Innovation
How work moves from a research hypothesis to something a customer pays for.
Terravek reinvests 17.4% of revenue in research and development. This page describes the path a capability takes from hypothesis to product, because the interesting question is not how much a company spends but how it decides.
The five stages
1. Hypothesis (Labs)
A written hypothesis, a pre-registered success criterion and a named scientist. Reviewed at six and eighteen months. Most work stops here and that is the intended outcome — four projects were stopped in 2025.
2. Demonstrated method (Labs)
The method works on real data, reproducibly, with a quantified uncertainty. It is published. At this point it is a paper, not a feature, and the honest position is that many good papers never become good features.
3. Productionisation (Labs + Engineering)
A research software engineer works with the scientist to make the method run unattended against arbitrary customer data. This is where most of the value is lost across the industry, and it is the reason Labs has research software engineers rather than borrowing them. Why.
4. Limited release (Product)
Shipped to a small number of customers who know they are early, with a model card documenting what is not yet known. The point of this stage is to find the failure modes that only appear in a real estate.
5. General availability
Documented, supported, in the quality model, and in the deprecation policy.
Recent examples
| Capability | Hypothesis | GA | Elapsed | Now in |
|---|---|---|---|---|
| Persistent scatterer interferometry | 2019 | 2022 | 3 years | ORBIS, TIDEWATCH, GRIDSCAN |
| Species-aware growth forecasting | 2018 | 2021 | 3 years | GRIDSCAN |
| On-board screening inference | 2022 | 2025 | 3 years | ORBIS 4, FIELDKIT |
| Nitrogen stress model (East Africa calibration) | 2022 | 2025 | 3 years | AGRISENSE Programme |
| Sub-pixel waterline extraction | 2020 | 2023 | 3 years | TIDEWATCH |
The consistent three-year figure is not a coincidence and not a target. It is roughly how long it takes to be confident enough in a measurement to put it in front of someone making a decision with it.
Patents
Terravek holds 23 granted patents, principally in sensor fusion, optical design and calibration methods. We do not assert patents defensively against open-source implementations of published methods. Licensing enquiries to legal@terravek.com.