Why Earth Observation projects stall in the data layer
Ludovic Auge
CEO, Dataionics · · 6 min read
The bottleneck is not AI. It is not analytics.
Earth Observation is finally meeting enterprise demand. AI applied to geospatial data, new satellite constellations, and a wave of use cases across agriculture, insurance, climate, energy and defense are pulling EO out of the research lab and into production engagements. Integrators and consulting firms are being asked to deliver, and they’re hitting a wall that has nothing to do with models.
The wall is the data layer. Every EO project goes through the same hidden grind before any real work can start: sourcing data across fragmented providers, building ingestion pipelines from scratch, harmonizing incompatible formats and resolutions, and handling preprocessing that has to be redone for every client. This is not glamorous work. It rarely shows up in proposals. And it quietly consumes 3 to 6 months before a single analytic runs.
The economics: how data engineering eats margin
The numbers are stark. RAND reported in 2025 that 42% of enterprise AI initiatives were abandoned, and independent benchmarks from MIT and Gartner converge on the same finding: 60 to 80% of AI budgets are consumed by data engineering, not modeling. In GIS teams, field data suggests roughly 90% of effort goes into data preparation rather than delivery.
For ESN and consulting firms, this has a direct P&L impact. Sector analysis for 2024-2025 shows EBITDA margins for integrator firms compressed to a five-year low around 9.8%. EO engagements routinely run 60 to 150% over budget. The root cause is almost never the analytic. It’s the invisible pipeline work below it.
Most integrators don’t lose projects on delivery. They lose margin in the data layer.
Reinforcement, not replacement
The partner model we propose rests on a clear separation of roles. The integrator owns the client relationship, the use case definition, the application layer and the business value. Dataionics owns the data layer: sourcing EO data across providers, orchestrating ingestion pipelines, preparing and harmonizing datasets, and delivering reliable inputs directly into the integrator’s cloud environment.
This is not a replacement model. It is a reinforcement model. The integrator keeps the client. The integrator keeps the application. The data logistics are industrialized below, once, and then reused across every engagement.
What “industrialized data layer” actually means
The phrase is thrown around too casually. In practice, industrialization of the EO data layer means four things:
- Discovery. Unified access to 2,500+ indexed EO collections across Copernicus, NASA, USGS, and commercial providers, without navigating disparate catalogs and authentication systems.
- Acquisition. Multi-source retrieval with automated handling of rate limits, retry logic, and provider-specific quirks that derail hand-built pipelines.
- Harmonization. Alignment of projections, resolutions, formats and naming conventions so data from different sources can be used together without manual reconciliation on every project.
- Delivery. Analysis-ready datasets indexed by geography and time, shipped in formats (GeoTIFF, COG, S3, API, GeoJSON) that plug directly into existing analytic stacks.
None of this is exotic engineering. What’s exotic is doing it once, running it as a managed layer, and amortizing the cost across every client engagement rather than rebuilding it for each project.
From pilot to scale: the path for integrators
The onboarding is deliberately short. Step one is a direct conversation, 20 minutes, no pitch, to review current EO projects and data constraints. Step two is picking a concrete use case from the existing portfolio where the data layer is the acknowledged pain point. Step three is a pilot: deploying an adapted data layer on that specific scenario so the integrator’s team can evaluate the integration path on real work. Step four is scaling the data layer across additional projects and clients.
The firms that win the next phase of EO will not be the ones with the best models. They will be the ones with the most scalable data layer. Regulation is tightening (EUDR, ESG, climate compliance). Demand is accelerating. Data complexity is exploding. The window to industrialize is now.