Perspective · 22 Aug 2026
Wildlife monitoring's bottleneck isn't the camera
The cameras and the recognition models are getting good. The chain from field footage to a record you can still trust after several seasons is the part that isn't.
It's easy to look at wildlife monitoring right now and conclude it's a solved problem. Cameras are cheap and everywhere, recognition models label species in seconds, and every season produces more footage than the last. But what a park, a province, or a research group actually needs isn't a pile of labeled images. It's a monitoring record: comparable across sites and seasons, auditable, and owned by the organization, that still makes sense after the people who built it have moved on. That record is the immature part. The sensors have raced ahead of the system around them. The field has gone data-rich; the management that should act on it is still decision-poor.
It's not that nothing exists
To be fair, this isn't a greenfield, and anyone who says it is hasn't looked. Good tools exist: research platforms that manage camera-trap studies, detectors that find an animal in a frame reliably, and an open standard, Camtrap DP, for moving data between them. What's missing isn't one more piece of software. It's the whole of it working as a layer an organization actually owns and runs: intake, effort, review, provenance, and export tied together, and still standing after the people and the seasons change. Most of what exists is research-first, or tied to one vendor, or assumes someone technical will glue the steps together by hand. The parts are here. The operational whole, mostly, is not.
More cameras is not more management
A camera is a sensor; a monitoring programme is an operation. The sensor keeps getting better: higher resolution, faster triggers, detection on the device. The operation, getting footage in reliably, keeping it organized, knowing what arrived and what quietly stopped, checking the uncertain calls, and turning all of it into numbers that hold up over years, has barely been built as a product. Most of that work still happens by hand, on someone's laptop, in a spreadsheet only they understand. Adding cameras makes the pile bigger; it doesn't make the record better.
Where the chain breaks
Ask the people who run camera-trap programmes where the time goes, and it's rarely the cameras. It's data management: the volume of images, the classification, the archiving, the quality control, the parts that rarely make it into a paper or a report. A handful of failure points show up again and again.
- Backlog. Collection outpaces processing. Drives fill; comparable records never form.
- Lock-in. Footage lives in a manufacturer's app or a closed cloud, in a format that's hard to get out of. Change tools and the history doesn't come with you.
- Missing effort. The counts get recorded; the survey effort behind them, which cameras were live, and for how long, often doesn't. Without effort, numbers can't be compared honestly.
- Weak export. What comes out is a CSV a vendor invented, not something another tool or a national repository can read.
- Seasons that don't connect. Each campaign is its own island. Staff change, the method drifts, and year three can't be lined up against year one.
In Europe the picture is more fragmented still: monitoring schemes differ by country, transnational integration is partial, and raw data is often neither standardized nor easy to reach. The camera-trap community has converged on open exchange formats like Camtrap DP precisely because the practice underneath isn't settled yet. Standards and governance tend to appear while a field is still sorting itself out, not after it's finished.
What "mature" would look like
A mature data-driven monitoring setup wouldn't be exotic. It would just do the unglamorous things reliably:
- footage arrives on its own, from any camera, and you can see what stopped arriving;
- it's organized by site and time, not by whoever downloaded it;
- uncertain identifications get a human check, and the correction stays with the record;
- survey effort travels with the counts, so seasons are actually comparable;
- every detection carries its provenance: which model, which version, confirmed by whom;
- and you can export the whole thing in an open format, at any time, because it's yours.
None of that is a research breakthrough. It's operational plumbing, and it's mostly missing.
Where we fit
That gap is what Untamed Data is built for. Not an ecology consultancy, not an AI demo, but the operational layer between the field and the systems you report into: intake over FTPS or by manual upload, organization by project and site, the ops view of what arrived and what went quiet, review where it's needed, metrics that hold over seasons, and export that includes the open camera-trap standards. Cameras stay your choice; the data stays under your organization's control; hosting is in the Netherlands by default. Recognition is a normal step in that pipeline, not the headline. It's software for real monitoring programmes, not a network to join.
The cameras are the easy part. Building the record they feed into, so it still means something after the grant ends and the team changes, is the work that's left. That's the part we're building.
