Skip to the content
Compass
AboutWhat can I enter?Sign in

The guide

Close

Data & AI

Computer Science

Finding patterns in data and building the models behind predictions and AI.

This suits you if…

You are curious about what a number actually means, patient with data that arrives inconsistent, and as interested in whether an answer changed a decision as in whether the model was clever.

Look elsewhere if…

You are here for the modelling. Most of the week is finding, cleaning and plumbing data, and a correct answer can be politely ignored by the organisation that asked for it. If that would grind you down, building software keeps more of the making and less of the persuading.

What the work actually is

What the work is actually like

Most of the week isn't modelling. It's finding the data, discovering it is inconsistent, deciding what a row actually means, and building the pipeline that keeps it arriving. Then a model, then the harder part: convincing a colleague who doesn't trust statistics that the number is real, and finding out whether it changed any decision at all.

The catch

The proportion is the catch. Practitioners routinely describe most of the job as data cleaning and plumbing rather than the modelling that attracted them. Analysis work can also be quietly ignored: you produce a correct answer, and the organisation does what it was going to do anyway. And this is the area where the tooling is being rewritten fastest, so today's specific skills have the shortest half-life on this list.

What people get wrong about it

That it is mainly about choosing clever algorithms. In practice the wins come from better data and a sharper question; a simple model on good data beats a sophisticated one on bad data almost every time.

The jobs inside this area

  • Data analyst
  • Data scientist
  • Machine-learning engineer
  • Data engineer
  • AI researcher

A list, not a recommendation. You narrow it, we don’t.

What this kind of work usually offers

Earning well · Making thingsA generalisation about the sphere, not a promise about a salary. Pay and security vary enormously by country and employer.

How you get there

The one-line version

CS, statistics or maths. Data and ML contests (Kaggle, Zindi) are how you prove it early.

Stage by stage

  1. While you are still at school

    Maths, especially statistics and probability. That is the part that transfers to everything. Learn Python and pull a real dataset about something you care about, then answer one honest question with it.

  2. What you study

    CS, statistics, maths or an applied science with heavy data work. Statistics is the durable core; the frameworks are learnable in weeks, but knowing whether a result means anything is a degree's worth of training.

  3. How the first years actually go

    Usually analyst before scientist: dashboards, reports, and answering other people's questions. That isn't a detour. It is how you learn what the organisation's data actually represents, without which no model of it is trustworthy.

Where this work lives, and how you reach it

What you would study for it

Each says what the first year is really made of, and who should study something else instead.

16 countries · 31 cities · 36 institutions named for this field. Each one states its catch as well as its appeal.

Without leaving home

3 of the routes that need no visa and no move lead into this field.

Test it this month

Free, and finishable in a few evenings

Enter a beginner Kaggle or Zindi competition and submit something bad deliberately, just to complete the loop once. Then improve it twice and notice which change helped. That is the whole feedback mechanism of the field.

Do the job for an afternoon, before you pick a degree

Employers build these to recruit, which is why they are honest about what the work is. You do the real tasks, unpaid and ungraded, and find out whether you can stand it. Free, no deadline, nothing to win.

  • BCGabout 5 hours

    Clean a messy client dataset, build a model that predicts something the business cares about, then explain it to people who will not read the code. That last part is the one everyone underestimates.

  • Deloitteabout 2 hours

    Read a spreadsheet a client actually sent, find the thing in it that is wrong, and turn it into a chart somebody can make a decision from.

  • Tata Consultancy Servicesabout 3 hours

    Turn a business question into a set of charts, then defend the choices: which cut of the data, which chart, what you left out.

These are on Forage. Search it by the employer’s name — what each one is called changes, and who built it does not.

Open Forage from the catalog

These are ways to try Data & AI before committing to it.