An AI-Oriented Approach to Project Management
Last updated: 13 September 2026
What is an AI-oriented approach to project management?
An AI-oriented approach to project management is a full-lifecycle model in which machine learning, language tools, and optimization sit inside planning, delivery, and control, while a person still owns the decision. It is not a chatbot bolted onto a Gantt chart. It is a claim that today’s tools are too fragmented to count as a method.
Yuliia Matvieieva, Yuliia Opanasiuk, and Dmytro Kurhankskyi, all at the Academic and Research Institute of Business, Economics and Management of Sumy State University, make that claim in issue 12 of Ukrainian Economic Journal (Український економічний часопис), pages 87 to 94, published 2 March 2026. The Ukrainian title is AI-орієнтований підхід до управління проєктами. The research article is on HAL. The same research article is on Academia.edu. DOI 10.32782/2786-8273/2026-12-14. License: CC BY 4.0. p-ISSN 2786-8273, e-ISSN 2786-8281.
This explainer is in English. The source paper is in Ukrainian, with an English abstract. The argument travels cleanly: AI tools for forecasting, risk, and resources exist, but they stay on isolated tasks. There is a methodological gap between what the tools can do and how project management is actually structured. The authors try to close that gap with a three-layer sketch, not with a new product.
"AI should amplify the manager’s judgment. It should not take the chair."
Why do current AI project tools feel incomplete?
Current AI project tools feel incomplete because they automate a slice of work, then stop. Traditional methods already struggle when projects are multidisciplinary, the environment is noisy, and decisions have to be both fast and accurate. Expert judgment, linear plans, and manual reporting run out of road. AI can score risk, shuffle resources, and simulate scenarios. In practice, the authors say, that help is still limited to planning, cost control, or a risk forecast. A whole-lifecycle model is still being born.
They also cite a second global survey finding that about 62 percent of managers need to expand their skills to stay effective in an AI-heavy setting. That number is not a software-engineering statistic. It is a management statistic. It is why the paper spends as much time on AI literacy and ethics as it does on algorithms. A tool that a manager cannot interpret is not a decision aid. It is a second source of noise.
The literature they survey already covers PM2030, ChatGPT use by project managers, risk methods, and software-project AI. The unresolved problem, in their reading, is the missing integrated model. Academic papers look at adoption. Consultants look at process efficiency. PMI and IPMA look at competence. ISO looks at governance. IEEE looks at explainability. Agile and DevOps look at delivery. Almost none of those families cover the full cycle with AI as a first-class method. That is Table 1 in the paper, compressed into one complaint.
What did the Scopus and VOSviewer map show?
The Scopus and VOSviewer map showed a field that is growing fast, clustered around a few themes, and geographically uneven. The authors searched Scopus on “Artificial Intelligence” and “Project Management,” then built a co-occurrence network. The two biggest nodes are those two phrases. Around them sit four color clusters.
The four clusters in Figure 1
- → Red: decision support, optimization, algorithms, decision-oriented systems
- → Blue: information management, risk, predictive analytics, digital transformation
- → Green: project management, resources, learning systems, machine learning, big data
- → Yellow: medical and clinical projects, diagnosis, pilot studies
Publication counts tell the time story. From 2003 to 2018 the yearly total sat roughly between 100 and 250 papers. After 2019 the line bends up. After 2020 it steepens. About 700 papers in 2023, more than 1,100 in 2024, about 1,600 in 2025. The authors list four drivers: breakthroughs in machine learning, NLP, and generative models; digital tools in business processes; AI as a competitiveness factor; and AI crossing into management, economics, and work psychology. If the slope holds, they say the yearly total could pass 2,000 by 2026 or 2027.
Geography is less even. The United States leads their Scopus slice with 1,473 papers. China has 1,051. Then the United Kingdom (663), Germany (643), India (627), Italy (614), Spain (421), France (364), Canada (317). Ukraine is on the map with 88. The authors treat that last number as a signal, not a slight: the topic is live in a transition economy, and still far from the volume of the digital-technology leaders. For a Ukrainian journal, putting Ukraine on the same figure as the United States is a research-policy point. Capacity and indexing both matter.
The map is not a model. It is evidence that the conversation is real, that it is spreading, and that it still lacks a single frame that ties lifecycle, management functions, and tool classes together. That is the job of the rest of the paper.
How did the authors build the argument?
The authors built the argument with four methods stacked on purpose: a synthesis of existing work, a systems view of the project as a socio-technical setup, a bibliometric and network pass on Scopus, and conceptual modeling. That mix is why the paper can say both “the literature is growing” and “the model is still missing” without pretending those are the same finding.
The comparative pass is the least flashy part and the most useful. They line up academic databases, PMI and IPMA competence frames, ISO 21500 and 21504, the ISO/IEC AI series, IEEE ethics standards, Agile and DevOps guides, consultant reports from firms such as Deloitte and McKinsey, vendor white papers, EU-style AI policy, and innovation-management articles. Academic work covers planning and risk with machine learning, NLP, RPA, and predictive analytics, then stops short of a joined-up model. Consulting work treats AI as a way to cut cost in operations, not as a project-management method. PMI’s PMBOK and Agile material cover the lifecycle and mention a tools ecosystem, without a deep AI layer. IPMA’s ICB 4.0 is starting to talk about AI competence and has not defined AI literacy. ISO project-governance texts allow predictive modelling and still do not treat AI as a methodological category. IEEE work on explainable and responsible AI is strong on governance and weak on the project cycle. Agile and DevOps papers are strong on delivery and weak on classic PMBOK functions. Vendor papers sell schedulers and Copilot-style assistants. EU policy talks about risk classification. None of those families, on their own, describe an AI-oriented lifecycle.
That table is the warrant for a new sketch. If every source family already solved the problem, a three-layer diagram would be decoration. Because each family solves a slice, the authors argue for a frame that can hold slices together: phases, functions, and tool classes, under a manager who can still refuse a recommendation. The conceptual-modeling step then draws those three dimensions on one page so later empirical work has something to test, rather than another slogan about “AI in PM.”
They also treat the project as a socio-technical system. Artificial intelligence is not an add-on sitting next to a Gantt chart. It is inside the management process, the formal functions, and the life of the project. That is why AI literacy shows up as a competence, next to critical thinking and ethical responsibility, instead of as a training afterthought. A manager who cannot read a model output cannot keep the human-centered rule the paper insists on.
How is the three-layer model built?
The three-layer model is built by lining up the project lifecycle, the classic functions of management, and classes of AI tools, then putting a human manager over the stack. Figure 4 in the paper is a box diagram. The outer box is the project environment: uncertainty, risks, constraints, stakeholders. Inside that sits human-centered management: the project manager as the decision center, with AI literacy, critical thinking, ethics, and control. Under that sits the integrated AI-oriented PM model with three interacting parts.
Layer I is the lifecycle. Initiation, planning, execution, monitoring and control, closure. On each phase the manager still has specific tasks. AI can support or reshape those tasks. The point is dynamic use across the whole cycle, not a plugin at planning time.
Layer II is management functions. Planning becomes scenario-based and predictive instead of a single linear path. Organization becomes adaptive and optimization-aware. Motivation becomes data-informed team support. Control becomes continuous AI monitoring of streams, not a weekly spreadsheet. Coordination becomes live recommendations for how people interact. The manager does not leave. The functions change shape.
Layer III is tool classes. Machine learning for time and cost forecasts. NLP for documents, reports, and communication. Optimization algorithms for resources and schedules. Generative models for scenarios and alternatives. Explainable and ethical AI for transparency and trust. Each class is supposed to map onto functions and phases, so a team does not buy five disconnected bots.
| Source family | What it covers | What it misses |
|---|---|---|
| Scopus and WoS papers | Planning, risk, forecasting | No integrated model, thin empirical base |
| PMI, IPMA, ISO, IEEE | Lifecycle, competence, governance, ethics | AI is a side note, not a method category |
| Agile, DevOps, vendor papers | Delivery, scheduling, insights | Product or tool focus, not PMBOK-style method |
The intended outputs of the model are listed at the bottom of the figure: better management efficiency, more predictive decisions, faster adaptation to change and risk, less fragmented AI adoption, and a stronger manager rather than a replaced one. That last clause is the paper’s ethical center. As systems get more autonomous, the risk of hidden errors and bias goes up. Literacy and the right to overrule are not decorations. They are how the model stays a management model.
What should a practitioner take from this?
A practitioner should stop treating “we bought an AI scheduler” as an AI-oriented method. Buy the scheduler if it helps. Then ask whether planning, control, and coordination also changed, and whether someone on the team can explain a bad recommendation. If the answer is no, you have a tool. You do not have the model this paper is arguing for.
The paper is conceptual. It does not report a field trial of the three-layer sketch. That is a real limit, and the authors say so: the work is a theoretical base for later empirical tests and for practical methods in complex environments. If you need a before-and-after cost chart, look at other articles on this site that review applied PM systems. If you need a way to argue that isolated bots are not a strategy, this is the clearer text.
Use this paper when
You are designing a PM operating model, writing competence requirements, or explaining why a tool catalog is not a lifecycle.
Do not use it as
A vendor bake-off, a software architecture, or proof that any one algorithm raises on-time delivery.
A compact checklist from the model
• Map every AI tool to a lifecycle phase and a management function, or drop it
• Keep a named human decision-maker for high-stakes calls
• Treat AI literacy and ethics as skills, next to scheduling and budgeting
• Watch the research boom without confusing paper counts with working practice
Read next to the English-language software-PM reviews on this site, the Ukrainian paper is the one that most clearly says “the missing object is a model, not another dashboard.” That is a useful sentence in a budget meeting. It is also a research agenda: someone still has to test the three layers on a live portfolio, in Ukraine and elsewhere, with the messy data the bibliometric map does not contain.
Until that test exists, the honest use of this work is as a design constraint. If your AI plan only covers forecasting, you are in the fragment the authors documented. If it covers the cycle, the functions, the tool classes, and a person who can still say no, you are at least aiming at the thing they named.
One more practical note from the trend section. The authors expect the next wave of papers to talk about autonomous decision systems, AI-driven governance, cognitive risk analytics, human-resource management based on predictive models, and ethical-legal frames for AI-plus-manager work. That list is a hiring and training hint as much as a research hint. If your PMO is still staffing only for schedule control, the competence gap they cited at 62 percent will show up as delayed decisions, not as a missing license. Train for interpretation. Keep the chair. Measure whether the three layers actually talk to each other on a live project, not only in a slide.
Frequently Asked Questions
What is an AI-oriented approach to project management?
It is a model that places AI across the whole project lifecycle, the classic management functions, and classes of AI tools, while keeping the project manager as the decision-maker. Matvieieva, Opanasiuk, and Kurhankskyi argue that today’s tools stay stuck on isolated tasks such as forecasting or cost control.
What gap did the Scopus and VOSviewer analysis find?
Publication volume on AI and project management rose from a few hundred papers a year before 2019 to about 1,600 in 2025. The research clusters around decision support, information management, project processes, and some medical projects. No cluster is a full-lifecycle integrated model.
Does this model replace the project manager?
No. The authors use a human-centered rule: AI is a cognitive amplifier, not an autonomous substitute. The manager still owns the call. AI literacy, critical thinking, and ethics are named as required skills, not optional extras.
Where was the paper published?
It appeared in Ukrainian Economic Journal (Український економічний часопис), issue 12, 2026, pages 87 to 94, under a CC BY 4.0 license. The authors are based at Sumy State University. DOI 10.32782/2786-8273/2026-12-14.