What Does a Data Scientist Do? How AI is Changing the Role

Data scientists have become central to how organisations understand performance, predict outcomes and make decisions. Their work combines data, statistics, programming and business knowledge to find patterns and develop models that can be applied to practical problems. 

Artificial intelligence is changing how much of that work is completed. AI tools can now generate code, write SQL queries, support analysis and accelerate model development. This is increasing the importance of the capabilities that require human judgement: understanding what the data represents, defining the right problem, evaluating outputs and explaining what the findings mean for the organisation. 

For current and aspiring data professionals, this is reshaping the skills needed to build a successful career. For employers, it is changing how data science roles should be defined and assessed. 

What Does a Data Scientist Do? 

A data scientist uses analytical techniques, statistical models and programming to extract useful information from complex datasets. Their work can support forecasting, customer analysis, operational improvements, risk management, automation and strategic planning. 

Australia’s official occupation classification describes data scientists as professionals who prepare and clean data, apply statistical and programming techniques, build machine learning and AI frameworks, evaluate models and present findings to decision-makers. It also recognises the strategic contribution data scientists make to organisational data initiatives. 

A typical data science project may involve: 

• Defining the business problem and determining whether data can answer it 
• Finding, cleaning and combining relevant data 
• Exploring patterns, anomalies and relationships 
• Developing and testing statistical or machine learning models 
• Evaluating the accuracy, reliability and limitations of those models 
• Translating the results into practical recommendations 
• Monitoring performance after a model or solution has been introduced 

The technical work is only one part of the role. A strong data scientist also needs to understand the operating environment, ask useful questions and communicate findings in a way that supports decisions. 

How AI is Changing Data Science Work 

AI adoption is increasing rapidly. In 2025, 88% of organisations globally reported using AI, according to the Stanford AI Index 2026. 

Australian adoption is lower across the entire business population, although it rises considerably with organisation size. The Australian Bureau of Statistics found that 12% of Australian businesses used AI during 2024–25. This reached 35% among large businesses and 22% among medium-sized businesses. 

Adoption was also concentrated in many of the industries that employ data professionals. AI was used by 38% of businesses in information, media and telecommunications, and by 24% of businesses in both professional services and financial services, according to the Australian Bureau of Statistics. 

As adoption grows, data scientists are increasingly using AI across several stages of their work. 

Coding and querying 

Generative AI can draft Python or R code, write SQL queries, explain unfamiliar code and suggest ways to resolve errors. This can reduce the time required to complete routine programming tasks and help data scientists explore different approaches more quickly. 

The value of the output depends on the context supplied to the model. A technically correct query can still use the wrong dataset, misinterpret a field or apply a definition that does not reflect how the organisation measures performance. 

Data exploration and model development 

AI tools can help identify potential relationships, recommend features, suggest modelling techniques and create initial documentation. They can also support experimentation by allowing data scientists to test ideas and compare approaches more efficiently. 

These capabilities increase the need for rigorous evaluation. Data scientists still need to determine whether a pattern is meaningful, whether the data is representative and whether a model will perform reliably outside its original development environment. 

Communicating findings 

AI can help turn technical analysis into summaries, reports and explanations for different audiences. This gives data scientists more ways to communicate their work across an organisation. 

Effective communication still depends on understanding the decisions that stakeholders need to make. Data scientists remain responsible for explaining assumptions, uncertainty and limitations clearly. 

What Spotify Shows Us About the Future of Data Science 

Spotify provides a practical example of how AI can change the work of data scientists. 

The company developed an internal data assistant that allows employees to ask questions in plain English. The assistant identifies the relevant context, writes a SQL query, runs it against Spotify’s data warehouse and returns the result alongside the query and its sources. 

Since August 2025, the assistant has been used by more than 2,100 Spotify employees across more than 13,000 conversations and 60,000 messages. More than one-quarter of its users had never written SQL. 

The system relies heavily on data scientists and analytics engineers. These specialists curate the datasets, business definitions and approved question-and-SQL examples that guide the assistant. 

Spotify tested whether its historical query data could provide these examples automatically. Domain experts approved only 12.5% of the suggested pairs. The remainder included exploratory queries, debugging work, outdated approaches, incorrect tables and queries that were technically valid but demonstrated an unsuitable pattern. 

This shows where the role of the data scientist is gaining strategic importance. Data professionals can spend less time answering repeatable questions and more time shaping the data context that allows others to find reliable answers. Their knowledge of definitions, systems and business requirements becomes part of the infrastructure behind the AI tool. 

Human Judgement is Becoming More Important 

AI systems are becoming more capable, although their performance remains inconsistent across different types of work. 

Stanford describes this as a “jagged” capability frontier. Models can perform exceptionally well on some difficult benchmarks and struggle with tasks that appear straightforward. In 2025, AI agents reached approximately 66% task success on a benchmark involving real computer tasks, leaving around one-third of attempts unsuccessful. 

There is also a gap between perceived and measured productivity in some technical environments. A 2025 randomised controlled trial studied 16 experienced open-source developers completing 246 genuine issues in repositories they knew well. When using AI tools, they took 19% longer to complete the work. Before the study, they expected AI to make them 24% faster. After completing the work, they still believed AI had improved their speed. 

The study relates specifically to experienced software developers working in established repositories, so its findings should not be applied directly to every data science task. It demonstrates why organisations need to measure the impact of AI tools and account for the time required to review, correct and integrate their output. 

For data scientists, this places greater value on: 

• Testing outputs against known results 
• Evaluating model performance across different scenarios 
• Identifying bias, leakage and unreliable correlations 
• Understanding the source and quality of the underlying data 
• Recognising when human review is required 
• Monitoring performance after deployment 

Will AI Replace Data Scientists? 

Current evidence points to significant changes in the tasks data scientists perform. 

The Anthropic Economic Index found that computer and mathematical work accounted for 34% of Claude.ai conversations and 46% of enterprise API usage in November 2025. AI is already deeply embedded in technical work. 

The same research found that 52% of Claude.ai interactions involved augmentation, where users learned, refined work or collaborated with the system. Automated interactions accounted for 45%. Anthropic also found that success rates declined as tasks became more complex. 

For data science, this suggests that routine coding, querying, documentation and exploratory work will increasingly be supported by AI. Greater responsibility will sit around problem definition, domain context, evaluation, data quality, governance and commercial application. 

This also helps explain why demand for data capability remains strong. The World Economic Forum Future of Jobs Report 2025 identified big data specialists as the fastest-growing role through to 2030, with AI and machine learning specialists also among the top three. The report found that 86% of employers expected AI and information-processing technologies to transform their business by 2030, while 39% of workers’ core skills were expected to change. 

Will AI Replace Data Scientists? 

The data scientists who create the most value will combine strong technical foundations with the ability to apply AI responsibly and commercially. 

Technical foundations 

Python, R, SQL, statistics, experimentation, machine learning and data preparation remain fundamental. Data scientists need enough technical depth to assess AI-generated work, identify errors and select an appropriate approach. 

AI-enabled delivery 

Experience with generative AI, large language models, model evaluation, prompt and context design, MLOps and monitoring is becoming increasingly relevant. The required depth will depend on whether the role applies existing AI tools or develops and deploys AI systems. 

Data quality and governance 

AI systems depend on reliable, well-understood data. Knowledge of data lineage, privacy, access controls, model documentation, explainability and responsible AI practices is becoming more valuable. 

Business understanding 

Data scientists need to understand the problem before choosing the method. This includes clarifying the decision that needs to be supported, identifying the available evidence and recognising when a simpler analytical approach may be more appropriate. 

Communication and influence 

Data science creates value when findings can be understood and applied. The ability to communicate uncertainty, explain limitations and influence stakeholders remains a defining capability. 

Data Scientist, Data Analyst or AI Engineer? 

The boundaries between data and AI roles can vary between organisations. The main distinction usually comes from the type of problem being solved and how the output will be used. 

Some organisations combine several of these responsibilities into one position. Clear role definition helps candidates understand expectations and helps employers identify the capability their project requires.

What This Means for Data Science Careers 

AI familiarity is becoming part of data science capability. Professionals will benefit from learning how to use AI tools effectively, evaluate their outputs and understand the governance requirements surrounding them. 

Domain knowledge will also become more important. Data scientists who understand the organisation, industry and decisions behind the data are better placed to recognise when an answer is useful, misleading or incomplete. 

The future of data science will be shaped by professionals who can combine technical expertise, AI capability and human judgement. As analytical tools become more accessible, the ability to create trusted context and turn outputs into sound decisions will remain highly valuable.