. At first, these were local initiatives: pilots in chatbots, analytics, or routine task automation. By 2025, however, most companies were treating AI not as a one-off technology, but as one of the tools for transforming business models.
Who Champions AI Inside the Company The main reasons companies began adopting AI: - Growing employee workload and demand for decision automation - The need to speed up processes, from logistics to customer service - Pressure from competitors and market technology leaders - Tool availability: advances in LLMs, API interfaces, and open-source solutions The most active roles in initiating and advancing AI projects are: - Chief Digital Transformation Officers (CDTOs) - CIOs and Heads of IT - Product teams - In some cases, business executives (CEOs, COOs) when the initiative is strategically important Nevertheless, in many companies AI is still seen as the domain of IT or R&D, not as a business growth tool.
This leads to strategy gaps and the lack of AI integration into key KPIs. Where AI is already being used Based on the interviews, AI is most widespread in the following areas: The tasks AI solves for the areas listed above:
| Application area | Task type |
|---|---|
| Logistics and warehouse operations | Load forecasting, routing, visual inspection |
| Finance | Predictive analytics, data reconciliation, reporting automation |
| Customer Service | Chatbots, answer generation, intelligent request routing |
| HR | Resume analysis, automated candidate assessment |
| Sales and Marketing | Offer personalization, description generation, customer behavior analysis |
Project maturity level Based on the interviews, companies can be roughly divided into 3 groups: 1. Those actively implementing and scaling AI (there are usually dedicated teams, processes, and internal expertise) 2. Those in the experimental phase (there is an MVP, but no consistency, scaling, or business adoption) 3. Those just starting to explore it (there is interest, but no infrastructure, expertise, or business demand) What is slowing development Key barriers: - Lack of a unified AI strategy - Distrust from the business side - Insufficient maturity of internal processes - Difficulties hiring specialists and training teams Despite the overall interest in artificial intelligence, companies are still at different stages of maturity.
Some are already integrating AI into core processes and seeing results, others are limited to pilots without scaling, and still others are only beginning to figure out where to start. At the same time even the most advanced participants still face systemic barriers: lack of a unified strategy, shortage of expertise, weak digital foundation, and distrust from the business side.
This shows that the path to mature AI use runs not only through technology investment, but also through reworking processes, training teams, and building trust in change.
Despite the general caution and the many experimental projects, in a number of companies AI is already delivering tangible results. We have gathered several examples where the technology was not only deployed but also proved its value in an operational or strategic context.
Project - Queue monitoring at retail locations The company implemented an AI model that analyzes shopper flows in real time. The system alerts staff when the checkout area is overloaded, allowing new registers to be opened quickly. This delivered a concrete result - 2% increase in turnover in the tested stores.
Project - AI platform for team support and LLM data analysis The company deployed an internal LLM-based AI platform that serves 70+ teams. The solutions include: - Automated analysis of data in legacy systems - Content generation - Support for call center processes Vision Labs is also actively used for custom AI products. At the same time, the company is working to educate users, both internal and external customers.
This helps temper inflated expectations of the technology and shape realistic use-case scenarios.
Project - Automation of marketing materials processing and response prediction The company uses AI to generate ad copy and visuals, as well as to predict communication performance. By training models on its own data, it automated part of marketers' routine work and shortened campaign launch cycles. Implementation was phased, from an MVP on selected product categories to scaling across key areas.
According to the representative, AI tools save up to 30% of the team’s time and enable faster testing of new hypotheses.
Project - Generation of integrations and documentation in a microservices architecture The company created its own integration bus with an AI module that: - Generates integrations based on prompts - Writes technical documentation in the customer's required format - Automatically handles errors and adapts to the customer's architecture According to the company representative, generation accuracy reaches 98-99% for simple integrations, resource savings of up to 80% of development time.
AI is also used in the MRP system: data is collected from machines, including Soviet-era equipment, and a predictive maintenance failure analytics. This helps avoid downtime and optimize spare parts procurement.
AI is seen today as a powerful tool for transformation, but in practice not every initiative leads to the expected results.
Even in advanced companies, some projects remain at the experimental stage and never find business application. We identified three typical groups of ambiguous cases:1.
Many companies launched MVPs or test scenarios, but did not go any further: -
One major retailer tried to automate the inspection of construction and installation work using AI, but ran into several constraints: unstable data and difficulties verifying results.
A company in the digital services sector ran pilots for text review analysis and speech analytics.
However, the projects were not scaled because there was no sustained business impact and infrastructure constraints. -
One federal retail chain considered using AI to track changes in legislation, but the initiative did not progress because of the lack of an internal IT mandate and strategic framework. 2.
Even when there is interest in AI, internal constraints can block implementation: -
One industrial company operates under strict regulatory constraints: public clouds and LLM services are prohibited, and all data within the organization is classified under a four-tier access system.
This significantly limits the use of external AI tools and requires local solutions. - In another digital solutions company, the main barrier to scaling AI was data quality: unstructured and incomplete information leads to unstable results, especially in content generation and analysis of customer inquiries. 3. Expectations that do not match reality
Sometimes the reason for failure is not the technology, but unrealistic expectations: -
Some participants expected AI to fully replace an expert - for example, by producing a comprehensive report without post-editing or conducting negotiations.
In practice, it requires validation, adjustment, and context-specific tuning. -
Several companies reported false positives or model hallucinations, especially in chats, report generation, and request handling.
As of 2025, most companies are still at early stages of AI adoption, either in pilot phase or in isolated automations.
However, the trends gathered in the study show a shift: AI is increasingly seen as part of the operating model rather than an external innovation. Companies that have already seen tangible results share common traits: - internal business-driven initiative, not only IT; - understanding where AI can be applied with clear value (cost savings, speed, control); - the ability to adapt teams and processes to new tools; - local success stories that help reduce skepticism.
Only 25% of companies are already seeing tangible results from AI - driven by business initiative, clear value, and adapted processes. The rest 75% are still at the experimentation or observation stage. At the same time, broader AI adoption is still held back by: - unprepared infrastructure and data; - fears and distrust, including at the management level; - a lack of methodology for understanding where AI is truly needed.
Even with strong interest in AI, very practical factors still stand in the way of broad adoption. 40% of companies face technical unpreparedness - this includes weak system integration, the absence of Data Governance, and difficulties accessing data.
More 35% of respondents reported perception barriers: distrust from managers, fear of losing control, and concerns about reduced team competence. 25% struggle to choose areas for AI implementation - there is no methodology or criteria that help the business distinguish real value from hype. Until these blockers are removed, moving from experiments to large-scale deployment remains difficult. Solving these issues is the key to leaving the experimental stage.
Even if you are just starting with AI, it is important to act systematically. Below are six steps that help build a resilient strategy, from early pilots to scaling. These recommendations are based on an analysis of the practices of companies that have already achieved results: 1. Define a basic AI strategy Even if there are no mature use cases yet, it is important to define goals, areas for experimentation, and the risk approach. This reduces chaos and helps establish focus. 2.
Define ownership for AI Without a decision-making center, projects remain fragmented. Responsibility may sit with the CDTO, CIO, or a dedicated AI team. 3. Invest in data preparation Without clean, accessible, and documented data, any AI initiative is doomed. The starting point should be prioritizing key data sources. If the data is fragmented, outdated, or stored in hard-to-access systems, even the best algorithm will not deliver results. 4.
Start with practical, high-pain use cases Where there is a tangible pain point (for example, labor-intensive integrations, manual tender review, production monitoring), AI ROI is clear already at the MVP stage. 5. Unite IT and business efforts Only joint work makes it possible to connect AI solutions to real metrics and secure adoption at the business stakeholder level. 6. Involve teams and build a culture of experimentation People fear what they do not understand.
Training, internal advocacy, accessible tools, and quick wins are the foundation of cultural transformation.
Analysis of interviews and research data confirmed that, despite strong interest in artificial intelligence, AI adoption maturity in CIS corporations remains uneven. Only a portion of companies have moved past pilots and begun scaling the technology. The rest are in the experimentation phase or exploring the potential. At the same time successful cases demonstrate real business value - from faster integrations and lower development workload to predictive analytics and production optimization.
However, scaling is constrained by a number of internal barriers: the absence of a strategy, weak integration with business processes, and a shortage of data and expertise. AI is still often viewed as an IT or R&D tool rather than a driver of business-level transformation.
Without a shift in managerial perception, its potential will remain local. To move from isolated pilots to large-scale AI deployment, corporations need to remove internal barriers, first of all at the level of data, processes, and management trust. Without this, even successful cases risk remaining local. Those building a systematic approach - connects AI to business goals, involves key teams, and drives impact at the company level, - are already gaining tangible competitive advantages today.
We thank all study participants for taking part in interviews, for their openness, candid answers, and practical focus, which made it possible to capture not only successful cases but also the real barriers to AI adoption. We also thank the experts who responded to open surveys; their observations from different roles, from CIOs to frontline specialists, helped round out the picture from multiple angles. We also thank everyone who helped prepare, transcribe, analyze, and visualize the data.
We hope the study's findings prove useful to those seeking a path toward mature, pragmatic AI adoption.
