How data analysts use AI to solve their tasks

My name is Stas Zolotarev, I'm an author of the courses «Data Analyst» and «Data Science Specialist» at Yandex Practicum. In this article I will tell you:

• what knowledge and skills an analyst needs to use AI for basic tasks without compromising quality,

how neural networks can be used in internal work processes.

Let's figure it out.

But first — the key idea for understanding the place of neural networks in the analyst's work:

AI is not a "magic button," but a technical tool, the same as Python or Excel. To be useful in work, an analyst still needs professional skills.

Important about analysts in movies and in life

This section is for those who are just starting their professional path as a data analyst or are still thinking about coming here. If you are already experienced or have worked in this field for at least half a year — you can safely skip it, you already know this.

The work of an analyst is always working with data in one form or another. It sounds interesting, but at the start (especially when a person is just coming into the profession) sometimes a dissonance between expectations and reality can arise.

Look, we are often shown this picture in movies: there is a big company, it deals with serious things, everything is professional and everyone is busy — analysts, programmers, managers, and so on. And some conditional analyst is assigned a task: study the data and find the relationship between different events or draw conclusions based on events that have already occurred.

Next, our cinematic analyst with a thoughtful look studies articles, draws tables with a marker on a glass partition, walks around the office, and squints out the window, mentally turning over connections and multidimensional data structures. Then he accidentally notices some small detail — and thanks to it, the whole picture immediately comes together. He circles a few lines on the printouts and triumphantly enters the boss's office, showing with his entire demeanor that he has solved this puzzle.

But in reality, things are usually quite different.

Data analyst work is not a Sherlock Holmes game, where everything is decided by insight and chance, but a methodical and purposeful effort, which often looks less glamorous from the outside than in the movies. Yet it is precisely this thoughtful, methodical, and sometimes routine approach that allows data analysts to solve the tasks set before them, and for companies to achieve business goals. Analysts are paid for this — for understanding, with the right approach and knowledge of their tools, what is happening now (and why), and drawing conclusions about what could happen.

In reality, what is shown in movies is more likely the work of a highly qualified analytics department specialist. In real life, such people also exist in companies, but they usually rely on information that has been prepared in advance by data analysts. Without them, their work would be many times more difficult.

What knowledge does an analyst need to effectively use AI

Now let’s talk about the knowledge and skills that data analysts most often use when working with AI. If you understand and master them at a good level, the efficiency and quality of analytical work with AI will increase many times over.

Understanding SQL and aggregation logic

A data analyst works with data — which is logical and follows from the name of the profession. And data is almost always stored in databases, and with about 90% probability it will be something SQL-like (and knowing the principles of working with SQL makes it easy to understand other DBMSs and their queries). From this follows the first important skill: an analyst must be able to "extract" the necessary data from the database and know how it can be processed in the workflow.

Beginner analysts often do this: they make a request to a neural network like “Write me a database query that will select information from these tables based on these criteria”. It sounds logical, but this is a risky approach.

AI can easily generate an SQL query, but without understanding the logic of aggregations, an analyst might not notice errors: duplicates, incorrect groupings, or invalid filters. By understanding GROUP BY, JOIN, COUNT, SUM, AVG, an analyst can accurately describe the task to the model and quickly check the result. Sometimes, this is even faster than asking the neural network and then verifying whether it did exactly what was needed.

For example, AI might calculate the average check as AVG(price) without considering the number of items in the order. An analyst knows that instead, they need to calculate SUM(total_price) / COUNT(order_id). Yes, this can also be described in a query, but from experience, I can say that it is much faster to do all this manually without having to double-check a machine.

Knowledge of business metrics and how they are calculated

AI often doesn’t know the context of your business or the specifics of your product, even if you load all your internal documentation into it (which is definitely not recommended, at least for corporate information security reasons). It can calculate a metric and it may look plausible, but it will have nothing to do with reality.

A classic example: if AI outputs Retention at 120%, even an analyst with basic knowledge will immediately understand that there is a logic error somewhere, because retaining more people than arrived is impossible.

Neural networks know nothing about your tasks or what the company you work for actually does. They often rely on some average values from the knowledge corpus they were trained on — but that doesn’t mean those averages apply in your specific case.

For example, the market average may consider a normal ROI in your sector to be around 15-30%. But as an analyst, you know that the scale of operational activity allows ROI to be considered differently (and it’s already considered good at 5-7%) — and the AI’s assessment here turns out to be completely incorrect.

Skill in formulating clear queries

Essentially, this is a skill of task framing — it doesn't matter who, a junior colleague or an AI. Without this, AI will either give superficial answers, or, worse, confidently generate incorrect solutions and present them as common knowledge.

Actually, treating AI as a first-year intern is a good approach — it immediately clears your mind and removes unnecessary illusions :-)

For example, a bad prompt sounds vague: "Calculate user churn", even if we provide all the tables with source data. What's bad here is that we didn't provide any criteria — and the neural network starts basing its work either on its own some notions about churn (and we don't know what those are), or solves this task in the most primitive way.

A good prompt, on the contrary, specifies the task as much as possible: "Write an SQL query for PostgreSQL to the sessions table (user_id, timestamp). Consider a user as churned if they haven't logged in for more than 30 days since their last visit. Exclude test accounts with domain @test.com". But even here it would be good to ask the neural network to provide its comments during its reasoning and logic.

The key thing in this skill is to never blindly trust AI answers and pass them through your internal filter of knowledge. Without this, a data analyst turns into a chatbot operator and their value as a specialist drops sharply.

AI as a tool for "internal" analyst processes

Today, analysts most often use AI to optimize their own workflows. The goal of this approach is to reduce time spent on routine and uncomplicated operations that the neural network handles about as well as a human, but significantly faster. Not "conduct analysis", not "find out why metrics dropped", but do purely mechanical work.

At the same time, common sense remains key: analysts only assign to AI those tasks that they are capable of performing themselves, so they can evaluate the correctness of the result and notice errors in time.

Understanding, refactoring, and writing code

For an analyst, SQL queries, Python scripts, and dbt models are an essential part of their work. AI is great at handling drafts, syntax suggestions, simplifying complex constructs, and finding errors.

Moreover, since analysts often work with existing processes and models, neural networks can help them quickly understand the logic of a query and explain what exactly is being calculated and where the metrics are coming from. This is particularly useful for legacy situations, when a new person joins a project and needs to quickly grasp what's happening.

This works well for generating explanations for complex SQL queries, syntax suggestions (window functions, CTEs, filters), simplifying nested queries, and finding typos and logical inconsistencies.

However, for tasks that require deep knowledge of business logic and relationships between database tables, AI can make critical logical errors that affect the business. A good analyst should be aware of these limitations and not trust the computer for anything more complex than basic tasks.

Automation of Routine Tasks

Routine tasks are part of an analyst's job, and some of them can be delegated to a neural network.

For example, AI works well for writing documentation for code and models, describing fields and tables—this reduces cognitive load and saves time for the analyst.

It’s important to note that AI does not replace the analyst, but acts as an accelerator for solving standard tasks (it’s crucial that they are standard tasks). The responsibility for the correctness of logic, interpretation of results, and business context still rests with the human.

For example, AI is not always able to correctly understand business logic: it might say that the column is_active is a “flag for activity,” but it won’t know that “active” means someone who made a purchase in the last 30 days. Of course, you can provide the computer with all the context and correct its answers step by step, but this will take significantly more time than doing the task yourself.

What’s next

Today we covered the basic part of applying neural networks in the work of an analyst — what can help in almost every direction. In the next article, we will look at how AI can be used to work with business users and talk about conversational analytics (it's even more interesting there).