Microsoft introduced a new feature that lets users ask questions about data from their Power BI reports and semantic models and conversationally get answers, something Microsoft itself would have found quite ambitious a few years back. For students exploring data analytics courses, this shift shows how quickly the tools analysts use are changing.
Google is doing the same thing. The Conversational Analytics features allow one to ask questions in natural language about a business, while new agentic workflows in Looker allow a user to observe a metric and investigate why it has changed.
The issue becomes quite challenging for someone planning a career in analytics.
What should an analyst learn when software can turn a sentence into a query, chart, or explanation?
An obvious answer would be verification.
A convincing answer can still be wrong
For example, suppose you have an analytics assistant who asks you:
“In which customer group did we retain the largest number of customers in the last quarter?”
It instantly produces a chart with an explanation within a few seconds.
The chart looks extremely professional. That doesn’t mean that the chart is right.
Or did “retention” refer to 30 days rather than the usual 90-day period? There might also have been trial users, or not enough customers in one particular group.
Not all these issues lead to creating an obviously wrong chart.
And that’s how they produce something much worse: a seemingly correct chart under a dubious definition.
Therefore, as data analytics courses become increasingly popular, it becomes essential to teach young students to question the answer rather than generate it.
SQL becomes a checking tool as well as a creation tool
SQL will not vanish with conversational analytics.
This gives SQL another purpose.
The analyst will have the ability to see the rationale of the AI result. What tables are being joined? Are there any duplicate rows? What filters have been applied? Was an inner join needed instead of a left join?
That holds in Excel, Python, and BI systems.
An analysis can be conducted either manually or, better yet, through automation. Understanding how the automation works helps analysts determine whether the result makes sense.
This distinction is now taking on greater importance from a business perspective. According to Reuters news in July, hiring in the IT industry for AI positions increased 16% YoY in June, while IT job positions declined 3%. In 14 industries, the number of AI and machine learning positions increased 25%.
Employers are certainly looking for AI talent. However, to be a competent analyst, having AI knowledge means understanding when it is useless to analyze automation. This is an important distinction that data analytics courses also need to address as AI becomes part of everyday analytical work.
Business definitions are part of analytics
How might I use this to teach a simpler problem?
Which item performed the best?
Since it’s not clear which one would be considered the best performer, there can be no analytical solution.
Was it the highest in sales?
The highest in revenue?
Was it the biggest margin contributor?
Or the fastest-growing item?
Or perhaps the lowest returning item?
In reality, even a good dashboard can provide the wrong analytical answer if the metric is not clearly defined.
The description of Google Conversational Analytics makes the point that it is about conversing using controlled enterprise semantics. Well said: advanced AI needs good definitions through the conversation.
For this purpose, the students need to become comfortable with formulating specific analytical questions from general stakeholder inquiries.
This is harder to gauge with a multiple-choice test.
Statistics becomes a defence against attractive stories
The AI system can offer explanations quickly. People have proved themselves capable of doing something similar.
After a marketing campaign, sales increased, meaning the marketing campaign raised sales.
Feature X customers do not churn, suggesting that the feature may help reduce churn.
One group has a higher conversion rate, meaning that their sales team should be better.
All the sentences could be right.
For now, there has been no proof from the given data.
A data analyst should have enough knowledge of statistics to detect sampling issues, confounders, misleading averages, and the difference between correlation and causation.
The courses should not teach everyone to become a statistician. Instead, the courses should make the learners uneasy about the conclusions which the data cannot support. This is particularly important for learners taking data analytics courses, where understanding the limits of data matters as much as producing an answer.
Perfect datasets teach the wrong lesson
Many projects are designed to be collaborative from the very start.
Dates are formatted correctly. Customer ID numbers are in agreement. Categorization is consistent. Missing values are readily manageable. A business question already exists.
However, this does not necessarily require students to think critically about the analysis.
A useful course project might instead contain:
- two conflicting definitions of an active customer;
- duplicate transactions that inflate revenue;
- missing values requiring a documented decision;
- a stakeholder request that is too vague to analyse immediately;
- an AI-generated interpretation that students must verify.
The final dashboard then becomes only part of the assessment.
The reasoning behind it matters more.
AI belongs inside the classroom
One would find it difficult to avoid the use of such tools.
Both Microsoft and Google are embedding conversational and agentic capabilities into their popular analytics suites. This is something that students will encounter in the near future when they go to work.
An excellent course would enable students to use AI to create SQL, examine trends and exceptions, extract information from dashboards, and generate explanations.
If that is true, they should be made to critically evaluate what they have created.
Where did the answer come from?
What were the underlying assumptions?
What could make you reconsider the conclusion?
Could the results be replicated without accepting the explanations provided?
Ask to see the mistakes students have to catch
One of the questions that arises while comparing data analytics courses is which tools they include.
These usual suspects, Excel, SQL, Python, Power BI, or Tableau, are still worth considering.
However, now one more question should be added:
Which errors in analysis can I expect to come across?
The course leaves out a very important task if all the exercises include one perfect dataset, one obvious metric, and one correct graph.
Analytical tools are becoming increasingly capable of providing answers.
The analyst whose job is still needed will be the one who can look at the nice and neat answer and say, “Before doing anything about it, I would like to check.” For students choosing data analytics courses, that ability to question and verify an answer may ultimately become one of the most valuable skills they develop.
Build smarter analytical thinking with AnalytixLabs, where you learn to question AI answers, verify insights, and make confident data-driven decisions.
