AI & the role of Data Analyst

The work before AI

Before AI entered the daily workflow, a large share of a data analyst's time went into work that created no direct value on its own. It went into things like:

• Writing and rewriting SQL to get a query into a usable shape
• Tracking down why two reports disagreed by a few percentage points
• Reformatting the same numbers into slightly different chart styles for different stakeholders
• Writing up what a dataset actually contains before anyone could ask a real question of it

The friction wasn't the analysis itself. It was everything that had to happen before the analysis could start: locating the right table, understanding what a column actually meant versus what it was called, checking whether last quarter's definition of "active user" still matched this quarter's.

None of that built a case for a decision. It just cleared the path to being able to make one.

Where AI already changes the day to day

A handful of tasks have genuinely shifted, not in theory but in the everyday routine of the role.

1. Query drafting and debugging. A first-pass SQL query, or an explanation of why an existing one is returning the wrong row count, now takes minutes instead of the better part of an hour spent scanning documentation or asking a colleague who's also busy.

2. Data cleaning and first exploration. Spotting outliers, flagging likely duplicates, and suggesting a first pass at how to handle missing values are now largely automatable, which matters because this step used to eat a disproportionate share of every analysis.

3. Draft documentation. Generating a data dictionary from a schema, or a plain-language summary of what a metric actually measures, is one of the more underrated shifts. It makes institutional knowledge more accessible to the next person, not just faster to produce.

4. First-draft narratives. Turning a table of numbers into a written summary a stakeholder can actually read, before a human tightens the framing and checks the caveats, has moved from a slow manual step to a starting point.

5. Pattern and anomaly flagging at scale. Surfacing where in a large dataset something looks unusual is now something AI does continuously in the background, rather than something an analyst discovers by chance while looking for something else.

Taken together, this is a genuine shift in where time goes, not a cosmetic one. A review of task-level productivity studies summarised in the International AI Safety Report 2026 found productivity gains in the 20 to 60% range under controlled conditions, and 15 to 30% in most real-world settings, with knowledge work and data-related tasks consistently among the categories showing the largest gains.

What continues to be deeply human

None of the above touches the parts of the job that actually carry weight.

Choosing what matters. Deciding which metric matters for a specific decision, rather than which one is easiest to compute, is a judgement call rooted in what the business actually needs, not in what the data happens to make convenient.

Telling signal from noise. Distinguishing a genuine anomaly from a data quality problem still requires knowing the system that produced the data, its quirks and its history, not just the number itself.

Navigating incentives. Negotiating what "success" means with a stakeholder who has their own incentives is a political conversation before it's a technical one.

Owning the number. Taking responsibility for a figure that will be quoted in a board deck, defending it under scrutiny, owning it if it turns out to be wrong, isn't a task that can be delegated to a tool. Accountability doesn't transfer the way output does.

How seniority changes with AI

The clearest pattern in recent research is that AI narrows the gap at the entry level and widens the gap in responsibility further up.

Using payroll data covering millions of workers, Stanford's Digital Economy Lab found that employment for 22 to 25 year olds in AI-exposed occupations fell by 16% relative to trend after the widespread adoption of generative AI, while employment further up the seniority ladder stayed broadly stable.

Field experiments point the same way. In a large-scale study across Microsoft, Accenture and an anonymous Fortune 100 company, developers using an AI coding assistant increased completed tasks by 26.08%, with less experienced developers showing higher adoption rates and greater productivity gains.

That doesn't mean juniors are being replaced. It means the work AI absorbs fastest is disproportionately the work that used to be how juniors learned the job and proved themselves.

As Brookings has argued, senior expertise is not just a stock of accumulated knowledge. It's the residue of years spent doing exactly the kind of routine work AI now handles. A firm that lets AI absorb that work without a deliberate plan for how junior staff still build judgement is narrowing, without quite noticing, the pipeline that produces its own future seniors.

For senior analysts specifically, AI doesn't reduce responsibility, it relocates it.

A KPMG and University of Texas field study of 523 early-career professionals found that the professionals who added real value alongside AI weren't necessarily the ones with the strongest technical skills going in. They were the ones who treated AI output as a draft to be checked against context, not a finished answer. That's precisely the muscle senior analysts are now asked to use more, not less: verifying work they didn't fully produce themselves, catching what a less experienced colleague didn't know to look for, and staying accountable for a faster pipeline of output.

Risks of using AI without context

Automation bias. The clearest risk isn't that AI gets things wrong occasionally. It's that people stop checking. The International AI Safety Report 2026 describes this directly as automation bias: the tendency to over-rely on automated outputs while discounting information that contradicts them, which discourages active reasoning and weakens people's willingness to act on their own judgement. It has been documented across aviation, medical diagnostics and everyday chatbot use, and there's no reason data analytics is immune.

Noisier output behind the speed. Productivity studies typically measure completed volume, not correctness. The Microsoft, Accenture and Fortune 100 study cited above measured completed tasks, not whether every one of them was right. That distinction matters in practice: a dashboard that ships faster isn't automatically a dashboard that's right, and most productivity metrics don't distinguish between the two.

A thinner bench, later. This risk compounds the seniority issue above. If AI increasingly does the work junior analysts once used to build judgement, and organisations don't deliberately replace that learning path with something else, the loss shows up years later, not immediately, as a thinner bench of people who can catch a wrong number before it reaches a decision, a version of the same risk covered in how to scale tech teams without increasing churn or delivery risk.

How KWAN thinks about this

KWAN builds AI-augmented engineering and data teams, not AI-replaced ones. The model stays people-first: every consultant has a dedicated People Experience Partner, and seniority is treated as something to be deliberately built, not assumed to appear once AI has absorbed the entry-level work.

The same questions in this article, where does AI genuinely help, and where does judgement still have to sit with a person, are the ones KWAN asks when building a team around a client's context, not a generic role description.

Frequently asked questions

1. Will AI replace data analysts?

No. AI has absorbed the mechanical middle of the role, extraction, cleaning, first-draft reporting, but judgement, accountability and stakeholder context remain tasks that don't transfer to a tool.

2. Which data analyst tasks does AI handle best today?

Query drafting and debugging, data cleaning and first exploration, draft documentation, first-draft narrative summaries, and pattern or anomaly flagging across large datasets.

3. Does AI make junior data analysts less necessary?

Not in the way it looks at first glance. AI absorbs the tasks juniors used to learn from, which raises a real question about how judgement gets built over time, not whether junior roles disappear.

4. Does AI reduce the responsibility carried by senior analysts?

No, it relocates it. Senior analysts increasingly spend time verifying AI-assisted work and catching what a less experienced colleague didn't know to look for, rather than producing every output themselves.

5. What's the biggest risk of using AI without context in data analysis?

Automation bias: the tendency to trust automated output and stop checking it. That risk compounds if junior learning pathways aren't deliberately maintained, thinning the pipeline of future senior judgement.

6. Is AI adoption in analytics mostly about replacing tasks or changing them?

Changing them. The tasks that consumed the most time, formatting, querying, first drafts, shift to AI. The tasks that carry the most weight, framing, trade-offs, accountability, stay exactly where they were.

More in this series soon, on other IT roles and where AI actually changes their day to day. If any of this resonates, we're happy to talk it through.