Decision-Ready Nature: Applying AI to What We Can Measure, and Act On Today
Session Description
Nature-positive commitments are increasingly running into a practical question: what can organisations actually measure reliably today, and what do they need to measure to act with confidence? This session, hosted by Gentian, set out to test that question directly, bringing together practitioners across ecological consultancy, corporate sustainability, nature finance, and policy to work out what's genuinely workable now versus what still needs further development.
Gentian opened with its own origin story, tracing back to co-founder Dusty Gedge's seven years spent manually counting green roofs across London to prove out the city's Green Roof Policy, before running through case studies of AI applied to high-resolution imagery: mapping habitat types in domestic gardens across six UK cities, tracking biodiversity uplift on a development site over seven years, detecting invasive sitka spruce on a Scottish peatland, and using AI to shortlist plausible endangered species on a site based on habitat type alone. Attendees then split into five groups (biodiversity net gain, sustainability reporting, nature restoration and rewilding, nature finance, and urban planning) to stress-test where AI could genuinely help their own work today, and where the gaps remain.
Speakers
Dusty Gedge, Co-founder, Gentian
Karen Day, Founder & COO, Gentian
Eleanor Thomson, Lead Technologist, Gentian
Watch the Session Recording
Key Takeaways
Gentian's own framing: AI's biggest value in high-resolution imagery work right now is speed and consistency at scale, not new insight. What one person could barely do manually for a single city in a month, AI can now do across much larger areas.
The BNG group's clearest point: AI is best used to speed up existing ecologist workflows (pre-survey assessments, monitoring over time), not to replace ecologists doing habitat assessment on the ground. Trust in AI-derived data has already taken some knocks in the industry, so positioning matters.
The sustainability reporting group flagged a structural problem more than a technical one: translating landscape-level, community-collected data into the standardised formats corporates need for frameworks like TNFD and CSRD remains a gap AI hasn't closed.
The restoration and rewilding group kept circling back to the same operational problems: which metrics actually indicate success, how to assess unfamiliar or hard-to-access sites before committing resources, and how to secure real community buy-in rather than assumed buy-in.
The nature finance group's discussion centred on trust and explainability rather than capability. AI-derived outputs are seen as opaque compared to field observation, and cost savings from AI aren't automatically matched by comparable gains in accuracy.
A recurring worry across groups: AI can't tell you whether the underlying reference data was ever any good. One example that stuck: a lot of forest carbon modelling still traces back to a single 1970s wood sample, repeatedly cited by later studies. AI applied on top of a weak baseline just scales the weak baseline.
Multiple groups landed on the same tension: AI is strong at collecting and correlating data at scale, but the moments that matter (reaching conviction, making a funding decision, persuading a stakeholder) still rest with people.
One capacity point raised in discussion: for a lot of organisations, the limiting factor isn't the tool, it's whether the organisation has the internal literacy to use it well. Otherwise these tools get tried once, look promising, and sit unused.
A smaller but concrete idea from the restoration group: using AI to digitise old handwritten ecological survey notebooks, unlocking historical data that was never digitised in the first place.
