According to McKinsey, 89% of companies worldwide are already using AI, but only 37% are seeing tangible financial returns, and just 6% are generating significant value, equivalent to more than 5% of EBIT. In other words, while nearly everyone has access to the technology, relatively few are turning it into measurable business results.
Simply introducing AI into business processes does not automatically translate into cost savings. Real savings come when companies set clear objectives and employees understand exactly which tasks AI can accelerate and which aspects of their work are suitable for automation. The most effective approach is to introduce AI gradually, starting with specific, well-defined, time-consuming tasks that can be partially delegated to AI agents.
HR offers plenty of opportunities for this approach, from sourcing and screening candidates, analyzing CVs and drafting job descriptions and correspondence to onboarding, training, employee assessments, HR analytics, and automating responses to routine enquiries.
At IT_ONE, the decision to integrate AI into HR processes was driven by practical business needs. The company wanted to speed up recruitment for highly specialized positions, reduce the time and cost of developing training materials, and improve the quality of technical onboarding. The team began by identifying the most time-consuming processes, determining which could benefit from automation and establishing where human involvement remained essential. Over six months, these efforts saved more than 4,000 working hours. During the same period, the company also acquired four new clients.
Case Study 1: Developing a training course in two weeks instead of three months
Employee training offers some of the greatest opportunities for reducing costs. In 2025, developing a single internship program cost the company approximately 200,000 rubles ($2,477). Experts handled the entire process, from designing the curriculum and preparing training materials to creating assessments and other resources, taking each program from the initial concept to the finished product.
AI has fundamentally changed this workflow. The process now begins with an analysis of current job requirements to create a profile of the ideal candidate, identifying both the skills they need from the outset and those they will need to develop. Based on this profile, AI generates a preliminary curriculum. Once the curriculum has been reviewed, the system produces drafts of detailed training materials, tests, guides and presentations.
Subject-matter experts then review and refine the content, removing unnecessary information, and preparing the materials for use. Instead of spending time creating everything from scratch, experts can concentrate on what they do best: evaluating, validating and improving the quality of the content. The routine work of producing initial drafts is left to AI. This approach has already been applied to more than 20 projects, including training programs in C#, Java, Python, DevOps and Data Engineering.
As a result, the average time needed to develop a course has fallen from three months to just two weeks, while the labor required has dropped from 702 hours to 15 hours. AI now handles 80% of the workload, compared with none previously, reducing the experts’ share from 100% to 20%. The project has delivered a return on investment (ROI) of 51%.
Case Study 2: Launching a new department
A company recently established a dedicated technical assessment unit responsible for organizing technical interviews and evaluations. To ensure the new unit operated effectively, its workflows and procedures needed to be developed and documented. These included formalizing the interview process, defining the bonus structure for technical experts, and establishing procedures for creating and updating assessment matrices.
AI helped facilitate the drafting of these documents. Staff uploaded all the information available to them, including informal descriptions of existing processes as well as transcripts of audio and video recordings of meetings discussing the work of the new department. AI generated initial drafts, which the team then refined manually. It also suggested potential performance metrics and provided templates for assessment matrices, checklists, and interview questions tailored to specific projects. As a result, within just a few months, the new department was able to document most of its processes and establish corresponding performance metrics.
Integrating AI reduced the time required to complete individual tasks by a factor of at least two to three. Overall, this has translated into monthly savings of approximately 30 hours and more than 20,000 rubles.
Case Study 3: Analyzing job vacancies for recruitment
Analyzing job vacancies is often a major bottleneck in recruiting for complex technical roles. Before AI was introduced, the process could take several hours. Recruiters would arrange meetings with technical experts or delivery managers to review job requirements and clarify details. Even after these discussions, only technical experts themselves could assess incoming resumes.
With AI, the process has become significantly faster. Recruiters upload a job description to an AI model and engage in a dialogue to clarify the requirements, and also use an AI agent to research emerging technologies, review solutions to technical assignments, and analyze anonymized sample resumes. This enables recruiters to assess incoming resumes independently and compile an initial shortlist without having to seek an assessment from a technical expert. Analyzing a job opening now takes just 30–60 minutes, compared with the previous 2–4 hours. The process has also been further enhanced by providing the recruitment team with technical training. Together, these measures save the company more than 100 hours and over 150,000 rubles per month.
Why AI requires strict oversight
This type of automation also has its drawbacks. The main risk is that employees may pass along AI-generated content without checking it. When that happens, the time saved at one stage of the process can quickly turn into additional work for a manager and lead to errors.
A second, less obvious problem is the tendency to assess AI-generated output in the same way as the work of a human colleague. For instance, if a high-performing employee writes a well-crafted opening to a document, a manager might reasonably expect the same level of quality throughout. The same assumption cannot be made about AI: it can make a mistake at any point, no matter how promising the opening may be. In this regard, AI can act like a “Trojan horse”: an error may remain concealed behind an otherwise convincing façade.
Both issues can be addressed through clear accountability and effective oversight. A key principle of AI implementation is to define responsibilities from the outset, establish checkpoints, and introduce an appropriate incentive structure for employees. The person who submits a document remains responsible for its contents, even if a substantial portion of the text was generated by AI. Nor can the use of a neural network, in itself, be accepted as an excuse for poor-quality work.
By Ksenia Manskova, Head of Resource Management, IT_ONE