Ask ten people what to learn first for a data job and you'll get ten confident answers, most of them autobiography. People recommend the route they happened to take, or the one they're currently selling a course on. Neither is dishonest. Both are a sample size of one.
There's a better source, and it's sitting in plain sight: the postings themselves. Employers write down what they want, in public, thousands of times a day. Luke Barousse's datanerd.tech scrapes those listings daily and publishes the share that mentions each skill, split by role. It isn't a survey of opinions. It's a count of what hiring managers actually typed.
Two of its reports are worth sitting with: Data Analyst, drawn from 83,517 postings, and Data Engineer, from 178,738. What they show isn't shocking. It's just noticeably different from the advice.
What employers actually list
| Skill | Data Analyst 83,517 postings |
Data Engineer 178,738 postings |
|---|---|---|
| SQL | 44.7% | 68.4% |
| Excel | 31.9% | — |
| Python | 30.7% | 66.4% |
| Power BI | 28.7% | 15.0% |
| Tableau | 22.1% | — |
| R | 15.6% | — |
| Azure | 6.2% | 39.7% |
| AWS | 5.6% | 41.4% |
| Snowflake | 5.2% | 25.7% |
| Spark | — | 37.6% |
| Databricks | — | 24.6% |
| GCP | — | 23.6% |
| Airflow | — | 17.2% |
| Kafka | — | 17.0% |
| Git | — | 11.7% |
| dbt | — | 10.6% |
| Average skills per posting | 4.5 | 6.8 |
Plenty jumps out of that table. But the most useful number in it isn't a skill at all.
The number nobody quotes
Average skills per posting: 4.5 for Data Analyst, 6.8 for Data Engineer.
Skills reports usually get read as shopping lists — which tool is winning, which one is dying. The per-posting average is more useful than any of that, because it describes the shape of the door rather than the contents of the room. It's a barrier metric.
A listing that asks for 4.5 things is something a focused person can meet inside a year of deliberate work. A listing that asks for 6.8 — SQL and Python and a cloud and Spark and a warehouse and an orchestrator — assumes you've already been inside the building. Same market, different admission price. That's roughly half again as many skills, and each one you add is another way to fail the screen.
It also explains the pay gap without needing a theory about it. Data Engineer pays more than Data Analyst for the same reason it's harder to break into: it's a wider ask, and breadth is what's scarce. The salary isn't a reward attached to the title. It's the price of the checklist.
Excel is #2, and everyone pretends otherwise
Excel appears in 31.9% of Data Analyst postings. That puts it ahead of Python (30.7%), ahead of Power BI (28.7%), and well ahead of Tableau (22.1%). Only SQL beats it.
This is the finding people most want to argue with. Excel is unglamorous, it doesn't version well, and nobody is building a personal brand on it. All true, and none of it moves the ranking. The postings aren't expressing a taste.
Skipping Excel is a filter you apply to yourself. Close to a third of the listings name it, and declining to learn it removes you from that third in order to protect a preference no hiring manager shares. The dismissal also tends to misread the word. On a lot of these postings, "Excel" doesn't mean SUM and a pivot table — it means Power Query, Power Pivot, a data model somebody has to maintain, and DAX that has to be right. That's the same modeling instinct you'd apply in dbt, running in the tool the finance team already trusts. Learning it properly isn't a step backward. It's the shortest path to seeing how a business reasons about its own numbers.
Python isn't the gate. SQL is.
SQL: 44.7%. Python: 30.7%. That fourteen-point gap is an entry strategy, if you let it be one.
Python is genuinely useful and I'd never argue against learning it. But it isn't the wall between you and a first analyst job, and there's a wrinkle inside that 30.7%: a good share of it sits on mid-level, senior, and data-science-adjacent listings — the same ones asking for statistics and modeling. That's a reading of the postings rather than a measured split, so hold it loosely. The direction is clear enough. As a share of the jobs a beginner can realistically win, Python is asked for less often than the headline number suggests.
The entry data points the same way. Roughly 85% of Data Analyst listings don't specify required years of experience, and about 39% are explicitly entry-level. That's a market hiring on demonstrated skill rather than tenure — good news, but only if you demonstrate the skill it's actually screening for. Treating Python as the gate is how people spend four months on pandas tutorials while their SQL stays shallow enough to break in the first technical screen.
The highest-paying role has the fewest doors
Analytics Engineer pays a median base around $158K. It's also where I want to end up. And it's a bad role to apply for first: the openings are few, and the ones that exist expect prior experience. Analytics Engineer is structurally a role you get after someone has already trusted you with a warehouse.
The pattern generalizes. Pay and accessibility pull against each other, because part of what the high number buys is the experience you don't have yet. Choosing your first target by compensation is how you spend a year applying to the smallest set of doors in the market.
Build broad, apply narrow
Once you've seen both lists, the temptation is to learn everything on them. That's how you end up with 40% of twelve skills instead of 90% of four — a profile that reads as unfocused to every screen it touches.
A recruiter filters by one role. The first pass is a job title with a checklist under it, run by someone holding a hundred other resumes and no reason to reconstruct your range from context. Breadth still matters: an analyst who understands what happens upstream asks better questions, and that breadth is what makes the next role possible. But breadth is what you grow into. It isn't what you lead with. You lead with one role and the evidence for it.
Where I'd start
Concretely, with the postings in front of me:
- SQL until you're not bluffing. Joins, window functions, CTEs, and knowing the grain of a table before you write against it. It's #1 for both roles, and it never stops being the thing you're paid for.
- Excel, properly. Power Query, a real data model, DAX. It's #2, and it's the fastest route to understanding how a business talks about its numbers.
- One BI tool, deep. Power BI (28.7%) if you're aiming at Microsoft shops, Tableau (22.1%) for startups and tech. One of them. Deep.
- Then Python, once you have something worth automating.
Cloud and warehouse skills come later on the analyst track — Azure, AWS and Snowflake sit at 6.2%, 5.6% and 5.2% of analyst postings. They're worth having, and you'll pick them up building the next thing, but they aren't what's keeping you out.
That's four things. The average posting asks for 4.5. That isn't a coincidence — it's the whole argument.