Abstract
Summary
The global services industry increasingly adopts software for intelligent process automation, including robotic process automation, low-code and no-code development solutions that are enhanced by artificial intelligence (AI) elements, as well as generative and agentic AI tools. While market growth forecasts are optimistic, organizations face significant adoption challenges. Through several years of field research across 33 European organizations, this article identifies 70 key automation challenges and provides analysis of the most critical. Practice-oriented research maps these challenges to specific project phases and organizational perspectives, at the same time providing actionable mitigation strategies. The article reveals how organizational context (size, industry, regulatory environment, culture) shapes automation outcomes and offers tailored guidance on issues such as sourcing external expertise versus building internal capabilities, and governing decentralized development models. This practical guide serves organizations currently automating processes, or planning to do so, helping them navigate complexity for more effective digital transformation.
In today's rapidly evolving technology landscape, intelligent process automation has become a strategic priority across multiple industries, driven by the emergence of deterministic technologies such as robotic process automation (RPA), as well as probabilistic intelligent systems powered by artificial intelligence (AI), including generative and agentic AI capable of operating user interfaces and handling unstructured data. 1 The accelerating adoption of software solutions for intelligent process automation reflects the promise to streamline operations, increase efficiency, and foster innovation in businesses and non-governmental organizations alike. 2 With the global market of intelligent process automation exceeding $3 billion, 3 organizations are rapidly adopting these technologies to transform operations. However, despite this optimistic outlook, our research reveals significant challenges accompanying process automation initiatives. Through our field research with 33 European organizations that lasted over 3.5 years, we identified 70 distinct challenges that can hinder automation adoption. We focus on and analyze the 37 most critical of these challenges, providing practitioners with actionable insights to navigate the complex automation landscape. These technologies range from RPA for automating repetitive tasks to AI-powered systems capable of handling complex, judgment-based processes, as well as increasingly autonomous AI agents that can orchestrate multi-step workflows across applications. 4
Despite the widespread integration of automation technologies with legacy systems, the road to their effective use, in which not only mundane and repetitive tasks are automated, is often affected by risks and issues. Key industry research agencies such as Gartner, Forrester, McKinsey, and EY report not only strong growth but also significant hurdles that organizations face during implementation. 5 Wrong approaches and management of automation may lead to what has been termed as “re-manualisation” 6 where processes revert to manual operations due to failures of automatic processing, highlighting a critical gap in the execution of digital strategies. 7 Research shows that governance challenges, including shadow IT and technical debt, frequently emerge when organizations rush to implement automation without proper evaluation and execution frameworks. 8 These risks are amplified as low-code and no-code solutions, as well as AI agents enabling more decentralized development, raising new concerns about how to govern “build versus buy” decisions, internal capability development, and long-term maintainability.
Our intended research contribution was to: (1) systematically identify and document key challenges in process automation implementation based on field research; (2) analyze the most critical challenges with practical solutions and mitigation strategies for each; and (3) provide a systematic lens that maps these challenges to specific project phases, enabling practitioners to anticipate and proactively address issues throughout the automation life cycle. Unlike existing theoretical frameworks, our practice-oriented article provides hands-on guidance grounded in real-world experiences to empower all stakeholders involved in the digital transformation journey to navigate these complexities more effectively. Hence, it is targeted at chief operations officers (COOs), chief technology officers (CTOs), department managers, process improvement and excellence managers, process owners, analysts, developers, citizen developers, and all stakeholders involved in process automation initiatives.
Research Approach
This work summarizes the outcomes of multiple data collections undertaken across a period of 3.5 years, between 2020–2023, conducted by lead author Damian Kedziora, who worked for over 10 years in the industry, including over five years at intelligent automation consultancies, and then transitioned to academia with a focus on process automation software. Over this period, Kedziora observed the evolution from traditional RPA toward intelligent automation solutions, including the early adoption of low-code and no-code platforms and, in the later stages, emerging generative and agentic AI tools. Over this timespan, more than 200 hours of data collection actions (i.e., interviews, observation notes, presentational material, etc.) were conducted focusing on process automation risks. Data collection methods included semi-structured interviews and observations, which allowed for an in-depth exploration of participant experiences, as well as open reflection questionnaires administered through surveying tools. 9 The dataset comprised interview transcripts, observation notes, and operational documents, which were triangulated to ensure validity. Data collection in each instance continued until saturation was reached, indicating that further participants would not provide substantially new insights. 10
In total, 33 organizations of different sizes, ranging from 11 to 30,000 employees, originating from the European Union (EU) took part in the study (see Supplemental Table 1 for details), including seven companies that acknowledged their employees being publicly mentioned in the research. Interviewee quotes have been edited for grammar and clarity while preserving the original meaning and intent of the statements (see Appendix for original citations of quotes). Organizations represented diverse sectors including banking, manufacturing, healthcare, and technology services. The companies included in the study were identified with purposive sampling based on their active engagement with process automation technologies. Many organizations were sourced through Damian Keziora's professional engagements in various companies and consultancies, where data was collected either during in-house projects or through interactions with clients that agreed on sharing their data. Additional organizations were recruited through his network, specifically for data collection to broaden the empirical foundation, aligning with the study's practice-oriented focus. 11 This approach ensured access to organizations with substantive automation experience and willingness to share both successes and failures. The roles of employees who took part in our study included Service Manager, Senior Developer, Department Manager, Head of Group Robotics, Global Head of Intelligent Automation, IT Service Manager, CEO, Product Manager, Operations Manager, IT Manager, and Head of Consultancy, Process Continuous Improvement Driver, and Head of Digital Transformation.
The collected data underwent systematic analysis by the research team using Braun and Clarke's 12 inductive thematic and reflexive approach. The analysis involved several iterative stages: (1) familiarization with the data through repeated reading of interview transcripts, observation notes, and materials; (2) initial coding to identify patterns and key features of the data; (3) grouping of codes into broader themes that reflected recurring concepts and issues; (4) iterative refinement of themes through collaborative discussions among the authors to ensure that interpretations were grounded in the data; and (5) critical reflexivity to minimize bias, supported by the team's combined industry and academic expertise. Selection of the 37 key challenges from the initial 70 identified issues was based on three criteria: frequency of occurrence across organizations (appearing in at least three companies), severity of business impact as assessed by respondents, and consensus among all three authors regarding strategic importance. Such process supported the objectivity of our findings and was additionally strengthened by the industry experience of Joanna Kedziora, who spent over 11 years at top multinational organizations of the Nordic banking industry. Her expertise supported the team with critical validation of findings and ensured practical relevance for both RPA initiatives, as well as low-code and AI-enhanced automation projects.
Findings and Discussion
These findings, both the challenges and risks encountered during process automation initiatives, focus specifically on challenges unique to process automation implementations, offering actionable insights for stakeholders involved in digital transformation efforts. The framework includes: 13 requirements, analysis, design, coding, testing, implementation, and maintenance, as well as the “entire DevOps cycle” if the issue occurs along the whole initiative.
Next, we discuss the collected challenges together with their triggers (background and antecedents), business impact (implications), proposed solutions, and project phases where the problems can occur, as well as selected illustrative quotes from interviewees. We provide perspectives with the following structure: 1) strategic decision making and financial management, 2) governance and competence building, 3) technical setup, 4) process documentation and management, 5) performance measurement and road mapping, and 6) human resource management. We pay particular attention to how commoditized RPA tools, low-code and no-code platforms, as well as generative and agentic AI reshape both implementation and post-implementation challenges, including the strategic “buy versus build” decision and the long-term governance of decentralized development. Building on these perspectives, this article connects the identified challenges and their implications to existing literature, theoretical frameworks, and the authors’ practical expertise across the full automation life cycle, from initial scoping and design to post-deployment optimization and workforce adoption.
Strategic Decision-Making and Financial Management
Entreprises with fewer than 500 employees (specifically companies 6, 7, 9, 17, 18, 19, 22, 32, 33) are particularly susceptible to trend-following behavior due to limited internal expertise for due diligence, while larger enterprises (>10,000 employees) faced this challenge primarily at departmental levels where middle management felt pressure to act innovatively. Hence, the significant challenge organizations face is succumbing to trends rather than making decisions on automating processes based on thorough evaluation. It is manifested differently across organizational contexts from their size perspective: smaller firms often lack dedicated evaluation teams, while larger organizations struggle with competing departmental interests and complex approval processes. Such tendencies align with the concept of "technology fashion" or "bandwagon effects" 14 which describe how organizations imitate peers to appear progressive, leading to the adoption of innovations without thorough evaluation that may result in misaligned investments and underutilization of technology. In our cases, this bandwagon behavior was particularly visible for RPA tools that were perceived as must-have commodities, encouraging decisions based on vendor popularity or peer adoption rather than strategic fit. The business impact of such an approach can be severe, resulting in increased costs, operational disruptions, and employee resistance. Organizations must conduct thorough assessments of value propositions and operational models before implementation, and explicitly distinguish between experimentation with commoditized tools and long-term strategic commitments. 15
Cost perception represents another serious challenge. As emphasized by the CEO of ECIT Veny AS: "As a manager, I quickly realized that focusing only on direct cost savings is short-sighted. The real value of automation lies in many other aspects." Technology should be leveraged for strategic advantage beyond mere cost reduction. 16 Moreover, the total cost of ownership (TCO) calculations poses considerable difficulties. In particular, manufacturing companies (companies 8, 15, 20, 23, 27, 29, 30) struggled with infrastructure scaling costs, while service-oriented firms underestimated training and change-management expenses. The Head of Robotics at Nordea Bank Apb highlights this issue: "It's common that every department wants just the benefits and savings from robotization, but nobody wants to take the cost of it. And it's easy to underestimate or simply forget some costs related to, for example, the time of process experts that they need to spend with automation analysts, when it comes to emailing, meetings, workshops, etc." Financial services companies (Nordea, companies 13, 25, 26, 28, 31) demonstrated more sophisticated TCO calculations due to regulatory requirements, yet still underestimated operational costs, suggesting that even heavily regulated industries benefit from systematic cost frameworks. Across organizations, many cost calculations focused on licenses and development effort while neglecting post-implementation activities such as software maintenance, model retraining for AI-based solutions, continuous optimization, governance overhead, and compliance updates. Failing to account for all direct and indirect costs, including initial investments, operational expenses, and maintenance, can lead to budget overruns. One article highlighted the importance of comprehensive cost-benefit analyses. 17 Additionally, the cost of not investing in process automation represents a crucial consideration. As the Product Manager of ECIT Veny AS emphasizes, "If you do not automate your processes, your employees will keep doing the same boring, manual tasks that will eventually make them unhappy or cause them to change jobs. Also, your customers and sales leads will never believe in your aim to transform them digitally and work on innovations in the future." This aligns with findings on the hidden costs of passivity in technological investments. 18
Technologies particularly susceptible to these strategic challenges include RPA platforms like Blue Prism and UiPath, which are often adopted based on market hype rather than strategic fit. At the same time, as the core RPA functionalities have converged and price competition has intensified, many decision makers in our research viewed RPA as a commodity, reinforcing the tendency to prioritize quick acquisition over careful alignment with local processes and capabilities. Similarly, LCDPs such as Zoho Creator and OutSystems may be selected following industry trends without proper evaluation of organizational readiness or TCO implications. Here, the strategic challenge extends beyond the initial “buy” decision for a platform to the broader “buy versus build” question: whether to rely on external consultants and pre-built solutions or to build internal capabilities, including citizen development. 19 Our cases show that internal build options can lower project costs but introduce new, often underestimated components of TCO, such as training, governance of decentralized development, management of technical debt and shadow IT, as well as ongoing maintenance of solutions created outside centralized IT. At the same time, as generative and agentic AI tools become more and more embedded in these platforms, they further shift cost structures (e.g., consumption-based pricing, requirements for specialized skills, and monitoring of AI behavior), making a holistic and life cycle oriented view of TCO essential for the strategic decision-making.
Governance and Competence Building
The lack of in-house competence presents a significant barrier to effective automation implementation. This was visible the most in traditional industries (manufacturing, automotive) where 78% of studied companies (companies 8, 15, 20, 23, 24, 27, 29, 30) lacked prior automation experience, compared to only 23% in technology-adjacent sectors. Without sufficient internal expertise, organizations risk improper implementation and increased dependency on external vendors. Organizations should either build internal capabilities or engage experienced consultants with proven track records and explicitly frame these choices as part of a broader “buy versus build” strategy for intelligent process automation.
Multinational organizations (Nordea, SSAB, companies 8, 15, 23, 29) faced additional governance complexity due to varying regulatory environments and cultural differences, requiring region-specific adaptation of their hub-and-spoke models. Hence, governance misalignment and unclear communication create substantial obstacles. 20 Effective IT governance, defined by clear decision-making structures and accountability, is essential for aligning technology initiatives with business objectives. 21 As noted by the Head of Robotic Execution at Nordea Bank AB, "We need to address the challenges related to the governance of RPA at Nordea. It feels that we need to seek the best possible balance in our hub-and-spokes model." This quote illustrates the challenge of balancing a central center of excellence (CoE) with decentralized responsibilities in business units. An unfitting operational model presents another significant challenge, as it may lead to inefficiencies and employee resistance. As further noted by the Head of Robotics at Nodea: "At the other end, (. . .) in the labyrinth of potential points of failure, information and collaboration are crucial." It is important to clearly define accountabilities and responsibilities within the CoE, to ensure transparent deliveries and reporting structures. 22 Such governance dilemmas become even more important when low-code and no-code platforms, as well as generative or agentic AI tools enable decentralized development by business users and citizen developers, increasing the risk of shadow IT and technical debt if decision rights and responsibilities are unclear. 23
Smaller businesses and family-owned organizations (companies 6, 7, 9, 17, 18, 22) demonstrated faster decision-making but required external validation more frequently, while large corporations needed extensive internal consensus-building that delayed implementation by an average of 3-6 months. The Business IT Team Manager at SSAB Europe highlights another critical challenge: "The business is speaking in their own language (. . .) so developers need to understand what the business really means when they are speaking about some case or some functionality." It points to a continuous IT business alignment gap, which can be particularly risky when business units independently configure low-code apps or AI agents without shared understanding. Such risk can result in solutions that fail to meet business needs, delayed project delivery, and reduced effectiveness. 24
Technologies requiring particular attention to governance include RPA platforms (e.g., Blue Prism, UiPath) where vendors often provide governance models that need adaptation to organizational contexts. As RPA capabilities become increasingly commoditized, organizations in our study tended to adopt vendor governance templates without fully tailoring them to their structures and cultures, which created ambiguity about who owns automation outcomes and post-implementation maintenance. AI tools, as well as low-code and no-code development platforms like Mendix, demand specialized knowledge, continuous monitoring, and explicit governance structures as, without them, organizations struggle to harness their full potential while managing associated risks. In particular, generative and agentic AI systems that can interact with user interfaces and process unstructured data shift governance concerns from purely technical standards to questions of acceptable AI behavior, transparency of decision making, and responsibility for updating or retraining models over time. 25 Our findings suggest that effective competence building requires not only initial training for developers and citizen developers but also ongoing post-implementation controls, such as reviews, architecture communities, and risk management, to oversee the evolution of intelligent automation solutions throughout their life cycle.
Technical Setup
Technical complexity is often underestimated. This is particularly critical for organizations with legacy systems older than 10 years (primarily in banking, manufacturing, and public sector), where integration challenges increased project timelines by 40-60% compared to companies with modern infrastructures. As noted by the Product Manager at Robocorp Oy: "Low-code software for process automation also needs some skills and knowledge about how to build and code solutions. You cannot use it when you are completely inexperienced." Such underestimation frequently results in extended timelines and higher costs. Technical project complexity consisting of many varied interrelated parts can lead to unforeseen challenges. 26 , 27 In the cases we studied, the shift from rule-based RPA toward low-code and no-code platforms and, more recently, generative and agentic AI systems further increased this complexity: while core RPA capabilities appeared commoditized, integration with heterogeneous systems, unstructured data sources, and evolving AI solutions created new forms of technical uncertainty that were often not reflected in initial plans.
Public sector organizations (City of Turku) and heavily regulated industries faced additional complexity due to compliance requirements and vendor approval processes, which required specialized technical roadmaps accounting for regulatory constraints. Scheduling and capacity management issues significantly impact project timelines and resource allocation. Infrastructure scalability concerns must be addressed to prevent performance bottlenecks and increased maintenance efforts. The difficulties of integrating new technologies with existing infrastructures can hinder performance and scalability. 28 Post implementation, such issues may persist in the form of capacity spikes when automation volumes grow, performance decline after platform upgrades, and the need to regularly update connectors, APIs, and AI models. In our cases, organizations that treated technical setup as a one-off project activity rather than an ongoing capability struggled to keep automations reliable and profitable over time.
Legacy system integration poses important challenges. As the program director at City of Turku points out, "There are a number of old school systems where there are no APIs, or development of such is so expensive and cumbersome." Companies with distributed operations (>5 locations) required significantly more robust infrastructure planning, with cloud-based solutions showing 35% better scalability outcomes compared to on-premise deployments in multi-site organizations. Organizations must map all system dependencies and vendor responsibilities before beginning automation initiatives. It becomes even more critical when generative or agentic AI is used to interact with user interfaces or process documents: besides traditional system dependencies, organizations need to document and validate data flows into AI services, training or fine-tuning artifacts, as well as monitoring mechanisms for AI outputs.
Strategic technology roadmap formulation is crucial. As the CEO of Tiilisi Oy notes, "Adding AI has several benefits. First, we save our own time, several hours per week. That time is instantly available for valuable interactions with our current and potential customers." However, incorrect roadmap formulation can lead to misaligned investments and delayed benefits realization. Our findings indicate that roadmaps often focused on which platforms to acquire (buy) rather than on which technical capabilities to establish internally (build), such as reusable components, integration patterns, and monitoring practices. This imbalance left organizations dependent on vendors for post-implementation changes and limited their ability to adapt automations when business requirements or AI capabilities evolved.
Critical technologies include RPA tools like UiPath and Blue Prism, which face scalability and scheduling challenges without proper infrastructure planning. Although many interviewees viewed these tools as interchangeable commodities at the feature level, their long-term operational characteristics, such as licence models, scheduling engines, monitoring dashboards, and upgrade cycles, had substantial impact on maintainability and operating costs. AI solutions such as IBM Watson and Google Cloud AI require robust infrastructure and careful integration strategies. LCDP/NCDP platforms like Mendix and OutSystems often encounter difficulties with legacy system integration, emphasizing the importance of technical compatibility assessment during planning. As these platforms increasingly embed generative AI assistants and support agentic workflows, technical setup must also address AI-specific concerns such as latency, observability of model behavior, rollback mechanisms, and safe defaults when AI-generated actions fail. In our study, the organizations that were most successful over the long-term treated technical setup as a life cycle spanning discipline, investing in reusable integration assets, monitoring and logging, and dedicated post-implementation support for their automation stack rather than merely configuring tools during initial deployment.
Process Documentation and Management
Poor process documentation can negatively impact the success of process automation. This is visible particularly in knowledge-intensive industries (consultancy firm companies 12, 16, 19, 32, 33) where tacit knowledge dominated, requiring more extensive documentation efforts compared to process-standardized industries like manufacturing. The IT Business Area Manager at SSAB notes, "Not rarely, the incomplete documentation of processes and requirements leads to product discrepancies with business expectations." Effective process documentation is essential for successful automation. 29 Organizations with high employee turnover (>15% annually, primarily in marketing services and consultancy) faced greater documentation challenges, as process knowledge frequently departed with employees, requiring more rigorous knowledge capture protocols. Organizations must establish stable handling procedures with clearly defined KPIs and control points aligned with both compliance and business perspectives. The business process management literature underscores the importance of well-defined processes for automation. 30 In our cases, organizations that invested in documentation management, regularly updated process descriptions, exception catalogs, and decision rules, were better able to sustain and evolve automations after deployment, whereas static one-off documents quickly became outdated once processes or technologies changed.
Processes lacking clear rules and exception handling procedures cannot be effectively automated. 31 As highlighted by the Expert at Turku Healthcare Centre, Welfare Division, "A robot working in 100 different languages is awesome, and such a service could also be deployed for other processes and types of queries." This demonstrates how structured input processing enables scalability across diverse applications but also creates a need to document which input types are supported, which are out of scope, and how exceptions will be handled over time.
Scope creep represents a persistent challenge. As Service Manager at Norian shares, "Once development started, somebody asked for one small change, then another. We felt this was a 'rolling implementation'." This highlights the critical need for fixed scope definition before project estimation and for clear change-management procedures once automations are in production.
Smaller organizations (≤200 employees) benefited from simplified process ownership structures with single decision makers, while larger enterprises required formal governance frameworks to manage multiple stakeholders and conflicting priorities. Process ownership clarity is therefore equally critical. As noted by Service Manager at Norian, "The key role was the process owner, so the person who can make decisions about the process." Without clear ownership, automation initiatives face delays and conflicting priorities. Technologies for managing these challenges include natural language processing and optical character recognition tools for structuring unstructured data. AI-enhanced platforms like Google Cloud AI excel at handling unstructured inputs but require precise configuration and clear process definitions to deliver value. With the growth of generative and agentic AI tools that can interpret free-text instructions, navigate user interfaces, and make context-dependent decisions, the focus of documentation shifts: detailed, rule-based flow specifications become less central than documenting prompts, training data sources, policies, and escalation paths. Several organizations in our study initially assumed that such AI systems would reduce documentation needs; instead, they discovered new documentation requirements related to transparency (e.g., explaining why certain outputs were produced), auditability (e.g., logging and reviewing AI actions), and post-implementation tuning (e.g., maintaining prompt libraries and updating decision guidelines). Effective process documentation for intelligent automation therefore spans the full life cycle, from initial process mapping to ongoing recording of changes in processes, data, and AI behavior, to avoid gradual misalignment between the documented process and the automation that executes it.
Performance Measurement and Roadmapping
Industry maturity significantly influenced measurement approaches: financial services demonstrated sophisticated KPI frameworks due to regulatory requirements, while emerging sectors (marketing services, consultancy) relied more heavily on operational metrics and required guidance in developing balanced scorecards. Indeed, organizations frequently struggle with automation accuracy concerns. This concern varies significantly by industry context: manufacturing companies could accept 85-90% accuracy due to human oversight capabilities, while financial services required 98%+ accuracy, necessitating different validation and testing strategies. While AI-based automation technologies may show varying accuracy initially, machine learning capabilities can achieve higher accuracy levels over time. 32 This requires patience and continuous improvement rather than expecting immediate perfection. With the increasing use of generative and agentic AI, accuracy also becomes multi-dimensional, encompassing not only correct outputs but also the frequency of hallucinations, the rate of human overrides, and the consistency of AI behavior across similar cases. The absence of clear performance metrics undermines ROI optimization. A balanced scorecard approach, integrating financial and non-financial metrics, works best. 33 Without clear KPIs, organizations cannot effectively evaluate the return on investment or identify areas for improvement. 34 Our cases show that organizations that defined a small, focused set of pre- and post-implementation indicators, such as processing time, error rates, volume handled, rework, and employee satisfaction, were better able to justify investments and iteratively improve automations after go-live.
Companies in competitive markets (automotive, financial services) demonstrated greater urgency in automation roadmapping but often sacrificed thorough planning, while organizations in stable markets (public sector, utilities) benefited from longer planning horizons but risked over-analysis paralysis. Incorrect prioritization and absent automation roadmaps result in misallocated resources and delayed value realization. Strategic roadmapping is essential for aligning technology implementation with business objectives. Indeed, technology roadmapping is a flexible technique for supporting strategic and long-range planning. 35 In our study, many roadmaps focused on initial deployment milestones (e.g., number of bots or applications to build) rather than on the post-implementation aspects, such as how automations would be scaled, retired, or enhanced as business needs and AI capabilities evolved. Incorporating explicit milestones for post-deployment improvements, such as scheduled performance reviews, retraining cycles for AI models, and refactoring of low-code components, helped organizations treat automation as a dynamic capability rather than a one-off project.
A critical oversight is attempting to automate processes without first optimizing them. As advised by the Product Manager at ECIT VENY AS, "If the process has not been precisely measured so far, it is good to do it even for one week (and then multiply by 52), or as long as possible before implementation." This baseline measurement is essential for demonstrating automation value. Equally important, however, is continuing measurement after implementation: several organizations in our study found that expected savings did not materialize until they used post-implementation data to remove bottlenecks, adjust workload allocation between humans and automation, or fine-tune AI decision thresholds. Measurement supports a continuous improvement loop across the whole automation life cycle rather than only providing a business case prior to implementation.
RPA platforms like Automation Anywhere and UiPath offer built-in performance measurement tools that remain underutilized without clear KPI frameworks. AI solutions such as IBM Watson and ChatGPT require rigorous testing protocols during design phases to address accuracy concerns. In practice, we observed that monitoring and analytics functionalities embedded in RPA, low-code, and AI platforms were often activated but rarely embedded into regular management routines (e.g., monthly dashboards or review meetings). As generative and agentic AI tools are introduced, organizations need complementary monitoring mechanisms, such as logging of AI prompts and outputs, review of failed or escalated cases, and thresholds for automatically disabling problematic automations, to ensure safe and reliable operation over time. Performance measurement and roadmapping are therefore closely related: well-designed metrics guide not only whether to start an automation initiative, but also how to steer and adapt it throughout its post-implementation life.
Human Resource Management
Cultural context significantly influenced change-management success: Nordic companies (Nordea, SSAB, City of Turku) leveraged collaborative decision-making traditions, as organizations in more hierarchical contexts required top-down communication strategies and formal change champions. The absence of comprehensive HR policies for automation-driven job changes creates significant organizational risk. This risk manifests differently across organizational contexts: unionized environments (manufacturing, public sector) required formal consultation processes lasting 6-12 months, while agile service companies could implement changes within 2-4 weeks but faced higher voluntary turnover. Research 36 highlighted that transformation efforts fail primarily due to neglecting the human dimension. As emphasized by the CEO at ECIT Veny AS, "We knew that automation of processes is the future and we need to listen to people about what they would like to do in their future available time, as they often had very good ideas." This proactive engagement transforms potential resistance into innovation opportunities. 37 In our cases, organizations that treated automation as an opportunity to redesign roles and career paths after go-live, rather than only as a cost-cutting exercise, were better able to sustain engagement and avoid post-implementation frustration.
Family-owned businesses demonstrated higher employee loyalty during automation transitions but required more personalized communication approaches, while large corporations benefited from systematic change-management programs but struggled with impersonal implementation. Under-resourcing creates a vicious cycle: employees cannot automate their work, because they lack time for it, being extremely busy with tasks that are easily automatable. This challenge is particularly visible in growing companies (50-500 employees) experiencing rapid scaling, where 67% reported resource constraints compared to 23% in stable, mature organizations. As the CEO at ECIT Veny AS stated, "Automation of processes is something that people should be invited to eagerly join and it should be celebrated as a common achievement." Post implementation, this under-resourcing often persisted: teams were expected to monitor, improve, or retrain automations on top of their regular duties, leading to change fatigue and limiting the organization’s ability to realize the full benefits of intelligent automation.
Automation changes the nature of work, requiring employees to adapt to new roles that often involve higher-order cognitive skills. 38 Organizations must invest in training to prepare their workforce for these changes . 39 Our findings highlight that this training need is not limited to the pre-deployment phase. As low-code and no-code platforms spread and generative and agentic AI tools are introduced, employees must continuously learn how to collaborate with intelligent systems, interpret their outputs, and escalate edge cases. 40 In several organizations, citizen developers and process owners became de facto maintainers of local automations without receiving systematic training in life cycle management, risk awareness, or AI-specific issues such as prompt design and bias.
Communication failures compound these challenges—organizational culture influences how technological change is perceived and adopted. 41 As the Global Head of Intelligent Automation at SSAB states, "For us, effective communication is a key to engaging all stakeholders." Transparent communication strategies help in aligning stakeholders and reducing uncertainties. 42 In our study, communication that focused only on the launch of new automations, without explaining how work would evolve over the following months (e.g., reallocation of tasks, opportunities for upskilling, expectations for monitoring AI outputs), tended to create rumors and resistance, particularly when generative AI tools were involved. From the other hand, organizations that established recurring meetings and forums, such as town halls, communities of practice, or feedback channels, were better able to surface post-implementation issues and adjust both technology and work design.
Technologies such as RPA initially cause apprehension among employees who fear job displacement. Success requires comprehensive training programs and clear communication about how automation enhances rather than replaces human work. E-learning platforms for reskilling remain underutilized, highlighting the gap between available tools and effective human resource strategies. The introduction of generative and agentic AI intensified these dynamics: while some employees welcomed AI as a coworker that could handle routine digital tasks, others perceived it as misleading and unpredictable. Organizations in our sample that explicitly framed AI-enabled automation as augmenting human judgment, by, e.g., positioning agents as first drafts or triage tools rather than final decision makers, and that offered ongoing support for employees experimenting with these tools, reported smoother and more sustained adoption use over time. Thus, human resource management in intelligent process automation extends beyond initial change communication to long-term capability building, psychological safety, and career development in increasingly AI-influenced work environments.
Summary
Our research across 33 Nordic and Central European enterprises identified 70 distinct risks in implementing intelligent process automation. The 37 most critical challenges analyzed provide a roadmap for practitioners navigating automation initiatives across the full life cycle, from initial scoping and design to post-implementation optimization.
Practitioners should tailor their automation strategies based on organizational context: small-to-medium enterprises (≤500 employees) should prioritize external consulting partnerships and simplified governance structures to compensate for limited internal expertise, while large organizations (>10,000 employees) must establish formal coordination mechanisms and invest in internal capability building to manage complexity. Industry-specific approaches are also essential: regulated sectors should extend implementation timelines by 40-60% to accommodate compliance requirements while leveraging their superior governance capabilities, manufacturing companies should focus on process-standardized activities before pursuing knowledge work automation, and service industries must allocate 30-50% more resources to change management compared to traditional sectors. Across all contexts, the increasing commoditization of RPA, and spread of low-code and no-code development platforms, as well as the emergence of generative and agentic AI tools make it even more important to move beyond tool-focused decisions toward context-sensitive strategies that account for governance, life cycle costs, and human factors. 43
International organizations must additionally develop region-specific adaptation strategies for their global automation frameworks, accounting for cultural decision-making preferences and varying regulatory environments. Success in automating processes requires addressing each of these challenges systematically while maintaining clear communication and stakeholder engagement. 44 Organizations should conduct comprehensive context assessments before selecting automation approaches, recognizing that one-size-fits-all solutions consistently underperform compared to contextually adapted strategies. Organizations must carefully balance technical requirements, human factors, and strategic objectives to achieve optimal results from their automation investments. 45 Rather than following industry trends, practitioners should evaluate automation opportunities against their specific organizational context, including size, industry maturity, regulatory environment, and cultural factors, to prevent impulsive decisions and ensure sustainable value creation. 46 This includes explicitly formulating sourcing choices as “buy versus build” decisions that weigh external solutions and consulting support against investments in internal capabilities, citizen development, and long-term governance.
Technical challenges, from legacy system compatibility to cybersecurity vulnerabilities, require proactive planning and robust governance frameworks. 47 Organizations with legacy systems older than 10 years should allocate additional 40-60% implementation time and consider cloud-based solutions for better scalability outcomes. By mapping problems to specific stages of the DevOps cycle, organizations can better anticipate challenges and implement mitigation strategies proactively. Post implementation, the same integration and security issues continue to surface as platforms get regularly upgraded, data volumes increase, and AI models evolve, underscoring the need to treat technical setup and monitoring as ongoing capabilities rather than one-off project tasks.
The most critical aspect of automation implementation revolves around managing the human element. 48 Flat-structured organizations should leverage their collaborative decision-making culture, while hierarchical organizations must establish formal change networks and top-down communication strategies. Companies must address resistance to change, skill gaps, and the need for continuous upskilling through comprehensive change-management practices and fostering an innovation-embracing culture. 49 Table 1 and Supplemental Table 2 summarizes key propositions for each of the six perspectives, i.e., strategic decision-making and financial management, governance and competence building, technical setup, process documentation and management, performance measurement and roadmapping, and human resource management, providing a concise, practitioner-oriented overview of our main findings.
Summary of key challenge propositions by perspective.
Practitioners must prepare for rapidly evolving automation capabilities that will fundamentally change implementation strategies. Looking ahead, the intelligent automation landscape shall continue to evolve rapidly. Emerging general-purpose AI agents that can operate user interfaces, orchestrate multi-step workflows, and process unstructured data 50 push the boundaries of intelligent process automation, blurring the line between rule-based automation and autonomous decision-making. Organizations must carefully consider the impact on user experience and ensure a seamless transition for all stakeholders. Digital leaders should begin evaluating these emerging technologies now: small-to-medium enterprises should focus on understanding interface automation capabilities that could reduce their dependency on technical expertise, while large organizations must assess how general-purpose AI agents will integrate with their existing governance frameworks. These developments extend intelligent automation beyond rule-based back-office processes to tasks that require interpretation, contextual reasoning, and collaboration with human workers. 51 The contextual factors identified in our study, i.e., organizational size, industry maturity, and cultural context, will become even more critical as these advanced tools require sophisticated change management and governance strategies.
Our study involved 33 companies, each contributing different levels of resources, time and insight. While some cases were studied extensively (e.g., 17 hours of interviews and observation), others were more limited in scope (e.g., 1 hour). To address this limitation, we focused on synthesizing overarching challenges and lessons learned that were applicable across diverse organizations and contexts.
Future studies should prioritize investigating context-specific automation outcomes through longitudinal studies that track implementation successes and failures across different organizational profiles and stages of automation life cycle, including post-implementation adoption. Also, include organizations across different industries and geographies to enhance generalizability and enable in-depth comparative analysis. We particularly recommend that organizations systematically quantify their cost of not investing, including measurable impacts on operational efficiency, employee well-being, and customer trust, as this will become increasingly critical for justifying automation investments in competitive markets.
By providing the comprehensive analysis of automation challenges and contextualized solutions, we equip practitioners with actionable frameworks to select appropriate strategies based on their specific organizational context, to govern “buy versus build” choices and decentralized development, and to manage post-implementation evolution of intelligent automation. It will position them to thrive in an increasingly automated future through evidence-based, context-aware implementation approaches.
Supplemental Material
sj-docx-1-cmr-10.1177_00081256261434509 – Supplemental material for Identifying and Overcoming Challenges in Intelligent Process Automation
Supplemental material, sj-docx-1-cmr-10.1177_00081256261434509 for Identifying and Overcoming Challenges in Intelligent Process Automation by Damian Kedziora, Dominik Siemon and Joanna Kedziora in California Management Review
Supplemental Material
sj-docx-2-cmr-10.1177_00081256261434509 – Supplemental material for Identifying and Overcoming Challenges in Intelligent Process Automation
Supplemental material, sj-docx-2-cmr-10.1177_00081256261434509 for Identifying and Overcoming Challenges in Intelligent Process Automation by Damian Kedziora, Dominik Siemon and Joanna Kedziora in California Management Review
Supplemental Material
sj-docx-3-cmr-10.1177_00081256261434509 – Supplemental material for Identifying and Overcoming Challenges in Intelligent Process Automation
Supplemental material, sj-docx-3-cmr-10.1177_00081256261434509 for Identifying and Overcoming Challenges in Intelligent Process Automation by Damian Kedziora, Dominik Siemon and Joanna Kedziora in California Management Review
Footnotes
Notes
Supplemental Material
Supplemental material for this article is available online.
Author Biographies
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
