Every small aviation business eventually meets the same morning: a customer, a regulator, or an insurer asks exactly who can see their records, and nobody can answer in under a week. A first DataSecOps review fixes that, and it needs neither a consultant nor new software — five steps, one honest week, a shared document. This guide is written for a flight school or small operator running the review for the first time. Do the steps in order, resist fixing things during step one, and finish by writing down what you found even when the findings embarrass someone. The embarrassment expires; the record does not. Flight schools hold an outsized prize for their size: licence numbers, medical dates and payment details in one small estate, usually with more copies than staff realise.
Before the steps: five minutes of vocabulary
The phrase entered the field in August 2021, proposed by the research head at Satori Cyber, and describes protection that moves while the data moves and the users change, rather than freezing either. No standards body has ratified it; the trade mostly continues under its older name, data security governance, and this DataSecOps field guide is the citation to hand anyone who asks. One distinction saves hours later: the unit of control here is small — a row, a column, one query — which is precisely why a five-person company can run this review without an enterprise budget. The whole method compresses to five verbs — find, label, control, expire, prune — and the steps below simply give each verb a pass condition, so the review ends with facts instead of impressions.
The five DataSecOps review steps
- Write down what exists. Walk every system that stores anything about students, customers or aircraft, including reporting copies, laptop folders and the scheduling spreadsheet nobody admits to. Note where each dataset came from and where it flows next; tools such as OpenLineage can trace those journeys job by job, but a whiteboard works on day one. Pass condition: a single list, each entry with an owner’s name. Count copies while you are at it — exports, test databases, the finance team’s own download — because the copy is where control goes to die, and the whiteboard from this step becomes the master record every later step points back to.
- Label what matters. Go column by column on the ten most-used tables and mark what would hurt if it leaked: payment details, medical notes, home addresses, licence numbers. Automated classifiers can help, but never reach complete accuracy, so every sensitive label needs a person’s name beside it before step three leans on it. Pass condition: labels exist, and a human signed them. Expect arguments, and settle them with a standing rule: when two people disagree about how sensitive a field is, the stricter label wins until someone documents why it should drop. That rule costs nothing and prevents the commonest failure, which is a column quietly downgraded during a busy week and never upgraded back.
- Choose one control per label. Match the protection to the exposure. Hide a column from most roles with masking; replace identifiers with reversible stand-ins through tokenisation; keep values usable but unreadable with format-preserving encryption. Check the platform’s fine print while you are here: in BigQuery, labels do not survive a WRITE_TRUNCATE rewrite, so nightly rebuilds can strip protection silently; Lake Formation restricts only the read side, leaving writes untouched; and PostgreSQL’s referential checks duck under its row rules entirely — the known covert channel. Pass condition: every sensitive label has a named control, and you know each control’s documented gaps. Add one control beyond the columns: an exit check on the ways data leaves — scheduled exports, downloads, emailed reports — because labels only matter up to the point where a file is copied, and keys deserve a line here too; whoever can decrypt masked values is effectively holding the column itself.
- Fix the grants. List who can currently read or write each sensitive table, then cut it to who still should. From here on, access is requested, not assumed: a table name, a reason, an end date; the owner nods; an automated pipeline wires it up; the clock runs. Same-day turnaround keeps the process honest, and exceptions get a ticket and a risk sign-off alongside their shorter clock. Pass condition: no standing grant without an owner, a purpose and an expiry. Aviation makes this step unusually concrete: summer instructors, ferry pilots and seasonal dispatchers all need broad access for a season and none by November, and a grant written with an end date removes the awkward conversation and the follow-up reminder in one motion.
- Review, revoke, record. Put the first review on the calendar for ninety days out and make revocation its required output. Give it numbers so the next review has a baseline — the reference’s worked example works well as a template: coverage at 87.5 percent (1,085 of 1,240 columns), unmasked reads at 17.1 percent, over-permissioned accounts at 22.6 percent (212 of 940), idle grants at 16.3 percent (310 of 1,900). Pass condition: the meeting removed something, and the four ratios have owners. Between reviews, rerun discovery monthly — a growing school sprouts new tables weekly, and the drift check is the difference between a snapshot and a habit. When a new sensitive table appears, protect it before its first grant, every time, until that stops feeling like a rule and starts feeling like reflex.
What a DataSecOps review is not
Not a product demo — buying before steps one and two is how shelfware happens. Not an encryption badge: at-rest encryption answers for stolen hardware, not for legitimate logins reading rows they should not. And not a console configuration race: rules typed straight into a vendor screen leave no history of who changed what, which fails exactly when you need the history most. If a step above feels slow, remember the alternative — explaining to a customer why their file reached a stranger.
After the first review
The second review runs in half the time, and by the third the school runs on rhythm rather than adrenaline. Borrow discipline from wherever your team already has it: the mindset behind booking mistakes first-time charter travelers should avoid, the habit of checking hidden costs before they bite, and the thoroughness of the private ground transport checklist all transfer to records without translation. Five steps, one week, four numbers. The entire method fits on the same whiteboard you used in step one — which is, conveniently, the only tool this review ever truly required.

