THE SKILLS AN HR ANALYST NEEDS BEFORE THEY NEED PYTHON

Job advertisements ask for Python first. Six other capabilities decide whether an HR analyst is useful, and they come before any programming language.

Emmanuel Duncan · 7 September 2026 · 6 min read

The people analytics job advertisements in this region increasingly ask for Python, SQL, and a machine learning framework. The training market has followed, and I meet HR officers halfway through a data science course, working on a public dataset about passengers on a ship that sank in 1912, wondering how it connects to the establishment list on their desk.

Programming is worth learning. It is also the fifth or sixth thing that makes an HR analyst useful, and teams that hire for it first end up with someone who can fit a model and cannot say whether the headcount is right.

Here is the order I would put them in.

One: Knowing What the Record Means

An HR dataset is a set of administrative traces. Every field was entered by somebody, for a purpose, under some pressure, at some moment.

The analyst who is useful knows that the appointment date field is sometimes the date of the letter and sometimes the date of first attendance. That grade changes are backdated after promotions are approved, so a promotion appears in the record months after it happened. That the unit code changes when the finance system is reorganised, which looks like a wave of internal transfers. That casual staff are sometimes entered with a placeholder identifier that is reused.

None of this is visible in the data. It is learned by asking the people who enter the records, and it is the difference between an analysis that describes the organisation and one that describes the filing.

This is the capability that transfers least well from a general data course, and the one that HR officers moving into analytics already have more of than they think.

Two: Turning a Vague Request Into an Answerable Question

Requests arrive as topics. "Can you look at attrition." "We need something on productivity." "The board wants to understand the age profile."

The work is a conversation that turns that into a question with a decision attached, a defined population, a defined period, and a stated comparison. Attrition among whom, over what period, compared with what, and what would you do differently depending on the answer.

An analyst who cannot run that conversation will produce technically correct work that answers nothing. This shows up in interviews if you ask a candidate to take a vague request and question you about it, which is worth more than any technical exercise.

Three: Arithmetic You Can Defend

Most HR analysis is counting, rates, ratios and simple comparisons. Doing that carefully is harder than it sounds and it is where most errors live.

Choosing a denominator and being able to say why. Average headcount against opening headcount changes a turnover rate materially. Knowing what a percentage of a small number is worth, and refusing to report a rate for a team of six. Understanding that an average tenure figure is dragged around by a handful of very long serving staff and that the median usually tells you more. Recognising when a change in a rate came from the numerator or the denominator, which is a question nobody asks and which explains a surprising share of mysterious movements.

None of this requires statistics beyond secondary school. All of it requires the habit of checking.

Four: Spreadsheet Work at a Real Standard

I would rather have an analyst who is genuinely strong in a spreadsheet than one who is halfway competent in a programming language, because the spreadsheet is where the organisation's data already lives and where the analyst's colleagues can follow the working.

Real standard means pivot tables without hesitation, lookup functions that fail loudly rather than silently, dates handled properly, a clean separation between raw data, working and output, and a file another person can open and understand. It also means knowing the limits, which usually appear when the same manual clean has to be repeated monthly.

That last point is what eventually creates the case for learning to code, and it is a much better reason than an advertisement asking for it.

Five: The Ethics and the Law of the Material

HR data is about identifiable people, much of it sensitive, and an analyst who is casual with it will cost the organisation more than they contribute.

The practical parts. Knowing what may be circulated and at what level of aggregation. Setting a minimum group size for reporting and holding to it. Understanding that health, disciplinary and grievance material sits under tighter restrictions than headcount. Knowing what Ghana's Data Protection Act, 2012 requires of the organisation as data controller, including the duty to hold personal data only for stated purposes.

And a judgement that is harder to teach: recognising when an analysis is technically possible and should not be produced, because the finding will be used to identify individuals rather than to change a system.

Six: Presenting to People Who Do Not Want a Chart

The last capability is getting a finding into a decision. A short written summary. One number that carries the argument. A chart that a director can read in fifteen seconds. An answer to the question of what should be done, offered as a recommendation rather than left implicit.

Analysts trained technically often present their working, because the working is what they are proud of. The audience wants the conclusion, the confidence they can place in it, and the choice in front of them.

Where Programming Does Belong

Once those six are in place, code earns its position quickly. Repeating a monthly build without hand editing. Joining files too large for a spreadsheet. Keeping a written record of exactly how a figure was produced, which is the underrated benefit and the one that makes analysis reproducible.

At that point, learn SQL first, because most of the work is retrieving and joining records. Then whichever of Python or R suits the environment. Learn them against your own organisation's data rather than a public teaching dataset, because the six capabilities above are exactly what a teaching dataset removes.

Where to Start

If you are building a team, write the job description around the first three items and test them in the interview with a real request and a messy extract.

If you are the analyst, take one figure your organisation reports every month and trace it back to the records it comes from, field by field, asking whoever enters each one what it means. That exercise will teach you more in a fortnight than a term of coursework, and it is the foundation the rest sits on.

Emmanuel Duncan is the founder of Aletheia Lunara Consult, a professional education and consulting practice working with organisations across Ghana and beyond on people analytics, workforce research and organisational capability.