The situation
One person owned all of it. Product decisions, vendor contracts, the cost line, whether the platform is up at 9 a.m. on a Monday, and the support request from a learner who cannot get in. No engineering team to escalate to, no operations team to absorb it. Whatever broke, broke on the owner’s desk.
The subject matter set the stakes. Aviation English is not general language learning; people study it because their work requires it. Tolerance for a platform that loses their progress or locks them out is close to zero.
The real difficulty in a small e-learning product is never the content. It is the administration around it: enrollment, access, records, scheduling, the customer who paid and cannot see the course, the report someone needs at month end. That work grows quietly, and it is what consumes an owner with no team. Every tool added was a permanent line the product had to carry.
The decision
Buy and integrate, do not build.
That call made the rest of it survivable. A custom platform was the more interesting option and the wrong one. The failure mode a sole owner cannot afford is being the only person alive who understands a codebase. Vendor tooling has a worse ceiling and a much better floor: someone else patches it, and replacing a vendor is a migration rather than a rewrite.
The second decision was sequencing. The administrative workflow got fixed before the product grew. New features on top of a broken back office produce a bigger broken back office. Administrative load is the real constraint on how many customers a one-person operation can hold.
Rejected: bespoke software, a longer feature list as the answer to every problem the product had, and any tool that could not be removed later without taking the customer records hostage.
What was built
Clear Sky Aviation English went from concept to production as a working service. Content delivered to learners. An administrative path from enrollment through access and records. A defined route for customer problems. A vendor stack contracted to sit inside what the product could pay for. Service operations continued after launch, which is what separates a product from a demonstration.
Germinar RH · OBPI ran on the same instinct in a different domain. A psychological-assessment experience used in commercial-pilot evaluation moved into a digital workflow, where the constraints are the process rather than the interface. The material is sensitive, the order of the steps is not decorative, and the output has to be defensible. Design work there was mostly removing chances to do it in the wrong sequence.
How it was proven
Proof at this size is operational rather than statistical. The questions were plain. Does the platform stay up during the hours learners actually use it? Does an enrollment complete without a human rescuing it? Is the vendor stack still inside what the product can carry? Those were reviewed continuously, because the person reviewing them also absorbed the answer.
Honesty about the instrumentation matters more here than a flattering summary. Success measures were not defined formally at launch, which is exactly why no numbers appear on this page. The discipline that arrived later, defining the measure before claiming the result, was learned partly here.
What it changed
The platform reached production and then ran as a live service, inside an engagement that held from March 2019 to August 2022. The same person owned discovery, implementation, vendor management and ongoing operations. Customer and administrative workflows got better while tooling, suppliers, reliability and cost all stayed on one desk. The assessment workflow shipped alongside it.
What it changed for the work that followed still pays. Owning every layer of a small service teaches vendor selection and the cost of a bad integration. It also teaches the difference between an outage that annoys a customer and one that costs them something.
What I’d do differently
Define the measure at launch. That is the first one and the expensive one. Years of real operating history sit behind this project. Almost none of it can be quoted, because nobody wrote down at the start what good would look like.
Write the administrative workflow down before choosing the tooling. Vendors were selected against a workflow that lived in one person’s head. That works right up until a vendor has to be replaced and the workflow has to be reconstructed from memory and invoices.
Document the vendor map and the runbook as if handing them over next month. Sole ownership is a single point of failure, and carrying that quietly is a choice rather than a fact of life. Two pages would have made the service survivable without its owner.
“If you do not define the measure at launch, you do not get to claim the result later. That one cost me an entire engagement’s worth of evidence.”