MPP Insights Blog

Will AI Replace Data Analysts?

Will AI Replace Data Analysts? cover showing a penguin diving from an ice cliff, by Dr. Sergei Shestakov
Everyone is worried that their job will be replaced by AI sooner or later, and data analysts have the same concern. Based on what I see every day in my work with data and AI, I believe AI will be good enough to handle more routine analytical tasks over the next three years. This will change the role of a data analyst, but it won’t replace them entirely. The question is how this change will impact data analysts.

The Biggest Change AI Will Bring to Analytics

Over the next 3 years we will live through the rise of autonomous AI agents. Currently, most AI tools work with human instructions, and humans control every step. AI agents work more independently and autonomously. Humans give AI agents a goal, then the agents can:
  • create a plan;
  • use tools;
  • examine the results;
  • find and correct mistakes;
  • continue working for a much longer period without human intervention.
On a 3-year horizon, autonomous agents will definitely be able to handle more complex analytical tasks and have a higher success rate than they do now. This will change the data analysts’ role: analysts will be involved in supervising the process and validating the results.
AI agents might not be as good as an experienced analyst, but they could be good enough for many tasks of average complexity, and could complete the work much faster.
All organizations care about the privacy and sensitivity of their business data. For this reason and other considerations like cost of computing, many prefer on-premise AI over a public cloud AI service. This is a huge trend - on-premise AI models prevail.
For example, NVIDIA Nemotron 3 Super may be less powerful than some frontier models, but it could be good enough and could be deployed on-premise. This helps organizations keep their sensitive data on their own servers and cut costs.

What Will Remain Human in Analysis

AI is great at doing boring and repetitive tasks. For example, data preparation, cleaning datasets, writing routine SQL, etc. This gives analysts the opportunity to focus on more valuable activities, such as setting goals and making decisions. In fact, AI can analyze data and calculate probabilities, but it can’t own a business decision. Humans set goals, define the vision, approve decisions and, more importantly, take responsibility for the consequences.
Accountability and ownership are especially important when external factors, such as macroeconomic shocks, make forecasting more complex and require human supervision. As we know, the real world and the processes happening in real life are non-deterministic and hard to model. This cannot be fully handled by computing machines, no matter how “smart” they seem.
I think a bridge role would be the most important for data analysts of the future. Human analysts will act as a bridge between AI agents and the business. This makes communication skills even more important, as analysts need to communicate with the team and turn their needs into prompts.

Challenges for Wider Adoption of AI

Organizations rush to leverage AI, and they are experimenting with it a lot, but there are several factors that may make AI adoption very challenging.

Hardware

Running AI on-premise is a major trend, and it requires hardware to run the model. Hardware costs and hardware shipment waiting times are significant blockers for many organizations that want to adopt AI.

Security

Most companies test AI freely in an isolated AI sandbox to make sure it is safe and yet provide freedom for innovations. However, the production environment has strict security policies, sometimes paranoid. So moving an AI application from the testing environment to production is the real challenge. Because the same application can work perfectly in the sandbox, but in production it literally won’t run as expected due to security restrictions.
Prompt injection is another example of a new type of security challenge that people will need to learn to handle. It happens when someone gives an AI agent a harmful instruction, such as ignoring rules or revealing private information. Organizations need to learn how to control the instructions the AI follows and prevent the agent from taking unsafe actions.

Data Quality and Governance

AI agents make analytics easier and faster. However, a wrong definition or a poor-quality dataset can make all the reports incorrect. For this reason, data quality and data governance will become increasingly bundled with AI and BI platforms. Our early internal testing shows token savings of 17%-20% when Agentic BI tasks run in a well-curated data governance environment versus bare-bones data marts.

Nobody Wants to Be First

Like with all innovations, a company can believe that AI adoption could be useful, yet it may not want to be the first company to make a major investment or accept a major risk.
A good example of this phenomenon could be hungry penguins crowding at the edge of a cliff above the roaring Antarctic seas. They need to jump to get fish to feed their young, but it's too dangerous. There are predators in the water waiting, and the waves are very high. But when the first penguin dares to jump, all the others follow.
Once several organizations show that AI can be deployed safely and can create measurable business value, other companies may become more confident about enterprise AI adoption. FOMO (“Fear Of Missing Out”) also helps.
Penguins crowding at the edge of a snowy cliff above rough Antarctic seas

What Skills Should Data Analysts Learn?

Data analysts should focus on learning fundamentals instead of fully relying on AI-wrapped technology.
Math skills are priceless, because they help analysts understand patterns and models.
Although AI generates SQL these days, the understanding of SQL will remain valuable. Understanding SQL helps analysts understand and unlock the full potential of relational data.
The same applies to Python. SQL and Python will be essential for analysts who want to understand what AI is doing. These skills are especially useful for code debugging and results validation. In the future it will be more about principles, not semantics.
Assembler languages are a great example of how the situation changes with time. Nowadays, developers don’t need to know assembler, but they generally understand the principles of machine level execution of low level computing instructions. In the future, analysts may not need deep knowledge of currently popular programming languages. But they will still need to understand the basics, including statistics, data modeling, business processes, and causal reasoning.

ROI on AI Investments

There is a very different ROI from investing in AI, case-by-case.
Let’s select 2 types of companies to analyze the ROI: software development companies and non-technology companies.

Software Development Companies

These companies use AI to speed up development, but they are also aware that blindly delegating work to AI can create a huge maintainability burden. Without human control, the system becomes difficult to understand, debug, update and repair later.
If too much work is delegated to AI, the most interesting and creative part of an engineer’s work may disappear. All-in-all, software development companies get significant gains from AI investment even now, and this will grow with years and more mature technology. And the results are manageable and measurable to a large extent.

Non-technology Companies

Many non-technology companies want to jump on the wagon because they are afraid of being left behind (the FOMO phenomenon). They may think that if they don’t invest in AI analytics, they will lose. This only creates pressure to invest quickly, even before the company has clearly defined the business problem.
These companies may have too high expectations from AI tools. They may believe that deploying AI tools will automatically solve some business problems, completely eliminate manual labor, etc. The concept is very much like “Let’s put a magic AI brick on a keyboard - it will do the job automatically”:
Brick with three domes resting on a laptop keyboard
Organizations must make AI solutions safe enough for production. This is a polarized “greed vs fear” issue. Some companies are lowering the security barrier to help IT teams experiment and deploy AI faster. Other companies are raising the security barrier to the topmost, “paranoid” level. This protects the organization better, but it can also block useful AI-based data analytics from being deployed in production.

The Future of Data Analysts

Clearly, software development and analytics have already changed because of AI. In the future, data analysts will remain responsible for setting goals, and they will be accountable for the final decisions.
Successful data analysts and BI professionals in the next few years are the ones who learn to work alongside AI, leverage it, know when to trust it and know when to step in, ultimately becoming a bridge between business and AI.

About the author: Dr. Sergei Shestakov is the CEO and Founder of MPP Insights, a data engineering, analytics, and agentic AI consultancy.

Related reading:

Opinion