Category: Case Study

About the company

A fintech startup.

Business Need

Fintech lenders face the critical challenge of ensuring timely loan repayments to maintain financial stability and minimize losses. However, predicting whether customers will pay their outstanding bills or debts within the designated timeframe remains challenging. Traditional collection strategies rely on manual processes and static credit risk assessments, leading to inefficiencies and increased operational costs.

To address these challenges, our client sought Altysys’ team to develop an advanced machine learning-driven solution to enhance its loan repayment prediction capabilities. The primary objectives were to:

• Accurately forecast the likelihood of a customer making a payment within 30 days
• Streamline collection workflows while reducing manual interventions
• Optimize debt recovery rates and operational effectiveness

Solution

The Altysys team conducted a comprehensive needs assessment, crafted a solution roadmap, and executed these key initiatives:

• Developed a classification model leveraging historical payment data, account status, and communication records (texts, calls, payments) to predict loan repayment likelihood.
• Integrated multiple data sources, including accounts, text logs, call logs, and payment history, to train an ML model with a recall of 40% and precision of 73%.
• Planned future enhancements by incorporating NLP techniques to analyze call transcripts and text interactions, utilizing large language models (LLMs) to improve prediction accuracy and model performance.

Tech Stack

Python, SQL, Mlflow, Databricks Runtime ML

Business Impact

• Higher collection efficiency, minimizing manual effort
• Increased payment success rate, covering 69% of the debt amount
• Boosted billing team productivity

About the company

A major bank based in the US.

Business Need

In the fast-paced world of banking, delivering seamless customer service is crucial for maintaining client trust and operational efficiency. However, banks must manage high volumes of customer inquiries while ensuring fast, accurate, and compliant responses. Traditional support channels often result in long wait times, inconsistent responses, and an overburdened human support team.

Our client faced challenges in meeting the growing demand for real-time assistance on investment options, financial instruments, and account management. To enhance customer satisfaction, ensure regulatory compliance, and optimize support operations, the client partnered with Altysys to develop an GenAI-enabled solution that could:

• Process thousands of inquiries simultaneously without performance issues
• Ensure all responses aligned with financial regulations
• Provide accurate insights from live financial data
• Maintain performance during trading surges

Solution

After conducting a comprehensive needs assessment, the Altysys team created a solution roadmap and executed the following key initiatives:

• Implemented GPT-4.0 for natural language processing, enabling accurate responses to customer inquiries.
• Integrated the bot with the bank’s CRM and trading databases, providing real-time access to ~50 million records.
• Deployed a real-time AI search engine, delivering precise investment recommendations and facilitating account openings.
• Incorporated compliance guardrails, ensuring all responses adhered to financial regulations.
• Enabled seamless escalation to human agents, routing complex queries for personalized assistance.
• Built a high-availability architecture to handle ~10,000 daily interactions with up to 500 concurrent users while maintaining 95% accuracy for standard queries and 85% for advanced trading-related questions.

Tech Stack

Databricks, AutoML, Mlflow, Databricks Vector Search, Delta Lake, ReactJs

Business Impact

• Improved response times by 70%
• Resolved 80% of the queries without human intervention
• Enhanced customer satisfaction

About the company

The fourth largest multi-national pharmaceutical manufacturer by revenue.

Business Need

Pharmaceutical companies rely on vast commercial datasets to drive market strategy, regulatory compliance, and competitive insights. However, accessing the right data quickly remains a challenge. Data science and analytics teams often struggle with retrieving relevant commercial data from complex datasets stored across multiple sources. The lack of an intuitive search mechanism leads to inefficiencies, increasing the time required to identify and utilize key datasets.

Our client had a similar issue with their dataset containing commercial data of medical drugs stored in a Snowflake database. To address this challenge, the client approached Altysys to develop an advanced AI-powered solution that could:

• Streamline the data search and retrieval process
• Enhance accessibility to their commercial drug dataset
• Accelerate and improve data discovery accuracy
• Enable deeper insights for business intelligence and analytics

Solution

Following a comprehensive needs assessment, the Altysys team crafted a solution roadmap and executed these key initiatives:

• Built a metadata-driven search framework within the Snowflake environment to accurately identify relevant databases and tables, ensuring precise query targeting.
• Deployed an SQL agent to automate data extraction, enabling users to retrieve specific data points efficiently through direct queries or predefined conditions.
• Integrated Retrieval-Augmented Generation (RAG) to enhance search accuracy for unstructured data, allowing AI-generated responses to incorporate relevant information from a vector database.
• Implemented cosine similarity search to improve query matching, ensuring that users receive the most relevant datasets based on their search intent.
• Leveraged Azure OpenAI embeddings and stored them in ChromaDB, optimizing data indexing and retrieval for faster, more accurate insights.

Tech Stack

GPT-3.5 Turbo, Azure OpenAI Service, FastAPI, React

Business Impact

• 50-60% reduction in manual effort for metadata creation and searching
• Improved search accuracy and result quality
• Accelerated data discovery
• Improved analytics and insights with AI-powered cognitive search

About the company

A major automotive electronics manufacturer.

Business Need

Customer data plays a pivotal role in strategic decision-making. For automotive electronics manufacturers, leveraging real-time insights is essential to staying competitive. To achieve this, our client sought to build a robust, centralized customer data platform that would provide executives—including AVP, DVP, VP, COO, and CEO-level stakeholders—with meaningful intelligence at a glance.

However, they encountered several challenges. Despite managing a vast dataset of over 10,000 customers, accurately forecasting revenue trends remained complex. The company also struggled with generating real-time queries across diverse metrics, detecting anomalies in operational and market data, and integrating sentiment analysis.

To address these needs, the client enlisted Altysys’ expertise to develop an AI-powered solution that would:

• Seamlessly aggregate and analyze diverse customer data
• Enable real-time visualization of insights through a centralized dashboard
• Deliver actionable intelligence
• Improve revenue forecasting accuracy

Solution

The Altysys team conducted a detailed assessment of the client’s needs and developed a structured solution roadmap. The team executed the plan through the following key initiatives:

• Implemented real-time SQL query generation using GPT-4.0.
• Integrated AI-driven sentiment analysis and forecasting models for predicting churn and revenue.
• Developed an intuitive data visualization framework that consolidates key metrics, including revenue trends, anomaly detection, and sentiment insights.
• Leveraged Azure AI and scalable cloud infrastructure to process high data loads.
• Enabled real-time anomaly detection to identify and flag irregularities in customer behavior and financial trends.
• Designed a scalable platform to support up to 50 concurrent executive users while maintaining high-speed data retrieval and processing.

Tech Stack

Azure OpenAI Service, Power BI, GPT 4.0, MS Azure

Business Impact

• Enhanced executive decision-making with a powerful, real-time customer insights dashboard
• Improved churn prediction accuracy by 15%, allowing proactive customer retention strategies
• 90% boost in revenue forecasting accuracy

About the company

A multi-national pharma distributor company.

Business Need

Pharmaceutical distributors must provide pharmacists and dealers with timely, accurate information on orders, product availability, and compliance updates. However, traditional customer support systems often face delays, human errors, and limited multilingual accessibility.

For our client, an attempt to set up customer assistance or query resolution channel proved ineffective, as navigating a private knowledge base of nearly two million customers—with thousands of lengthy documents—was beyond human capacity. Manual sorting of files, text, and blobs was inefficient, and untrained operators further hindered query resolution. These inefficiencies disrupted operations, delayed decision-making, and created communication gaps between stakeholders.

Therefore, the client approached Altysys to develop an GenAI-powered customer assistance solution that would –

• Automate data processing
• Improve response accuracy
• Provide multi-lingual assistance 24/7
• Enhance distributor-dealer communication

Solution

Following a comprehensive needs assessment, the Altysys team crafted a solution roadmap and developed a smart, mobile-friendly, GenAI-powered customer care bot by executing these key initiatives:

• Built the bot on Microsoft Azure infrastructure, leveraging OpenAI models like ChatGPT+ to handle high query volumes.
• Built a real-time delivery tracking system to keep customers informed and an autonomous system to handle routine queries.
• Integrated real-time case deflection mechanism to route only exceptional cases to human agents.
• Utilized AI-powered NLP to enhance response accuracy and personalization for English, Hindi, and Spanish speakers, and enabled 24/7 assistance across multiple time zones.
• Automated invoice generation for quick and hassle-free transactions and integrated customer feedback collection into the workflow to improve service quality.

Tech Stack

MS, MS Azure, Azure Open AI, GenAI model: Llama, Embedding Model: text-embedding-ada-002

Business Impact

• 85% of customer queries resolved through the bot
• Improved customer experience and positive customer feedback and presence over social media
• Minimized human errors

About the company

A reputed US-based consumer electronics company.

Business Need

Freight, Distribution, and Routing (FDR) facilities handle vast parcel volumes from multiple sources, but fluctuating inflows often lead to inefficient staffing, delays, and higher costs. To improve efficiency and ensure compliance, organizations are turning to AI-enabled solutions to implement data-driven forecasting of next-day volumes. This helps optimize workforce planning, reduce costs, and enhance resource utilization.

Therefore, the client reached out to Altysys to develop a predictive modeling solution that would –

• Manage the flow of parcels
• Streamline staff allocation

Solution

The Altysys team conducted a detailed assessment of the client’s needs and developed a structured solution roadmap. The team executed the plan through the following key initiatives:

• Built four predictive models for each channel—Manifest, Returns, and Inter-Facility—to enhance volume forecasting and operational efficiency.
  • Model 1 – Developed a cycle time prediction model using CatBoost Regressor to estimate parcel transit time from source to destination and aggregate shipment volumes.
  • Model 2 – Designed a volume forecasting model to predict the next day’s incoming shipments by analyzing past induction volumes and in-transit parcel data.
  • Model 3 – Implemented a time-series forecasting model using ARMA-GARCH to estimate incoming facility volumes based on historical induction trends.
  • Model 4 – Deployed a Monte Carlo simulation model with Geometric Brownian Motion to account for uncertainties in volume predictions and improve planning accuracy.
• Utilized Python-based modeling frameworks to build, test, and optimize the solution, ensuring seamless integration into the existing logistics workflows.

Tech Stack

Azure ML Pipeline, Azure Cosmos DB, Azure Storage account, MS Azure, Python

Business Impact

• Built an ML model with 90% accuracy and 84% precision
• Cost savings per facility of over 20% as a result of optimized staff allocation

About the company

A leading US-based healthcare provider.

Business Need

Chronic disease management requires healthcare providers to closely monitor and support patients over extended periods. Clinics handling patients with conditions such as diabetes and depression must track routine visits, manage resources efficiently, and ensure continuous care. However, without a clear prediction of patient visit volumes, clinics face challenges in resource allocation, staff scheduling, and overall workload management.

To enhance operational efficiency, the client enlisted Altysys to develop a AI-powered solution that would –

• Forecast the monthly workload based on repeat patient visits
• Accurately predict visit volumes
• Optimize workforce planning
• Reduce administrative strain
• Improve patient care continuity

Solution

After conducting a thorough needs assessment, the Altysys team developed a strategic roadmap and implemented it through the following key initiatives:

• Analyzed 17,000 patients and 50,000 to 60,000 appointments over a period of 2 and a half years
• Captured a comprehensive list of data elements and parameters, including clinical data from lab reports, patient encounter parameters such as number of doctors and nurses, equipment and medicine usage, and staff utilization and duration, insurance type and status, etc.
• Developed a mathematical modeling approach to optimize the care managers’ time allocation and assess the costs and benefits of Collaborative Care.
• Designed a Markov Dynamic Program to model Collaborative Care at the clinic level, considering an infinite planning horizon.
• Established an objective function that balances total patient QALYs (Quality-Adjusted Life Years) and clinic profitability.
• Categorized patients based on insurance payment structures, resource utilization costs, and disease progression of comorbid diabetes and depression
• Utilized historical patient data to:
  • Estimate the duration patients remain in complex health states
  • Predict resource consumption and associated costs
  • Forecast revenue generated from patient care

Tech Stack

Azure ML Pipeline, Azure Cosmos DB, Azure Storage Account

Business Impact

• Achieved 95% model accuracy
• Boosted revenue per patient by 23%, where revenue per month increased by 4% and profit per month by 38%
• Augmented care manager time per month by 9%
• Improved financial planning with detailed cost and revenue projections
• Simplified treatment policy and workload optimization recommendations
• Optimized quality care and profitability with data-driven decision-making