Geospatial AI: connect location data to decisions
Geospatial AI combines location-related information with modelling to support a defined decision. Earth observation can contribute information about land, buildings or environmental conditions; business and operational data explain how those features relate to your objective. We test the value of each data source and validate whether results transfer to new locations.
Experience: BeSpatial.AI, Fresh Corner and solar siting
BeSpatial.AI is a completed ESA demonstration with HolisticRM as prime contractor and GeoX as subcontractor. The official account describes customer-location affinity using Earth-observation and geodemographic inputs. It reports a 91% good-versus-bad location classification result for a particular financial-services customer. That is a project-specific classification result; it is not a promised sales uplift or a universal accuracy level for new customers.
Our MOL Group / Fresh Corner reference addresses a related but different question: which retail format fits each service-station location. The published company case describes GeoX spatial enrichment, business context and model-based format selection. The OPT4SOL.AI demonstration instead concerns solar-site decision support; it remains ongoing, with GeoX as prime and HolisticRM as subcontractor.
Choose the right spatial question before the data layers
A customer-affinity map, a retail-format decision and a solar shortlist cannot share an unquestioned success measure. We define the target geography, the actionable decision and the relevant exclusions first. Then we compare the added value of location, imagery and business information. More spatial layers may improve context, but they also add acquisition, licensing, integration and maintenance work.
For buyers, the useful demonstration shows a baseline and the enriched approach on unfamiliar locations. Ask where the model is uncertain, which source dates apply and which constraints remain subject to specialist approval.
Choose the approach that fits the decision
| Option | Useful when | Trade-off to test |
|---|---|---|
| Business-data baseline | Existing customer or asset data supports the decision | Coverage gaps and historical selection effects |
| Geospatial enrichment | Neighbourhood or access conditions matter | Resolution, freshness and reliable spatial joins |
| Earth-observation model | Visual or environmental features add useful information | Sensor coverage, clouds, seasonality and validation |
Define the geography, candidate and exclusion rules
Bring candidate locations, historical outcomes, operating constraints and source dates. A Fresh Corner format decision needs commercial context; a solar shortlist needs land, infrastructure and engineering exclusions. Spatial resolution and licensing must fit the decision, rather than being selected for their apparent detail.
Test the contribution of each spatial layer
- Create a business-data baseline for the same geography and decision.
- Align spatial and temporal references before adding imagery or geodemographic features.
- Compare enriched predictions on unfamiliar regions and review uncertain or missing-data cases.
- Deliver a shortlist or map with assumptions, source dates and required specialist checks.
Measure screening value before investment returns
Track useful candidates per analyst hour, exclusions missed and stability under reasonable scenarios. Later commercial outcomes need separate observation. A common geographic evidence base can improve discussion between commercial, engineering and planning teams; it does not remove permissions, site surveys or investment judgement.
Sample: an illustrative evaluation plan
Sample — illustrative scenario, not a client result. For illustration, compare 100 candidate locations using the current business-data baseline and an enriched spatial model. Hold out an appropriate geographic area, then review a 20-site shortlist for missed exclusions, analyst time and uncertainty. Retain extra layers only when they improve the decision. These are Sample design figures, not measured project outcomes.
What you receive
A source and licence register, geographic data specification, baseline and enriched model comparison, uncertainty assessment and a reviewable map or shortlist. The handover explains source dates, resolution, geographic coverage and operational exclusions. Relevant GIS integration is agreed against the user's actual workflow.
Questions buyers ask
- Is geospatial AI only for space companies?
- No. Location-informed decisions arise in energy, retail, banking, infrastructure and other sectors.
- Can a model trained in one region work elsewhere?
- Possibly, but transfer must be tested. Customer behaviour, infrastructure, imagery and data coverage can differ substantially.
- Do more data layers always help?
- No. We retain additional data when it improves a relevant decision enough to justify acquisition, processing and maintenance effort.
The official BeSpatial.AI demonstration record identifies HolisticRM as prime contractor. For OPT4SOL.AI, our role is subcontractor to GeoX in an ongoing demonstration. Explore relevant industry applications or our AI development approach.
Discuss your geospatial AI project. Bring a short, non-confidential description of the decision, current process and available data.
We build AI models for the European Space Agency ESA in various domains such as Navigation (Navisp) and Earth Observation (EO). We use AI built on space assets to determine the best locations for advertising.
BeSpatial.AI helps businesses in financial services and other industries to boost their ads performance with customer trained machine learning model (HolistiCRM).
Training is done based on internal datasets, Earth Observation provided by European Space Agency ESA and GIS data (GeoX). In our first use-case machine learning model accuracy reached 91%.
Machine Learning supported targeting ROI reached 1127% thanks to lower CPC and higher conversion rates targeting the right (potential customer dense) locations.
BeSpatialAI is a B2B service. We are creating value for companies in the financial services and retail sectors through making their digital advertising efforts more efficient. Based on discussion with customers, the direct users of our product will be the departments responsible for sales and marketing across digital channels.
Advertisers are looking for tools to better target their messaging to reach their customers. The current best practice in digital campaign management is to train a new model for each product offering, however this is a time and resource-consuming process and uses internal data that is disconnected from the Facebook/Google ad systems.
Companies often run many parallel digital campaigns without a machine learning component to optimize performance using both internal knowledge and geospatial information. There is widespread recognition that better integration and utilization of data analytics can be a real competitive advantage.
BeSpatial.AI enhances targeting accuracy based on geolocational similarity and customer affinity maps. It reduces the cost and the resources required to generate a new, specific model for a target group because its automated MLOPS process doesn’t require man months of scarce internal resources (IT, BI, CRM) involvement.
Navigation (ESA NAVISP)
We have developed a learning solution to model the EGNOS ground-segment processing computations. The machine learning model predicts EGNOS performance in terms of predicted ionospheric and orbit and clock models (GIVE and UDRE), which can then be translated to predicted protection levels that are important for aviation applications and possible other safety-of-life uses. The project was coordinated by Integricom, HolistiCRM developed the AI model, Iguassu developed the interfaces to EGNOS and UI, GEOX developed extract transfer load modules.
OPT4SOL: Master Utility-Scale Solar Siting
The rapid expansion of renewable energy requires precision at scale. OPT4SOL replaces ad-hoc, manual site analysis with an AI-driven, comprehensive national overview for PV project developers.
Developed in partnership with GeoX and supported by ESA, this all-in-one platform evaluates terrain using close to 70 different spatial data types. By utilizing Copernicus Sentinel-1 and Sentinel-2 imagery alongside critical local data, OPT4SOL trains custom AI models to pinpoint locations mirroring your most profitable existing sites.
Multi-layered Spatial Profiling: Integrates topography, land value, drought indices, and medium/high voltage grid networks into a single dashboard.
Accelerated Permitting: Automatically filters out Natura 2000 protected areas and excluded developmental zones before capital is spent.
Scalable Architecture: Ready for pan-European deployment based on local data availability and regulations.
E-Spacement & Mobiliti: Unlocking EV Infrastructure ROI
In the electric vehicle rollout, choosing the wrong location results in sunk costs. Our analysis reveals a staggering 15-20X performance gap in daily energy dispensed (kWh) between the lowest and highest-performing charging stations.
E-Spacement ensures you only build in the top quartile. Powered by multi-layer spatial data—including GNSS-based traffic patterns, grid constraints, and Copernicus Earth Observation datasets—our machine learning models (using Gradient Boosted Decision Trees and Graph Neural Networks) predict utilization potential with up to 91% accuracy.
- Unmatched Speed: Cuts site planning and validation time from weeks down to days.
- Mobiliti Success: Deployed for Hungary’s largest EV charger network to eliminate non-performing deployments using geospatial index scoring.
- Future-Proof (HDV Module): Actively developing Heavy-Duty Vehicle corridor planning tools aligned with EU AFIR regulations for freight logistics.
Earth Observation Based Next Best Action / Next Best Offer Recommender
Transforming Earth Data into Hyper-Personalized Commerce Moving beyond traditional mass marketing, we fuse aerospace technology with advanced machine learning to deliver exactly what your customers want, driving customer-driven communication based on individual behavior patterns.
The Problem: The “Jam Study” Paradox Presenting multiple generic offers to a non-affinity base wastes your PPC budget. The famous “Jam Study” proved that offering 24 choices attracted 60% of shoppers but resulted in only 3% buying. Conversely, offering just 6 curated choices attracted 40% of shoppers but resulted in 30% buying. Too much irrelevant choice hurts conversion and increases PPC costs.
The EOPAIR.AI Solution Current recommendation engines use only internal data. Our Earth Observation Next Best Action / Next Best Offer Recommender (EOPAIR.AI) breaks this limitation by using both internal customer data (former baskets, available products) and powerful external spatial data.
Overcoming the “Cold Start” Problem: Traditional recommendation engines struggle when there is no historical data for new users or new geographic markets. By leveraging space data, EOPAIR.AI instantly understands the physical, demographic, and mobility context of an environment, enabling highly accurate predictions and hyper-personalized offers from day one.
Unprecedented Depth: We integrate micro-level geodemographic data, mobility data, shop distance/catchment areas, and Earth Observation imagery.
Rich Spatial Layers: The system utilizes 103 distinct data layers , including 15 high-resolution (10m x 10m) ESA Sentinel-2 RGB satellite layers.
Automated MLOps Pipeline: Our automated data processing and machine learning pipeline cuts campaign preparation time from the standard 3-6 months down to just 2 weeks.
Proven Use Cases & Digital Growth Success Stories
Foodpanda (Netpincér): The “Next Best Action / Next Best Offer” Engine We implemented a powerful Next Best Action / Next Best Offer Recommendation Engine for Foodpanda to answer a crucial question: What is the Next Best Offer?.
Using a “Matchbox Recommendation” approach, the engine successfully combined collaborative filtering, geodemographics, and content-based filtering.
The Result: Delivered highly targeted recommendations that achieved a 2.5X Email Open Rate and a 3.5X Click-Through Rate.
Global Recognition: This innovative approach even caught the attention of Microsoft CEO Satya Nadella. Speaking to over 1,000 people at the Microsoft Future Decoded conference in Hungary, Nadella highlighted Netpincér as an inspiring example of democratizing AI and cloud-driven intelligence. According to Microsoft’s official news site, Nadella praised the machine learning models for being so effective at predicting customer cravings that 80% of users order the exact food that was recommended to them.