A Charter Desk, a Half-Empty Load and What the Team Changed

A Charter Desk, a Half-Empty Load and What the Team Changed

The story below is invented. No real operator, crew or client appears in it, and every figure about the imaginary team is left vague on purpose. Its purpose is to show how a DataOps mindset changes the way a charter operations desk finds and handles data trouble. Where I cite numbers, they come from a published reference, and I say so.

The morning the availability board looked fine

Picture a small charter brokerage. Each night an automated job gathers aircraft availability, positioning legs and crew duty windows into one table, and the sales desk reads that table at the start of the day to quote clients. The job runs at an early hour, finishes without complaint and turns green on the scheduler.

On one particular Monday, the job finished green and the table was missing a large share of its rows. Upstream, a partner feed had delivered a partial file. A salesperson quoted a client a repositioning leg using the thin table, found no suitable aircraft, and told the client the date was unavailable. Two hours later a colleague, looking at a different report, noticed that several aircraft were in fact free. The client had already called a competitor.

The post-mortem was short and uncomfortable. The job had succeeded. The data had not.

What the team learned from the reference

The team’s analyst hunted for a framework and landed on a DataOps reference in the Ops Atlas. Two ideas from it landed straight away. The first was that data differs from code because it changes on every run, with no deploy, so each load needs its own checks. The second was an uncomfortable statistic in a worked example: four incidents added up to 43.75 hours of outage, and three quarters of it (32.75 hours) went on simply noticing what was wrong. Consumers raised the alarm first on half of them.

The analyst read that as a mirror. The desk’s problems were not slow repairs. They were slow discoveries, and the people discovering them were clients.

Change one: put the nightly job in version control

The scripts had lived in a shared folder, edited in place. The team moved them into a repository, asked for a review on every change, and recorded an owner for each table. That is the change loop, the part of the discipline closest to ordinary software work.

Change two: check every arrival

The second change addressed the run loop. For the availability table, the team added three checks. Was the table ready before the sales desk opens? Was the row count within the usual band for that day of the week? Were aircraft identifiers present and unique? The Atlas notes that a monitor based on three standard deviations suits one stable process, whereas a business with a weekly rhythm has more than one. The team therefore kept a separate baseline for weekends.

Change three: audit before publishing

The team reserved a stricter pattern for this one table, since it feeds live quotes. New rows are written to a hidden copy, the checks run there, and only a clean load replaces the table the sales desk reads. A failed load is held back, and the desk keeps the previous day’s version with a visible banner showing its age. The Atlas names the pattern write-audit-publish and warns of its cost: added latency and, when publishing means copying, double writes. Because of that, the team applied it to one table, not to all.

Change four: agree what the partner feed promises

The partial file was the real culprit, so the team wrote a short agreement with the feed’s provider, covering which fields would be present, what the status codes mean, how many aircraft the file should cover and by what hour it would land. The Atlas describes this kind of producer-owned data contract as a way to stop silent upstream changes. It adds that structure-only checks cannot catch a change in meaning, which is why the agreement included definitions as well as field names.

What changed in the weekly rhythm

Each Friday, the team reviewed a small log of every data problem, noting how it was found, how long it took to notice and how long to fix. Over the following months the most useful number was simply the proportion found first by someone outside the desk. The team’s goal was to push that proportion down, which meant more problems caught by monitors before a client could hear about them.

Other aviation teams reading similar material have wondered how far to go. A sensible limit comes from the tier model in the Atlas: critical pipelines get contracts, blocking audits and an on-call owner, while exploratory analyses get version control and nothing more. A charter desk’s pricing analysis for a possible new route can live in the third tier happily.

The wider lesson

Operators far from aviation wrestle with the same pattern. For a take on how a small business protects something precious from silent failure, see the article on keeping family money secure with a simple app. And for readers who want to build technical skills around these ideas, a guide to growing a career through a tech training school is a reasonable starting point.

Back at the imaginary desk, the green light on the scheduler no longer settles any argument. The test is whether the numbers on the board would survive a skeptical look. Since the changes, they usually do, and when they do not, the people who learn first are the ones who can fix it, not the client holding the phone.