Artificial intelligence driving operational excellence across different industries

The ability to transform data into rapid, informed decisions has become one of the key factors driving competitive differentiation in modern organizations. In a context where the volume of available information is growing at an unprecedented rate, the central issue is no longer the quantity of data: it is the ability to interpret, structure, and mobilize it at the right moment.

This case study covers projects developed in organizations across different sectors, all united by a common challenge: transforming scattered data, manual processes, and intuitive decisions into intelligent, automated systems. In each case, the implemented solution combined artificial intelligence models, geospatial analysis, machine learning, or generative AI with careful integration into the organizations’ existing processes and technologies, enabling measurable results in terms of operational efficiency, quality, information reliability, and innovation capacity. The following cases illustrate this transformation logic across distinct operational contexts.

Detection of leaks and consumption anomalies in a water supply network

The organization operates in municipal water supply. The leak detection process was carried out by various teams based on field inspections and valve checks; it was a complex, time-consuming task that was rarely completed with the necessary speed, resulting in significant water losses with considerable economic and environmental consequences.

The approach to transforming leak detection

To address these challenges, an artificial intelligence solution capable of classifying leak risk and identifying anomalous consumption in real time was developed and implemented, integrated into the existing management platform.

The solution is based on an AI model that combines internal and external data to generate a dynamic risk map of the network. Among the data used, the following stand out:

• Internal data: multi-year history of breaks, spatial distribution of the supply network, network pressure and performance data.
• External data: temperature and precipitation, population density, and soil type.

The developed system processes internal and external data using AI causal models, generating as outputthe probable location of breaks, an updated risk map, and real-time alerts for the detection of anomalous consumption. The system has self-learning capabilities and integrates with the organization’s existing systems.

Figure 1 – Example of the implemented system

Results achieved

The implementation of the AI solution for leak detection and consumption anomalies had a significant impact on key operational indicators:

• 80% of damaged pipes were correctly identified by the model.
• A 25% reduction in unbilled water, with a direct impact on operating costs.
• 10% of the network’s pipes were classified as high-risk.
• Reduction in the resources allocated to the leak detection process.
• Greater efficiency in real-time detection of leaks in the network.

Automated inspection of blister packs in the pharmaceutical industry

The company in this case operates in the pharmaceutical industry, where the blister packaging process requires rigorous verification to ensure that each blister contains the correct medication, which is a critical requirement for patient safety and regulatory compliance in the sector.

The operational challenge

The existing equipment required between 25 and 40 minutes of manual calibration for each new batch or format change, which represented a significant operational cost in terms of line availability.

Conventional inspection systems relied on simple contrasts, such as color and weight, which proved insufficient in more complex scenarios or when dealing with equipment from different generations.

Inspection automation as a solution

A computer vision solution based on Deep Learning models was developed to inspect blister packs in real time without prior calibration. Using images captured on the production line and a database of different formats, the model automatically detects each cell and classifies it as “Full” or “Empty,” issuing alerts for immediate rejection.

The solution works for both aluminum and transparent plastic packaging without the need for reprogramming, with the potential to identify more complex defects, including pigment contamination, roughness, cracks, and sectional damage to the tablets, paving the way for a more comprehensive quality inspection.

Results achieved

The implementation of the AI-powered automated inspection solution had a significant impact on quality and operational efficiency metrics:

• Setup time reduced from approximately 25 minutes to just 0.
• 100% accuracy rate in classifying filled vs. empty blisters in the tests conducted.
• The solution can be implemented on any production line, regardless of the number of blisters or the tablet shape.

Geospatial analysis for the expansion of a gas station network

With an established geographic presence, the company sought to expand its network in a structured way, based on objective data and predictive models, rather than decisions based solely on empirical criteria.

The process of identifying new gas station locations was conducted without using geospatial techniques or geographic context data, which limited the objectivity and accuracy of investment decisions. Additionally, the potential of each location to predict the station’s future performance was not quantified, making it difficult to compare alternatives and prioritize opportunities.

The organization identified the following gaps in the existing process:

• Lack of sales forecasting models by location, preventing an objective assessment of each location’s potential.
• Lack of integrated geospatial data (mobility, points of interest, demographics) to support decision-making.
• Inability to compare each gas station’s actual performance with its estimated potential.

The approach to identifying Ideal locations

An intelligent solution was developed to support site selection, based on machine learningand geospatial analysis models. The solution integrates internal and external data to build a robust predictive model. Internal data includes the portfolio of existing stations, such as location, product offerings, and fuel and non-fuel sales. Based on these inputs, the model performs four main analytical processes:

• Clustering of existing stations by surrounding profile, grouping stations with similar geographic contexts.
• Development of sales forecast models for each clusterusing machine learning algorithms.
• An optimization model to identify the best locations for new gas stations.
• Comparison of each station’s actual performance with its estimated potential, identifying opportunities for operational improvement.

The solution was made available in the form of an interactive map, where teams can obtain sales forecasts for any geographic location of interest.

Qualitative results achieved

The implementation of the smart location solution had a direct impact on the organization’s strategic and operational capabilities:

• Identification of the best locations for network expansion, based on objective and quantifiable criteria, replacing the previous empirical approach.

Figure 2 – Map exemplifying the network expansion

• Systematic comparison of actual performance versusestimated potential for each station.
• Creation of differentiated product offerings by location.

Data governance: from quality to data-driven decision-making

The company analyzed in this project operates in the cellulose pulp and forestry products industry, a capital-intensive sector heavily driven by operational data. It had multiple information systems, with disconnected data silos and inconsistent data quality, which led to mistrust in the data used to support decision-making. Data management was predominantly ad hoc, lacking standardized processes or clear responsibilities, and data quality varied significantly across platforms, with no uniform validation standards.

The approach to implementing data governance

A structured methodology was adopted to build a reliable database ready to leverage artificial intelligence solutions. The initial phase involved assessing the organization’s data management maturity, defining the governance strategy, and designing the target data architecture, culminating in the development of a 24-month strategic roadmap.

The implementation spanned multiple dimensions: the creation of certified reports, the implementation of self-service capabilities, data management, and the development of an organization-wide data literacy program. The following support elements were created:

• Data strategy report and internal page with program information.
• Roadmapand communication plan published internally.
• Datagovernance frameworkmade available to the entire organization.
• Monthly data committee meetings, regular training sessions and workshops.
• Business glossary, reporting catalog, and documented demand management processes.
• Alignment between the technology roadmap and the data roadmap.

Achieved results

The implementation of the data governance program has led to measurable improvements in information quality, process efficiency, and organizational culture:

• More than 200 KPIs with guaranteed reliability, ensuring greater consistency in the organization’s critical indicators.
• A reduction of 185 hours per month spent on report preparation.
• Establishment of a single, reliable source of data, with certifiedreportingand complete traceability of information.
• Clear definition of responsibilities and qualitystandardsthroughout the organization.
• AI-ready data infrastructure.

Data-driven production planning: synchronization between plants and reduction of inefficiencies

The company operates in the manufacturing industry and has two separate plants, one dedicated to the production of semi-finished goods and the other to customization and packaging operations, which were managed by independent planners. Consequently, the strong interdependence between the two units made aligning the production plan an ongoing operational challenge.

The complexity inherent in coordinating the two plants led to recurring inefficiencies that affected the reliability of the plan and the utilization of installed capacity.

Additionally, the organization had limited visibility into the impact of plan changes across the entire supply chain, which hindered a rapid and informed response to operational disruptions.

Production planning transformation

An integrated and automated planning solution was developed, based on optimization algorithms and inter-plant synchronization logic, which eliminates the need for manual coordination and generates reliable plans in real time. The planning platform processes demandinputand system data and applies three core analytical capabilities:

• Synchronization logic to automatically align the plans of the two plants.
• Inventory-aware routing, optimizing material flow based on available stock.
• Optimization based on capacity constraints, setups, and operational limitations, generating a feasible plan for semi-finished goods, customization, and packaging.

Breakthrough results

The implementation of the planning solution had a significant impact on operational reliability and resource efficiency:

• A 20-percentage point increase in the fulfillment rate between the production of semi-finished goods and customization and packaging.
• A 15-percentage point improvement in adherence to the production plan.
• An 80% reduction in the time spent on the planning process.
• A reduction of 4 days in the capital expenditure (CAPEX) decision-making process.
• Real-time coordination between the two planners and the two factories, reducing communication errors and rework.

Web Scraping with AI for R&D: smart monitoring of strategic information

The organization operates in the advanced composites industry, a sector where competitive advantage is based on continuous innovation and the ability to apply sustainable materials to demanding technical markets. It has multiple R&D teams that rely on continuous access to external knowledge to drive innovation. However, this information was collected manually, in a fragmented and non-scalable manner, consuming significant resources without ensuring systematic coverage of relevant sources.

The high volume of unstructured external data, coming from multiple sources and in different languages, made manual collection inefficient and incomplete. Duplication of effort among teams and the lack of a standardized screening process exacerbated this problem.

Therefore, the lack of standardization and structuring of theinsightsobtained by the different teams created information silos and reduced the ability to leverage the accumulated knowledge. The organization lacked effective mechanisms to share and consolidate monitoring results across departments.

Additionally, the absence of systematic mechanisms for validating the collected information jeopardized the accuracy, consistency, and reliability of the data used in R&D decisions, exposing the organization to errors based on unverified sources.

Automating R&D monitoring

An end-to-endAI-powered web scrapingsolution was developed, based on two intelligent agents that communicate with each other, which automates the collection, validation, structuring, and distribution of strategic information to R&D teams. The system uses two complementary AI agents:

• Agent 1 (collection): performs intelligent web scrapingwith HTML preprocessing, identifies relevant news articles by keywords, and validates content based on defined criteria.
• Agent 2 (analysis): conducts an in-depth analysis of the full articles, generating structured executive summaries using generative AI models.

Theoutputs generated include a structured database of identified news items and opportunities, an automatically generated PDF report summarizing the relevant findings, and integration with Microsoft Teams to enable teams to interact and collaborate on the collected information.

The workflow is orchestrated by an automated “scheduler” integrated with Power Automate, ensuring the monitoring process runs continuously without manual intervention. The solution is fully integrated into the organization’s existing technologystack.

Results achieved

The implementation of the AI-powered web scrapingsolution had a significant impact on the operational efficiency and innovation capacity of the R&D teams:

• Automated and structured monitoring of more than 300 information sources.
• A reduction of 150 hours per week spent on collecting and synthesizing strategic information.
• Elimination of tasks with no added value.
• Processing of repeatable and validated information, with traceability ensured by AI agents.
• Acceleration of innovation through the anticipation of global trends.

AI-based forecasting for construction planning

The company in question is a leader in the construction sector, specializing in industrial facilities, and faced challenges in accurately estimating task durations during the budgeting and initial planning phases of complex industrial construction projects.

The lack of precision in initial planning directly impacted project schedules and costs. Compounding this challenge were undetected risks during key phases of construction, as well as difficulty in adjusting plans in response to external variables.

Approach to a more accurate forecasting

A solution based on AI models was developed, capable of predicting task durations and identifying delay risks using various data sources. Among the data used, the following stand out:

• Historical project data: recognition of duration patterns by task type, location, and project features.
• Macroeconomic and regional indicators: labor availability, inflation, and other relevant local economic factors.
• Weather history and forecasts: impact of weather conditions on weather-sensitive activities (e.g., foundations and roofing).
• Subcontractor and supplier capacity: data on lead times and capacity constraints that increase forecast accuracy.

The coordinated solution also enables automatic task scheduling based on project characteristics, the issuance of real-time alerts during execution indicating potential delays, and the continuous updating of the model with actual data from project execution.

Figure 3 – Example of a schedule

Qualitative results obtained

The implementation of the AI solution for forecasting and planning had significant qualitative impacts on the company’s planning capabilities:

• Ability to simulate scenarios, which enabled better contingency planning and more efficient resource allocation.
• Real-time visualization of risks and monitoring of deviations.
• Faster and more informed decision-making.
• Continuous learning by the model with each project, creating a constantly evolving knowledge base.

Conclusion: the transformative impact of artificial intelligence across different industries

This case study demonstrates that adopting artificial intelligence and advanced data analytics solutions is not an end in itself, but rather a means to build organizations that are more agile, more reliable, and better able to respond to the demands of a constantly changing market.

The journey toward a truly data-driven organization is an ongoing one, and that is precisely where the Kaizen Institute and its specialized digital solutions make a difference.

We are committed to respecting our clients’ confidentiality. While we have altered or omitted their names, the results are genuine.