At HolisticRM, we’re proud to showcase a portfolio of forward-thinking projects that push the boundaries of AI and machine learning. Our references include cutting-edge solutions in geolocation, location-based AI recommendation systems, branch visit and store sales prediction, as well as cross-industry affinity models and AI innovations in manufacturing.
For largest CEE oil-retail service stations
Fresh Corner Optimal location and format identification with
GeoX spatial data enrichment and Machine Learning models.
Solution: Molgroup as part of its 2030 strategy increases its retail presence at their service stations. The objective of Machine Learning model was to identify based on mainly external geolocational information the right retail Fresh Corner format for every service station. That fits micro locational target market. The model among few hundreds of GeoX features takes into account profitability and Nash equilibrium related variables. For the right restaurant format fast food chain locations and performance.
Technology: Solution and results has been delivered with Azure ML, Power BI, Azure Datafactory
Customer: MolGroup largest CEE Oil Company
Business Domain: Retail Network Expansion
Delivered Solutions
• Holistic AI Algorithm Ensemble
• GeoX 100×100 catchment area analysis
• Competition analysis Nash Equilibrium
• Foursquare Google based Bayes Average Visitor Prediction
• Automated machine learning model training
Hyper-personalized recommender engine
Delivery Hero/“Net Pincér” Email Resposne Rate Boost and
Recommendation Engine with Machine Learning
Solution: NetPincér is the largest food delivery service wanted to increase its email campaign response rate. To support this, we have built multiple Machine Learning binary classification models. Based on the success of initial response rate analysis we have developed complete recommender engine to increase click thru rates on top of increased open rates with personalized offers. We leveraged Azure Matchbox Machine Learning recommending model boosted with GeoX geolocational indexes.
Solution: selected the best models and provided an automated learning scoring process.
Result: A+2.5X Open Rate, +3.5x Click Thru Rate Presented by Satya Nadella as case study at Microsoft ConferenceTechnology:Azure ML, Power BI, Azure Datafactory
Customer: Netpincér / Foodpanda / Delivery Hero
Business Domain: E-commerce, marketing
Key Deliverables
• 250% Open Rate increase
• 350% Click Through Rate increase
• Endorsement by Satya Nadella CEO of Microsoft
• Automated Data Processing
• Machine Learning predictive response model (champion challenger model selection)
• Automated retraining
• Automated hand off file generation with specific output format for email sending
• Power BI Dashboard track emails, map visualization, campaign tracking
Customer level branch visit prediction for largest CEE bank
Predicting number of customer branch visits with Machine
Learning on a customer and branch level based on internal
transactions and Geolocational features OTP Bank Data
Discovery Challenge
Solution: Predicting retail banking customer and branch pair level visits based
on past transactions, channels, and product ownership information (34 Features). As part of feature enhancement, we have enriched initial features with geodemographical (GeoX) indexes and added calculated features. As a result of increased number of (1483) features managed to improve accuracy significantly. Multiple models has been trained for different subsegments. We have combined for each segment different models and approaches into an ensembled model.
Result: In global competition with 40+ participants launched by OTP (largest retail bank in CEE) within 6 days 5th position globally 2nd place among Hungarian teams.
Technology: GeoX 100×100 Geolocational databases. SQL, SPSS, Azure ML
Customer: OTP Bank
Business Domain: Digital, X-Sell, Billboard advertisment
Key success criteria
• Feature enhancement from 34 features to 1483 GeoX features
• Space time distance calculations
• Workplace identification based on transaction k nearest neighbor centroid
• Applied gravity model
• Holistic Machine Learning hyper tune
• Holistic Algorithm Ensemble
• Collaborative filtering
Affinity based cross-sell for inbound and outbound in banking
Citibank Next Best Offer – Customer Affinity based offering.
Solution:Next Best Offering framework covered whole lifecycle of retail banking
customers. NBO framework covered all the main product (20+) and service offering. We have built separate customer affinity models for each product and service. This enabled offer hierarchy in inbound and best mix offering for outbound channels. This helped maximizing revenues with given outbound sales capacity. The Machine Learning framework also supported preferred channel mix selection.
Results: 6X+ Booking Rate improvement. EMEA X-Sell award
Technology: SQL Server 2008R2, Analysis Services, Integration Services, Database and Reporting Services
Customer: Citibank (EMEA Region)
Business Domain: X-Sell, Callcenter capacity optimization, marketing cost, revenue and channel mix optimization
Key Features
• Optimized inbound and outbound Cross sales performance
• Training of Analysis Services 27 Machine Learning models
• Automated Score and campaign eligibility scoring
• Continuous test and control monitoring
• Data Driven campaign success reporting
• Self service reporting multidimensional cubes
• Incentive calculation
Credit Card Retention
Card Retention Program
Machine Learning based Retention Program
Solution: State of the art customer lifecycle holistic retention model. Customer
Retention offers based on customer level churn/attrition probability and future value. Customer future value calculated based on affinity for further products and product lifetime revenue. Machine Learning influenced the effort and value based incentive structure. Automated scoring and reporting provided transparency and continuous management control.
Technology: SSQL Server 2008R2, Analysis, Integration, Database and Reporting Services
Customer: Citibank (Hungary, Romania)
Business Domain: Segment, Product & Portfolio management
Key Deliverables
• 250% Open Rate increase
• 350% Click Through Rate increase
• Endorsement by Satya Nadella CEO of Microsoft
• Automated Data Processing
• Machine Learning predictive response model (champion challenger model selection)
• Automated retraining
• Automated hand off file generation with specific output format for email sending
• Power BI Dashboard track emails, map visualization, campaign tracking
Customer potential based AI retention
Personal Installment Loan Retention Program
Machine Learning Retention System
Solution: Machine Learning model based Attrition probability, combined with
machine learning based top-up response rate ML models. Framework covered post
lifecycle customers with a special Machine Learning model built around “Back 2 Bank”
campaign responsiveness for ex Personal Installment Loan customers. Solution
included model for best time to call. On top of the models the whole Retention
Program Reporting, effort and value driven incentive was part of solution.
Technology: SQL Server 2008R2, Analysis, Integration, Database and Reporting Services.
Customer: Citibank Hungary
Business Domain: Customer Product and Portfolio management
Key Features
• Machine Learning predictive models (Early Repay, Back2Bank,
Top-up, Grandfather)
• Campaign Management
• Portfolio management, Controlling
• Effort Driven (1- Affinity score) Incentive calculation
AI optimized outbound capacity
Cofidis Next Best Offer – Customer Affinity based offering.
Solution: Next Best Offer framework covered all X-Sell acitivities of retail Cofidis customers. NBO framework covered all the main products (Renewable Loan, Low Cost Loan, Debt Settlement Loan) involved in XSell offering. We have built separate customer affinity models for each product offering and service. This enabled offer best mix offering for outbound call centers. This helped maximizing revenues with given outbound sales capacity. The Machine Learning framework also supported preferred channel mix selection and replaced post contact protocol rules with preventive contact protocols.
Results: 40% Booking Rate improvement vs control group.
Technology: Python 3.6 Scikit Learn, IPython Notebooks, XGB model, Logistic Regression, LGBM, Ensemble models. Tensorflow 2.0 Keras
Customer: Cofidis Hungary
Applied Domains: X-Sell, Callcenter capacity optimization, marketing cost, revenue and channel mix optimization
Business Domain: Retail Network Expansion
Key Features
• Training of 3 sets of ensemble Machine Learning models with GeoX Indexes
• Score and campaign eligibility scoring
• Customer Future Value
• Next Best Action Calculation
• Campaign Enhancements
• 40% Booking Rate improvement vs control group
Finding best locations for electric vehicle charging stations
Electric Vehicle Optimal Charging Location
Solution: We built a machine learning model to predict performance of a potential location for a charger based on past chargers performance and surrounding area geospatial characteristics (GeoX index database). We have built multiple models and selected the best performing models in terms of accuracy. This enabled customer to validate a new location before investing into deployment of a charger to a potential non performing location. GeoX built an mobile application where end user can pinpoint on the map to verify location score (1 best location, 0 no body will charge there).
Technology: Anaconda, Python 3.7 Scikit Learn, IPython Notebooks, XGB model, Logistic Regression, LGBM, Ensemble models. Tensorflow 2.0+Keras, DNN, LSTM, Flask
Customer: Mobiliti (Largest EV charger chain in Hungary)
Applied Domain: EV charger location optimization
Key Features
• Training Ensemble Machine Learning models with GeoX Indexes
• API for real time Scoring
• GeoX Heatmaps
Customer visit and sales prediction for retail stores
Rossmann Customer visits and store sales prediction
Solution: Rossmann operates over 3,000 drug stores in 7 European countries. Currently, Rossmann store managers are tasked with predicting their daily sales for up to six weeks in advance. Store sales are influenced by many factors, including promotions, competition, school and state holidays, seasonality, and locality. With thousands of individual managers predicting sales based on their unique circumstances, the accuracy of results can be quite varied.
Results: 5% accuracy increase with external weather data
Technology: R Studio, Ensemble models
Customer: Rossmann Germany
Business Domain: Store Sales Prediction
Key Features
• Training Ensemble Machine Learning models
• Leveraging external School holiday / Weather features
Location based beer consumption in HoReCa
Borsodi Pub marketing support strength based on consumption
potential.
Solution: Predict Borsodi brand product consumption market potential of different location type (Hotels, Restaurants, Pubs) based on 229 location attributes: Days since open, format, opening hours, location unemployment rate, income, age group, photos taken nearby.
Results: 78% Accuracy just based on external variables.
Technology: Python 3.6 Scikit Learn, IPython Notebooks, XGB model, Logistic Regression, LGBM, Ensemble models. Tensorflow 2.0 Keras, CNN
Customer: Borsodi Hungary Hungary
Applied Domains: Marketing support based on potential.
Key Features
• Training of 3 sets of ensemble Machine Learning models with
GeoX Indexes
• Heatmaps based on consumption
• Key variables drivers of “good locations”
• Decision Tree
• Ensemble model
Insurance cross-sell for utility customers
Insurance X-Sell Machine Learning Model.
Solution: Predict response of electricity or natural gas customers to family accident insurance offer. Machine learning models built based on previous responses to various campaigns and product owned and services used. Feature set has been enriched with GeoX spatial, geodemographic indices. Separate models built for based on product ownership for specific customer segments.
Results: 17% Booking Rate improvement.
Technology: SQL 2019, Python 3.6 Scikit Learn, IPython Notebooks, XGB model, Logistic Regression, LGBM, Ensemble models. Tensorflow 2.0 Keras, CNN, LSTM
Customer: Energia Hungary Applied Domains: X-Sell, Call-centereting capacity optimization, marketing cost, revenue and channel mix optimization
Key Features
• Training of 3 sets of ensemble Machine Learning models with GeoX Indexes
• Score and campaign eligibility scoring
Predicting signal errors in ESA EGNOS system
Predicting GPS signal errors based on Satellite Location
Solution: We use machine learning to model the EGNOS ground-segment processing computations. The model will be used to predict EGNOS performance in terms of predicted ionospheric and orbit and clock models (Grid Ionospheric Vertical Error/GIVE and User Differential Range Error /UDRE), which can then be translated to predicted protection levels that are important for aviation applications and possible other safety- of-life uses.
Technology: TensorFlow, Convolutional Neural Network
Customer: European Space Agency (ESA).
Applied Domains: Computer Vision, Deep Learning, Image Classification
Key Features
• Predicting EGNOS GIVEI values based on Satellite location, space weather and local weather
• Predicting UDREI values based on Satellite location, space weather and local weather
Boost ad performance with earth observation based AI location profiles
BeSpatial.AI: Micro Locational Ads Targeting Optimization
Solution: BeSpatial.AI provides the locations to our Clients where they are the most likely to reach their target audience so they can target their advertising efforts more efficiently.
The system does this by analysing – through the use of Artificial Intelligence – the locations of existing customers and finding new locations that have similar spatial characteristics. When doing the analysis and comparison the system takes into account very large numbers of variables that can be used to describe a given location.
These can include sociodemographic variables, land use, mobility patterns, distances from POIs, earth observation data, housing types, etc.
We have successfully developed a set of machine learning models to classify and identify good and bad locations with a stunning 91%+Accuracy distinguishing good and bad locations from customer density perspective.
Technology: TensorFlow, Convolutional Neural Network
Customer: European Space Agency (ESA)/Cofidis.
Applied Domains: Computer Vision, Pattern Recognition, Deep Learning, Image Classification
Key Features
• GEOX GIS Layers (68 layers)
• ESA, images, remote sensing data, indexes (35 layers)
Opt For Sol
All-in-one tool for site selection developed for the needs of the
Hungarian, Czech and German market.
Solution: OPT4SOL is an AI-driven site selection tool designed for utility-scale solar projects in Hungary. It integrates satellite imagery, GIS, and over 70 spatial data types to identify optimal locations quickly and accurately. Users can create custom AI models and conduct manual analysis with built-in GIS tools. Developed by GeoX and HolsitiCRM with ESA support, OPT4SOL streamlines solar development and offers a scalable methodology applicable to other European markets. The solution was developed by GeoX and HolsitiCRM in cooperation with, and the support of the European Space Agency (ESA).
Technology: AI-GIS platform
Business Domain: Solar site development
Key Deliverables
• AI-powered platform for detecting and selecting optimal utility-scale solar sites and tested in multiple markets.
• Integration of 70+ spatial data layers from Hungarian and EU sources for comprehensive analysis.
• Custom AI model feature to identify similar high-potential locations based on user input.
• Built-in GIS tools for advanced manual analysis and visualization.
• Scalable methodology adaptable to other European countries with available data.
Purchase Order Delay Prediction
Solution: Manufacturers depend heavily on timely deliveries to maintain efficient supply chain operations. Currently, human vendor managers predict purchase order delays based on limited historical data and intuition, leading to inconsistent accuracy. Our AI solution analyzes over 11,000 variables, including vendor performance, logistical patterns, seasonality, and external data, to reliably predict whether purchase orders will be delayed. It also categorizes delays into specific time frames for precise supply chain management.
Technology: Mixture-of-Experts AI Model, Extensive Feature-rich Dataset, Delay-Time Classification
Customer: Global Manufacturing Companies
Business Domain: Supply Chain Management, Procurement Optimization
Key Features:
• AI model predicts delays with 74% accuracy, compared
to 44% accuracy by human managers.
• Achieved a 168% improvement over manual predictions.
• Training Ensemble Machine Learning models,
• Leveraging extensive internal and external datasets, including seasonality and logistical features
Results
• AI model predicts delays with 74% accuracy, compared
to 44% accuracy by human managers.
• Achieved a 168% improvement over manual predictions.
Automated Testing Code Generation with LLM
Solution: In manufacturing, the manual creation of automated testing scripts from extensive product specifications (often exceeding 50 pages across multiple documents) is highly resource-intensive and prone to errors. Our solution leverages advanced Large Language Models (LLMs) to automatically convert these detailed specifications into over 60,000 lines of accurate testing code in under 20 minutes. Utilizing multiple LLM models and tailored fine-tuning methods, the solution efficiently interprets technical documentation while ensuring the integrity and consistency of critical IDs and configurations during code generation.
Technology: Multiple Large Language Models (LLMs), Document Processing Framework,
Function-specific Tooling
Customer: Global Manufacturing Companies
Business Domain: Quality Assurance, Automated Software Testing
Key Features:
• Multi-LLM model integration
• Specialized fine-tuning of language models
• Automated document processing
• Function-specific tooling
• Integrity preservation of IDs and configurations
Results
• 60-80% reduction in manual coding efforts for automated testing.
• Rapid generation of 60,000+ lines of testing code within 20 minutes.
• Significant improvement in test script accuracy and consistency.
Best Practices Identification with Product Similarity Clustering
Solution: Manufacturing companies with multiple sites often encounter variability in efficiency and best practices. Our AI-powered clustering solution identifies similarities among products across different sites by analyzing Bill of Materials (BOM) descriptions through multidimensional clustering. The system uncovers opportunities for cost reduction, productivity enhancement, and standardized practices by highlighting the most effective methodologies within each cluster. Additionally, it provides insightful impact analyses concerning cost and time improvements achievable by adopting identified best practices.
Technology: Machine Learning Clustering Algorithms, NLP for Material Description Analysis, Impact Analysis Tools
Customer: Global Manufacturing Companies
Business Domain: Production Optimization, Best Practice Standardization
Key Features:
• Multidimensional clustering based on BOM material descriptions
• Product similarity analysis across multiple manufacturing locations
• Cost-time impact analysis
• Identification of actionable improvement opportunities within clusters
Results
• Clear identification of cost-saving and productivity-enhancing opportunities.
• Data-driven insights for standardizing best practices across multiple manufacturing sites.
• Improved efficiency and reduced variability in production processes.
Industry : E-commerce
In the e-commerce sector, we significantly boosted order numbers with our conversion-optimized campaigns. We run cross-border campaigns as well.
Customer: ROKSH, Mogyi
Used platforms:
• Google Ads – display, search
• Meta – facebook, Instagram
• Tiktok
Google Ads – Display
Industry : Manufacturing, Services
In the manufacturing sector, we saw a substantial rise in website traffic, with a large portion of orders now coming from our online campaigns.
Customer: EDELHOLZ, Solar Express, HAN Spaces Hungary
Used platforms:
• Google Ads – display, search
• Meta – facebook, Instagram
• Pinterest
Google Ads – Display
Facebook – Google
Industry : Healthcare
In the healthcare industry, our campaigns led to notable revenue growth as well. We continuously monitor and optimize our campaigns, making recommendations for budget adjustments, keyword changes, and creative swaps as needed.
Customer: ReproGenesis, Széchenyi Thermal Bath,
Pesterzsébeti Jódos-Sós Gyógy- és Strandfürdő
Used platforms:
• Google Ads – display, search
• Meta – facebook, Instagram
Google Ads – Display
Google Ads – Display
Industry : Recruitments
We can effectively apply online campaigns in the recruitment sector as well. We can acquire leads at a fifth of the cost compared to traditional headhunting firms.
Customer: UNIQA, ROKSH, EDELHOLZ, NFSZ, EURES Hungary
Used platforms:
• Google Ads – display, search
• Meta – facebook, Instagram
• LinkedIn
Marketing Tableau Dashboard focusing for Personal loan product
Erste marketing department requested a comprehensive report on the personal loan application process, starting from which landing page the interested party originated, all the way to the final status of the loan request, including how many days the application was reviewed and the amount the customer received.
Customer: Erste Bank Hungary / Smart Design Digital
Used datasources
• Google Ads
• Google Analytics
• Erste Credit system
Marketing Power BI Report
WellWay Clinics aimed to integrate all its systems into a single dashboard to track the source of patient arrivals (e.g., campaigns, walk-ins), analyze doctor statistics, monitor call center activities, record examination results and associated revenues, and assess the outcomes of sent emails, among other metrics.
Customer: WellWay Clinics
Used datasources
• Google Ads
• Google Analytics
• Social (Meta)
• Patient traffic data
• Call Center system
• CRM system
Marketing Power BI Report focusing for Google reviews
MBH Bank aimed to increase both the quantity and quality of its Google reviews. To achieve this, they launched a competition among their branches. Our task was to create a report that would allow them to monitor each branch’s ratings, detailing how many reviews each branch received and the nature of those reviews, as well as how the ratings changed compared to the situation before the competition.
Customer: MBH Bank
Used datasources
• Google My Business
• MBH branch data
