Anyone who runs data pipelines for a living already performs a good share of a traditional IT operations group’s work. You get paged when a load fails, argue about whether a schema tweak needs a review, and keep a mental list of which tables feed which dashboards. The difference is mostly vocabulary and inheritance. This comparison puts ITOps and the day-to-day of a data team side by side, so you can borrow the habits that fit and skip the ones that do not.
First, what we are comparing
The Ops Atlas from xOps School defines the discipline in its ITOps reference page as the group that keeps technology services available. Its remit: answer user requests, respond to outages, dig out whatever keeps causing the same breakage, gate alterations to live systems and keep a current inventory of what the estate contains. The usual working style borrows from ITIL-flavoured service management.
Notice that this is a function, meaning people and roles, rather than a toolkit or a methodology. The Atlas separates the three on purpose. The function is ITOps, the practices are ITSM, and the tooling that watches systems is ITOM. Data teams tend to blur all three into “the pipeline”, and that blurring is the source of a lot of confusion when a data engineer sits down with an infrastructure colleague for the first time.
Round one: what counts as an incident
On a classic operations desk, an incident means a service that stopped or slowed without warning. The measure is whether people can still do their work. In a data team the equivalent is often subtler. A job can succeed and still deliver wrong numbers, so the pipeline is green while the business reads nonsense.
That gap is worth studying, because the Atlas flags a related trap: treating the event and monitoring layer as a non-critical dependency, so that a silent failure hides an outage. For a data engineer the lesson transfers neatly. Your quality checks are production, too. If the check that guards a table quietly stops running, you have lost your alarm without anyone noticing. A write-up on validating data with an open-source expectations library covers how such checks are usually built, and the operations lens simply asks who owns them and how you know they are alive.
Round two: change handling
Change is where the two worlds feel most alike and most different. An IT desk sorts changes into three types, as the Atlas lays them out. Standard changes are pre-authorised and low risk. Normal changes are assessed before they proceed. Emergency changes go immediately, such as an urgent fix for something already under attack.
Data teams usually have only two modes, merge it or do not, with the review living inside a pull request. The ITOps view adds a useful middle idea: some changes deserve a recipe that is approved once and then reused. Adding a column to a staging table in a known pattern is a candidate for a standard change; rewriting a core revenue model is not.
The Atlas offers a worked example of scoring. A change coded CHG-302 reached a score of 6, built from 3 points for its tier, 2 for touching a schema and 1 for landing outside the agreed window. That total sent it to a committee. A data team can reproduce the spirit of this in ten minutes: list three or four risk factors, give each a weight, and decide in advance what total triggers a second pair of eyes.
There is a debate worth knowing about. The Atlas reports that the software-delivery research programme found no evidence that outside approval lowers change failure rates and recommends peer review instead. It also points out that a European financial-sector regulation still expects an independent approval function and after-the-fact approval of emergency changes. Two different documents, two different expectations; if your data platform serves a regulated firm, check which one governs you.
Round three: knowing what you have
Operations teams keep a configuration database. Data teams keep a catalogue, or more often a collection of wiki pages and tribal memory. Both exist to answer the same question: if this thing breaks, what else is hurt?
The Atlas gives a small health check for such a register. In its ten-item lab, nine records were fully filled in, two had sat untouched beyond sixty days, one pointed at nothing, and one reused another item’s serial number. Replace “serial number” with “table name” and the failure modes read like a data catalogue audit, with entries nobody has touched, entries pointing at nothing and entries that appear twice.
The warning against trying to capture everything applies in both camps. The Atlas says a register should hold what decisions require. For a data team that might mean the handful of tables that feed finance reports and customer-facing features, not every scratch table created during a long weekend.
On the infrastructure side, the Atlas also mentions that a widely used set of security controls puts hardware and software inventory first. The ordering is a hint that knowing the estate is a foundation for everything else.
Round four: promises and clocks
Data consumers care about freshness: is the dashboard ready by nine? That is a service level in disguise. The ITOps vocabulary offers a tidy way to build it. A promise to the business (SLA) rests on internal promises between teams (OLA) and on promises from outside suppliers (UC).
The Atlas arithmetic example adds up an eight-hour commitment from a desk triage step, a diagnosis step, a carrier step and a buffer. For a nightly pipeline, the parallel chain might be ingestion, transformation, quality checks and a refresh of the reporting layer, each with its own owner and its own allowance. When the final promise is shorter than the sum of the parts, you know the number is aspirational.
Clock rules matter just as much. One Atlas lab tracks a ticket open for 92.5 hours on the wall clock, of which a mere 4.5 hours were billed to its target, since weekend and holiday pauses stopped the count. If your freshness promise silently excludes weekends, say so in writing, or someone will be surprised on a Monday.
Another view worth borrowing comes from an overview of orchestrating data workflows with an asset-based tool, which shows how dependencies can be made visible. Visibility of dependencies is what makes the service-level arithmetic possible in the first place.
A quick side-by-side
Before the frameworks, here is the comparison at a glance. Use it to prompt a chat at your next retrospective, not to settle an argument.
- Unit of failure: a service outage on the desk, a wrong or late dataset in the data team.
- Review style: a tiered, scored route for changes versus a single pull-request gate.
- Register: a configuration database versus a catalogue, both prone to rot if nobody audits them.
- Promise: response and restore times versus freshness and completeness.
- Typical pitfall: process added after every failure versus checks added after every bad number.
Frameworks, briefly and without the hype
Teams often ask which framework to adopt. The Atlas offers a short, neutral lineup. ITIL 4 is optional guidance with 34 practices. ISO/IEC 20000-1 is the one you can be certified against. COBIT 2019, from ISACA, frames governance as 40 objectives. FitSM is a free lightweight standard that came out of an EU project. A tiny data team will rarely need more than a few borrowed ideas from any of them, and plenty of teams get by on one page of rules they wrote themselves after a bad week.
The warning from the Atlas is worth repeating: imposing a heavyweight service management setup on one team running one service is an anti-pattern. If your whole operation is five pipelines and three people, adopt a lightweight register, a change score and a written clock rule, and stop there.
So who wins the comparison?
Nobody, and that is the useful answer. The desk side has decades of refined habits for sorting, scoring and recording. The data side has newer instincts for testing and automation. A friendly rule of thumb: take the operations desk’s discipline about ownership and clocks, and bring your own habit of testing everything.
And one last nudge on vocabulary. Whenever someone says “DORA” in a meeting, ask whether they mean the research on software delivery or the European regulation. The Atlas warns the two are easily confused, and a clarifying question saves a lot of crossed wires.