The services trap
A drone services business sells flight hours. Every contract starts from zero. The data it captures belongs to the customer (rightly), and the next customer, even in the same industry, pays for a fresh capture. The business is linear in pilots, linear in aircraft, linear in calendar. It has no asset that grows with volume.
Every drone operator eventually hits this ceiling. Revenue scales. Margin does not. The fifth year looks like the first year, with more overhead.
What changes when it becomes a platform
A platform operator captures the same flight data but retains three assets that compound.
A labelled ground-truth corpus. Every survey that is reconciled against a weighbridge, a drill result, or a geophysical observatory produces a labelled (input, ground-truth) pair. That corpus trains feature-detection and change-detection models that the first customer could not access. The tenth customer inherits the work of the first nine.
A reproducible processing pipeline. A services business reprocesses from scratch for every project. A platform commits a processing recipe that is hashable, auditable, and replayable. Every recipe run costs less than the one before it. Every bug fix improves all historical outputs.
An integrated delivery surface. A services business hands over a PDF. A platform hands over a persistent, queryable view into the customer’s assets that updates every cycle. The platform becomes the system of record for the measured asset.
Why the moat is real and non-obvious
Skeptics argue the moat is weak — anyone can fly a drone; processing software is commodity; LLMs will commoditise interpretation. All true at the component level and wrong at the system level.
The moat is not in any component. It is in the joint asset — the labelled corpus, tied to the processing recipe, tied to the delivery surface, tied to the customers who continue to generate new labelled data by operating the platform. No competitor can recover the first 18 months of accumulated ground truth without doing the work from zero, in a market where the early customers are already locked in.
How to tell, as an investor
Five tests that separate platform from services in this category.
First, does revenue per pilot-hour rise year-over-year, or stay flat? Flat is services. Rising is platform, because software leverage is displacing human time.
Second, does the company retain processed outputs across customer cycles, or start from zero each engagement? Retention is necessary for compounding, and customer contracts that prevent it are a structural ceiling.
Third, do the customers log in to view data, or receive it in an email? Engagement surface is proof of platform product. Email delivery is services with a veneer.
Fourth, is there a ground-truth dataset that improves the product, documented and versioned? If yes, the company has the data asset. If not, it does not.
Fifth, is the second-year customer cheaper to serve than the first, holding everything else constant? If yes, the platform thesis is real. If no, it is a story.
The Flybi frame
Flybi is built as a platform in this specific sense. The corpus — aeromagnetic surveys reconciled against drill results, stockpile measurements reconciled against weighbridges — is retained, versioned, and used. The processing pipelines are hashable and replayable. The delivery surface is software, not PDFs. The thesis is testable against the five tests above, and we welcome that test.
Related reading
From reading to doing
Bring us a real site, a real block, a real question.
Most of our engagements start here — a reader who became a customer inside one scoping call.