Robotic Process Automation (RPA) has long been a standard for efficiency in many organizations. It enables the automation of routine, repetitive tasks, freeing up specialist resources for more complex assignments. However, this year has already seen a qualitative leap in RPA development in practice, driven by deep integration with artificial intelligence. This synergy transforms automation from a tool for simple tasks into a platform for optimizing complex, cognitive processes.
The central thesis is that the integration of AI into RPA is transforming business processes this year and in the coming ones, but the success of this transformation depends on data readiness and effective management of AI-related risks.
AI in RPA: The evolution of business process automation
Traditional RPA excels at automating structured, rule-based processes. However, its capabilities are limited when dealing with unstructured data, context-based decision-making, or adapting to change. This is where AI comes in. The integration of AI agents allows RPA bots not just to follow instructions but also to learn, analyze, understand natural language, recognize images, and make predictions. This opens the door to automating processes previously considered exclusively human prerogatives.
According to Gartner, 40% of enterprise applications will incorporate integrated AI agents for specific tasks this year, a significant increase from less than 5% in 2025. This trend indicates that integrating AI into enterprise systems, including RPA, is becoming an industry standard rather than an experimental technology.
Challenges of using data for AI analytics in the public sector
The public sector, like any large organization, faces specific challenges when implementing AI in RPA. The primary issues revolve around data: fragmentation, inconsistency, gaps, and outdatedness. Data is often scattered across various non-integrated systems, making it difficult to create a unified, reliable information space. Without quality, accessible data, even the most sophisticated AI models cannot function effectively. This is particularly sensitive for critical public services where accuracy and reliability are paramount.
A common mistake: training AI on poor-quality data
One of the most common and costly mistakes in implementing AI in RPA is attempting to train models on poor-quality, incomplete, or inconsistent data. The result is AI systems that make incorrect decisions, generate erroneous forecasts, or amplify existing biases. This not only undermines trust in automated processes but can also lead to significant financial and reputational losses. Before investing in complex AI models, a reliable foundation of quality data must be ensured.
Scenario: Automating citizen request processing with AI and RPA
Let’s consider a typical scenario in the public sector: automating the processing of citizen requests. Traditionally, this process requires significant human resources for registration, classification, routing, and responding to inquiries. With AI in RPA, this process can be significantly optimized.
In practice, this looks like this: when a request arrives (via email, web form, or messenger), an AI agent integrated into a document management system (DMS) like Scriptum (a low-code BPM platform on UnityBase from InBase) analyzes the request text. It uses Natural Language Processing (NLP) to extract key information: the topic, type of request, and mentioned organizations or individuals. The RPA bot, in turn, uses this information to automatically register the request in the system, classify it, assign responsible personnel, and send an automated confirmation to the citizen. For more complex requests requiring human involvement, AI can prepare a draft response or gather relevant information from various internal systems, for example, using the integration capabilities of the UnityBase platform (an open-source low-code platform developed by InBase).
This approach not only speeds up processing but also improves the quality of responses, reduces errors, and allows employees to focus on resolving unique and complex cases that require empathy and in-depth analysis.
AI risk management in the context of RPA
As AI systems in RPA become more complex and autonomous, the associated risks also grow. Managing these risks is becoming mandatory, especially for critical infrastructure and applications using generative AI. The U.S. National Institute of Standards and Technology (NIST) has developed the AI Risk Management Framework (AI RMF 1.0), which structures management around four functions: Govern, Map, Measure, and Manage. This approach helps organizations identify, assess, and mitigate risks related to the design, implementation, and use of AI.
One specific example is the risk of Prompt Injection, which OWASP LLM Top 10 ranks first for LLM/GenAI applications. This risk involves an attacker manipulating input data (prompts) to force an AI model to perform unwanted actions or disclose confidential information. For AI in RPA, this could mean that a bot interacting with external systems might be compromised through manipulation of the text it processes.
The success of AI implementation also depends on organizational factors. As noted in the 2026 Work Trend Index Annual Report, factors such as culture, manager support, and talent management practices explain twice the impact of AI compared to individual efforts. This means that technology alone does not guarantee success; a comprehensive strategy is needed, including staff training, changes in corporate culture, and active leadership support.
Data readiness as the foundation for AI implementation
The most crucial step before any AI project is ensuring data readiness. This includes not only data collection but also cleaning, structuring, standardization, and quality assurance. Implementing Master Data Management (MDM) systems and data governance policies is a mandatory requirement. For instance, alliance members like Softengi and InBase have experience in building such systems, providing a single, reliable view of data for complex enterprise solutions.
Only after data has been brought into proper condition can one expect the effective and secure implementation of AI in RPA. Ignoring this stage will inevitably lead to project failure or the creation of systems that cause more problems than they solve.
Data readiness checklist for AI in RPA
- A data quality audit (consistency, completeness, accuracy) has been conducted for key business processes.
- Unified dictionaries and master data (MDM) for critical entities (customers, products, services) have been defined and implemented.
- A data governance policy is in place, defining responsibilities for data quality and usage.
- Processes for data cleaning, transformation, and validation have been automated before their use in AI models and RPA scenarios.
- Logging of AI model and RPA agent decisions is provided for audit and analysis (drift, hallucination rate).
- AI-specific metrics have been defined to assess the effectiveness and security of solutions (e.g., classification accuracy, false positive rate).
- A ‘human-in-the-loop’ mechanism is in place for critical decisions made by AI agents.