This is a guest post by Vitalii Kamynin, an expert with more than 15 years of experience in developing software solutions for the digital transformation of the pharmaceutical industry. Founder and CTO of OMNI Digital and Infolek, specializing in creating innovative products that automate and optimize business processes of pharmacies and pharmaceutical companies.
AI is rapidly becoming a core component of modern analytics workflows — from data preparation and transformation to report generation and decision support. According to McKinsey’s The State of AI in Early 2024, 72% of organizations have implemented AI in at least one business function, with a growing share reporting regular use of generative AI. Analytics is consistently cited among the most impactful areas for these technologies.
But as the tools evolve and enthusiasm rises, so does the need for a more critical look at where AI truly delivers value and where it still needs human oversight. Fast data processing and user-friendly interfaces do not always ensure the reliability of outcomes. Models trained on historical data often struggle with aspects that go beyond patterns.
In this article, we will examine the areas where AI adds measurable value to analytical work and those where its use requires increased caution.
In which analytical tasks AI delivers the greatest efficiency
- Data Analysis and Preparation
AI enables the automation of labor-intensive processes, which is particularly valuable in cases where data is high-dimensional, comes from heterogeneous sources, or contains hidden issues.
Rapid Data Diagnostics: AI makes it possible to evaluate the quality and structure of a dataset almost instantly. In a matter of seconds, it can perform a basic “health check” of a dataset: identifying columns with missing values, detecting anomalies, flagging rare or inconsistent records, analyzing the distribution of numerical indicators, and highlighting potential data quality issues. This functionality becomes especially important when data must be assessed quickly, such as during audit preparation or reporting cycles. For example, in a case study from Turing, a healthcare organization, the implementation of AI tools helped reduce audit preparation time by 50% and improved regulatory compliance. Automated data diagnostics became central to their broader data quality management process.
Pattern and Anomaly Detection: According to McKinsey’s 2023 report, AI pattern detection is among the most commonly used tools in customer and product analytics. AII can group similar records, flag deviations from typical patterns, and surface unusual behaviours, whether an unexpected user action, a sudden spike in activity, or an unusual combination of attributes. These insights help analysts form hypotheses faster and identify emerging trends earlier.
Data Preprocessing and Cleaning: Before any meaningful reporting or modelling can begin, data must be organized and cleaned. AI can assist by identifying which fields are numerical, categorical, and plain text; filling in missing values (e.g., with the mean); highlighting duplicates or incorrectly formatted entries; and helping to standardize names, dates, and textual values.
This significantly cuts down preparation time, mainly when working with multiple, inconsistent data sources.
- Working with Text, Code, and Interfaces
AI is reshaping analytics not just by processing data, but by making interaction easier: helping to formulate queries in natural language, accelerate code development, and simplify work with unstructured text.
Generating SQL Queries from Natural Language: AI improves speed by transforming plain-language requests, such as “show revenue by category for the last quarter,” into executable SQL queries. This is especially useful when working with complex database structures, when the analyst is unsure about the exact syntax, or when a query needs to be quickly adjusted for a new task.
In one of our own SaaS platforms, where clients can view processed data (e.g., distributor and pharmacy chain reports) through a web interface, we integrated an AI assistant to simplify user interaction.
A user unfamiliar with SQL or report configuration can enter a natural-language query, such as: “Which pharmacy chains sold Somatropin the most during spring 2024?” or”Show YTD package sales by brand for May 2024.”
Working with Text Data: Handling unstructured
AI plays a key role in accelerating the development of analytical code. It allows for faster transitions from concept to prototype, which is especially valuable when the focus is on exploring insights rather than implementation. text remains one of the more challenging aspects of analytics, but AI has made significant progress in this area. Modern language models can automatically detect language, identify topics, and assess sentiment, whether it is open-ended survey responses, customer support tickets, CRM notes, or online reviews. They can classify tone as positive, negative, or neutral, enabling quick assessments of customer sentiment and the early identification of emerging issues.
Beyond basic sentiment analysis, AI can extract key ideas, summarise a text’s core message, and group similar entries by theme. This dramatically reduces the time needed to interpret large volumes of data and helps analysts focus on the most critical insights.
Many tools also incorporate machine translation, simplifying work with multilingual sources. In global environments where feedback arrives in different languages and must be unified within a common analytical framework, this is incredibly handy.
The system uses a two-stage prompt refinement process, which accounts for the database structure and reference directories. It then generates and runs the appropriate SQL query, returning results directly to the user.
- Automated Visualization and Analytical Insight Generation
AI is increasingly used to automate the creation of dashboards and visualisations from structured data. Models can independently select appropriate forms of visual representation, intelligently configure axes and aggregations, and generate basic dashboards in BI platforms or code environments. Moreover, AI can highlight notable patterns in the data, such as trends, anomalies, and correlations, offering initial commentary to support exploratory analysis.
In user data analysis, AI is often applied to segmentation — grouping entities based on behaviour or shared attributes. These segments can support personalisation strategies, performance tracking, retention analysis, and A/B testing. However, algorithmic groupings are not always meaningful in business terms. Final interpretation still requires expert review to ensure the segments are relevant and actionable.
AI also plays a role in the early exploration of new datasets, helping analysts quickly identify patterns, structural anomalies, or relationships worth investigating further.
Analytical Scenarios Where AI Remains Limited
With all its progress, AI is still far from being a universal solution in analytics. In specific scenarios, what looks like automation on the surface quickly breaks down, especially when the data lacks structure, context matters more than patterns, or the answers are not purely technical.
1. Data Integration from Heterogeneous Sources
Merging data from CRM, ERP, and web analytics systems is still challenging for AI. Differences in structure, encoding, naming conventions, and the lack of shared unique identifiers often produce unreliable results. Automated matching without consideration of business logic frequently leads to errors, either by combining unrelated data or by losing critical information. These issues are particularly notiсeable when working with duplicate records that are similar but not identical.
2. Conflict Resolution and Duplicate Management
AI really does have a hard time deciding which version of a duplicate record to keep. Conflicts often arise when the same customer or transaction appears across different systems with discrepancies in fields like contact details, IDs, or timestamps. Relying on fully automated selection in these cases risks distorting results and compromising data quality.
In practice, probabilistic deduplication algorithms merge unrelated objects or fail to identify apparent matches. This is particularly common without a consistent, unique identifier across datasets. As a result, proper record consolidation often requires manual rule configuration, alignment with business logic, and expert oversight to assess the relevance and integrity of the data accurately.
For example, consolidating customer data is one of the key challenges in automating the processing of reports from distributors and pharmacy chains. The same pharmacy might be listed in one system as “Pharmacia Vita; Kishinev, Vasile Alecsandri Street, 54,” and in another as “SRL Pharma; Kishinev, Bucharest Street, 45.” In reality, both refer to the same location situated at a crossroads and having both a branded and a legal name. Different sources, while correctly providing the address, may refer to different streets or use different names, leading to discrepancies in describing the same establishment.
In both cases, full automation remains out of reach. Errors in matching can significantly distort analytical outputs and affect business decisions, underscoring the continued importance of expert involvement.
3. Explaining Complex Models and Outcomes
AI can produce outputs that are technically correct but difficult to interpret. Models may confuse correlation with causation, rely on internal parameters that are not easily explained, or use simplified methods that overlook hidden variables.
This challenge is mainly present in complex architectures like neural networks, which are often perceived as “black boxes.” In such cases, a lack of explainability limits trust and adoption, especially in industries with high demands for transparency, such as finance, healthcare, and law enforcement.
4. Low Sensitivity to Rare and Unique Events
Most AI models are trained on patterns seen in typical datasets. As a result, they tend to struggle with rare or unusual events. These systems may miss critical anomalies or flag infrequent but perfectly valid cases as errors. This becomes a particular issue in domains with high variability, such as logistics, manufacturing, or B2B analytics.
Another limitation is the lack of embedded business context. AI evaluates events based on statistical deviation rather than practical relevance. Without human insight, rare events risk being misread or missed altogether. Getting it right often means fine-tuning models regularly and relying on expert judgement to separate what is unusual from what truly matters.
AI has already cemented its role in modern analytics, but its impact still hinges on one key factor: how well the task is framed and who is guiding the process. As the technology matures, we can expect models to become more adaptable to non-standard situations, understand the business context more deeply, and offer greater transparency in how decisions are made.
Even so, expert interpretation is essential. For now, and likely for years to come, human judgment will be the factor that turns algorithmic output into real insight.
