Data Analyst Workplace Guide · 2026
What Does a Data Analyst Actually Do at Work?
request → clarify → analyse → verify → explain
Go beyond the job description. Follow real workplace situations—from a manager’s urgent question and a Finance mismatch to a broken weekly report—and see how a Data Analyst turns raw records into an answer people can trust.
The software changes. The responsibility does not.
Quick answer
What does a Data Analyst actually do at work?
A Data Analyst turns questions and raw records into checked evidence that people can use. The work can include clarifying a manager’s request, retrieving data, checking quality, calculating and comparing results, reconciling numbers, updating reports, investigating broken dashboards, documenting definitions and explaining findings. Microsoft’s current Power BI Data Analyst framework likewise covers stakeholder requirements, data preparation, modelling, analysis, visualisation, data quality, refresh and governance—not dashboard design alone. [1]
- Clarify the real question
- Check the right records
- Analyse and reconcile
- Explain what the evidence supports
- Maintain repeatable reporting
What actually lands on a Data Analyst’s desk?
Real requests are rarely written like textbook exercises. A manager may ask, “Why did sales fall yesterday?” Finance may say, “Your dashboard does not match our total.” A scheduled report may stop refreshing five minutes before a meeting. The analyst has to determine what kind of problem has arrived before choosing a tool.
Recurring reporting
Daily, weekly or monthly reports that must use consistent definitions and dependable data.
Typical concern: Did the refresh complete, and do the totals still reconcile?
Ad-hoc questions
A new question that is not already answered by an established report: “Which branch caused the decline?”
Typical concern: Define the question quickly without jumping to a cause.
Data-quality problems
Duplicates, missing records, changed categories, impossible values or two systems reporting different numbers.
Typical concern: Is the apparent business problem actually a data problem?
Report maintenance
A file name, column, credential or upstream process changes and a once-reliable report no longer behaves as expected.
Typical concern: Find where the reporting chain failed and retest it.
Before analysis: clarify the request, define the number and understand the records
Good analysts do not begin with software. They begin with meaning. Microsoft’s current analyst framework explicitly places business requirements and data acquisition before preparation, modelling and analysis. [1]
1. Clarify vague requests
Suppose someone says, “Customers are reducing.” That is not yet a safe analytical question. Which customers? Compared with which period? Is a customer someone who registered, someone who ordered, or someone who completed a purchase?
Turn the complaint into a measurable question
“Did the number of customers completing at least one purchase decline in October compared with September, and which branches contributed most to the change?” Now the analyst has a measure, a period and a scope.
2. Define what the number means
Words such as sales, active customer, profit, completed order and late delivery can have several valid definitions. Sales may mean gross sales, net sales after refunds, invoiced sales or money actually received. Two departments can therefore calculate different answers from the same organisation without either formula being mathematically wrong.
3. Know what one row represents
This is sometimes called the grain or level of detail of the data. Consider this fictional extract:
| Order | Product | Quantity |
|---|---|---|
| 1001 | Cement | 4 |
| 1001 | Paint | 2 |
| 1002 | Cement | 5 |
On a small screen, swipe the table sideways to view every column. There are three rows but only two orders. One row represents an order item. Counting rows as orders would produce the wrong answer even if the formula itself worked.
Practical workplace situation 1
Finance says your dashboard is wrong. What does the analyst do?
Imagine a fictional sales dashboard shows October revenue of KSh 8,420,000, but Finance reports KSh 8,010,000. The difference is KSh 410,000.
A professional analyst does not defend the dashboard simply because it is automated or attractive. The task is to reconcile the disagreement.
Confirm the definition
Does each side mean invoiced revenue, paid revenue, gross sales or revenue after refunds and cancellations?
First question: Are we measuring the same thing?
Confirm the period
Check transaction date, posting date, time boundaries and whether late-posted items fall into different months.
A date rule can explain a difference before any formula changes.
Inspect filters
Look for an omitted branch, excluded category, test transaction, cancelled order or unexpected status value.
One hidden filter can change an otherwise correct report.
Reconcile record counts
Compare source counts with the analytical dataset. If 12,480 transactions become 24,960 rows after a join, investigate the join before trusting any total.
A query can be syntactically correct while multiplying records.
Break the difference down
Compare branch by branch, date by date, category by category or status by status until the disagreement becomes small enough to trace.
Do not stare only at two grand totals.
Correct and retest
Suppose the fictional difference is KSh 410,000 of cancelled transactions left inside the dashboard data. Correct the rule, rerun the checks and document the change.
The job ends with a verified correction, not merely a diagnosis.
A useful final message
“The dashboard included KSh 410,000 of cancelled October transactions. The reporting rule has been corrected to exclude cancelled transactions, bringing the October total to KSh 8,010,000.”
When numbers disagree, reconcile the whole path
A useful checking model is Source → Analytical result → Published result. Agreement between only two stages does not prove correctness.
If the SQL result and dashboard both show 6,250 while the source contains 3,125, the dashboard and query agree with each other but may both be wrong. A duplicated join could have doubled the records before the dashboard ever saw them.
Analyst habit
Do not ask only, “Do these two numbers match?” Ask, “What should this number match, and why?”
Practical workplace situation 2
An urgent management question arrives before a meeting
It is 11:15 a.m. A manager has a 2:00 p.m. meeting and asks: “Sales dropped badly yesterday. Can you tell me what happened?” The analyst cannot investigate every possible explanation in a few hours. The skill is to scope the question intelligently.
Choose a sensible baseline
Compare with the same weekday, a recent average or the planned target—not automatically with a day that behaves differently.
Find where the decline sits
Break the total down by branch, product, category, channel or time until the movement is concentrated.
Test obvious explanations
Check transaction count, average order value, missing records, cancellations and other operational signals available in the data.
State what is known
Give management the strongest supported finding and name the next evidence needed instead of inventing a cause.
Suppose the fictional total decline is KSh 320,000 and Branch C alone accounts for KSh 270,000. Transaction count is broadly normal, but three normally high-value products recorded almost no sales.
“Yesterday’s decline was concentrated in Branch C. Transaction volume was broadly normal, but sales of three normally high-value products fell sharply. The sales data identifies where the decline occurred; it does not show whether stock availability caused it. Inventory records should be checked next.”
Not every question needs a new dashboard
A manager may ask, “Can you build a dashboard showing this?” Before building it, the analyst should ask two simpler questions: What decision will it support? and How often will this question be asked?
Use the smallest useful output
A checked table, short explanation or one chart may be better than creating a report that nobody needs again.
Make the answer repeatable
If the same request arrives every Monday, consider a saved query, reusable report or approved automation with validation checks.
Practical workplace situation 3
The weekly report suddenly breaks
A report has updated every Monday for six months. Today it fails. The Data Analyst investigates the whole reporting chain instead of assuming the dashboard software itself is the problem. Microsoft’s current analyst objectives explicitly include changing data-source settings, resolving import errors, resolving null and data-quality issues, and configuring scheduled refresh. [1]
Did the source arrive?
The expected file or database load may simply be missing.
Start upstream before rebuilding downstream.
Did the name or path change?
An automated process may expect one exact filename, folder or connection.
A small naming change can break a reliable routine.
Did the columns change?
Branch becoming Branch_Name can break a transformation that still expects the old field.
Schema changes are common causes of broken refreshes.
Did the data type change?
A date that arrives as text may fail where a transformation expects a true date value.
Inspect types, not only visible values.
Did access change?
Credentials, permissions or a gateway connection can fail even when the report itself has not changed.
A valid report still needs valid access to its source.
Did it refresh but become wrong?
This is the more dangerous case. A report can refresh successfully while an upstream source silently omits a branch or changes a category.
Successful automation is not proof of correct data.
The dangerous report
A report that crashes announces its failure. A report that refreshes successfully and quietly publishes the wrong number can be more dangerous. Validation still matters after automation.
A Data Analyst’s day is not one fixed timetable
There is no universal “9:00 SQL, 10:00 dashboard, 11:00 meeting” schedule. A better way to understand the job is to recognise the kinds of days that analytical work creates.
Reporting day
Refresh, verify, update and publish recurring figures. The key skill is consistency.
Investigation day
A number moved and nobody knows why. The key skill is analytical reasoning.
Data-quality day
Duplicates, missing records or broken joins make the numbers unreliable. The key skill is verification.
Communication day
Explain findings, agree definitions and answer questions. The key skill is clarity.
Does a Data Analyst spend most of the day cleaning data?
Sometimes data preparation takes a large share of the work; sometimes it does not. Avoid universal claims such as “every analyst spends 80% of the day cleaning data.” A mature organisation with governed databases can behave very differently from one where several branches exchange manually edited spreadsheets. The reliable lesson is that data quality is a real part of the job, but its share varies by organisation, system and role.
What if a value looks suspicious?
Suppose normal daily sales are between KSh 400,000 and KSh 600,000, then one day shows KSh 5,800,000. Do not delete it merely because it looks unusual. It could be a duplicate, a posting problem, several days loaded together—or a genuine large transaction. Investigate the records that created the spike before deciding what to do.
Missing does not mean zero
A blank score, payment date or delivery time can mean absent, not entered, unknown, not applicable or import failure. The analyst must understand the business meaning before replacing a missing value. Microsoft’s analyst objectives explicitly include resolving unexpected and null values as data-quality work. [1]
What tools does a Data Analyst use—and do they spend all day coding?
No single tool defines the job. The best tool depends on where the records live, how often the task repeats and what the audience needs.
| Workplace need | Common tool | What the analyst is trying to do |
|---|---|---|
| Quickly inspect a manageable file | Excel or another spreadsheet | Filter, calculate, pivot, spot-check and compare. |
| Retrieve relational records | SQL | Select, filter, join, group and aggregate structured data. |
| Repeatable cleaning or automation | Python or another scripting tool | Make transformations repeatable and reduce manual work. |
| Recurring interactive reporting | Power BI or another BI platform | Publish measures, trends and monitored views for authorised users. |
| Explain a finding | Report, presentation, email or meeting | Turn the result into language another person can act on. |
On a small screen, swipe the table sideways to view every column. A real workflow may use one tool or several. The objective is a reliable answer, not maximum software use.
What does SQL actually do during the day?
When important records are stored in a relational database, SQL can retrieve selected rows, join related tables and calculate grouped summaries. PostgreSQL’s official tutorial covers querying tables, joins and aggregate functions as core SQL operations. [2]
But SQL cannot decide what “completed sale” means for your organisation. A query can run successfully while using the wrong date, wrong status, wrong join or wrong business definition.
Want the SQL part step by step?
Do Data Analysts attend meetings?
Often, yes. Meetings can prevent analytical errors: clarifying the request, agreeing a metric definition, understanding a surprising result, presenting findings or deciding what evidence to collect next. A ten-minute discussion about what “active customer” means can save hours of analysing the wrong definition.
What does the Data Analyst actually give the manager?
The final product is not always a dashboard. Sometimes the most useful output is three verified numbers and a clear explanation. A strong handoff usually contains five pieces.
The answer
What happened?
The evidence
Which calculation, comparison, chart or records support it?
The definition
What exactly was measured?
The limitation
What does the available evidence not establish?
The next action
What should the reader do or investigate next?
Documentation is part of the work
If a dashboard says Active customers = 12,410, another analyst should not have to guess what “active” means six months later.
| Documentation item | Example |
|---|---|
| Metric | Active customer |
| Definition | Customer with at least one completed purchase in the previous 90 days |
| Source | Orders database |
| Exclusions | Cancelled and test orders |
| Refresh | Daily |
| Last definition change | 12 October 2026 |
On a small screen, swipe the table sideways to view every column. Documentation makes a recurring number easier to check, explain and maintain.
When should a Data Analyst automate a task?
If somebody asks every Monday, “Please send last week’s sales by branch,” repeating the same manual steps forever may not be sensible. The analyst might save and parameterise a query, create a reusable report or automate an approved refresh.
Before automating
Make the logic stable first
An automated wrong definition simply produces the wrong answer faster.
- Confirm the metric definition and source.
- Make the transformation repeatable.
- Add checks for expected dates, row counts or duplicates.
- Monitor refreshes and source changes.
- Keep a way to investigate unexpected totals.
Where does AI fit into Data Analyst work in 2026?
AI can help with parts of the workflow. Microsoft Power BI Copilot, for example, can generate summaries and answer questions about report content, while Microsoft’s current analyst objectives also include Copilot-assisted report creation and summarisation. [1] [4]
That does not transfer responsibility for the final number to the tool. A generated query, summary or visual still needs the same questions:
Check the logic
Correct tables? Correct join? Correct filters? Correct date? Correct treatment of missing values?
Check the evidence
Does the wording match the report, its filters and the actual figures? Microsoft documents limitations and report-design considerations for Copilot summaries. [4]
Simple 2026 rule
Use AI to accelerate approved analytical work—not to outsource responsibility for whether the answer is correct.
Data Analysts also have responsibilities around personal information
Customer, student, employee, patient and financial datasets can contain personal information. Kenya’s Data Protection Act requires principles including lawful, fair and transparent processing; explicit and legitimate purpose; data minimisation; accuracy; and appropriate limits on retention and transfer. [3]
Practical questions before sharing or exporting data
Do you actually need names for this summary? Should phone numbers appear in the analytical file? Who is authorised to see the raw records? Can the report use aggregated information instead? Follow the organisation’s approved privacy and access procedures.
What does a junior Data Analyst actually do?
A junior analyst is often given more clearly defined pieces of the wider workflow. That can include checking a refresh, cleaning an export, running or modifying an established query, reconciling two totals, updating a recurring report, investigating a small discrepancy, preparing a chart or documenting a metric.
Check
Confirm dates, row counts, missing fields and obvious inconsistencies.
Learn what reliable organisational data looks like.
Update
Refresh or prepare established reports using agreed definitions.
Build consistency and repeatability.
Investigate
Trace a mismatch or unusual value without jumping to a conclusion.
Build analytical judgement.
Explain
Turn the result into a clear note, table, visual or short presentation.
Build stakeholder communication.
As experience grows, the questions usually become less defined and the analyst takes more ownership of deciding what should be measured, how it should be checked and how the result should be used.
Looking for qualifications, career progression and Kenya job context?
The complete Data Analyst workplace workflow
The exact order can loop—an unexpected result may send you back to the source or definition—but this is a useful mental model for real work.
Request
Someone needs an answer.
Clarify
Turn vague wording into a precise question.
Define
Agree what the metric and terms mean.
Scope
Choose the period, records and groups that belong.
Retrieve
Get the required data from approved sources.
Inspect
Look for missing, duplicated or suspicious records.
Analyse
Calculate, compare and investigate.
Reconcile
Check important totals against trusted sources.
Interpret
State what the evidence supports—and what it does not.
Communicate
Explain the finding in language the audience understands.
Deliver
Use the right output: report, dashboard, workbook, query, email or presentation.
Follow up
Answer questions, document changes or automate recurring work.
What beginners often imagine versus what happens at work
| Beginner expectation | Workplace reality |
|---|---|
| “Someone will give me a perfect dataset.” | Records may be incomplete, inconsistent or poorly documented. |
| “Someone will give me a perfect question.” | Requests are often vague and need clarification. |
| “SQL returned results, so the answer is correct.” | A valid query can still use the wrong records, join or definition. |
| “A dashboard is the analysis.” | The dashboard is usually one output from a larger checking and reasoning process. |
| “Missing means zero.” | Missing information can have several meanings. |
| “An unusual value is an error.” | It might be an error—or the most important finding. |
| “Automation means the report is safe.” | Automated reports still need monitoring and validation. |
| “AI can verify the answer for me.” | The analyst remains responsible for the definition, data, logic and conclusion. |
On a small screen, swipe the table sideways to view every column.
Before you press Send
Five final checks before sharing an analysis
A technically impressive result can still fail
Use this final check to catch a wrong question, wrong record set, weak verification or an overconfident conclusion before another person acts on it.
- Am I answering the question that was actually asked?
- Do I understand what the records represent and which records belong?
- Did I verify the important number against an appropriate source or independent check?
- Does my conclusion go further than the evidence supports?
- Can the reader understand what happened and what should happen next?
Frequently asked questions
What does a Data Analyst do every day?
The work can combine recurring reports, ad-hoc questions, SQL or spreadsheet analysis, data-quality checks, dashboard maintenance, reconciliation, meetings, documentation and communicating findings. The exact mixture varies by organisation and by day.
Do Data Analysts code every day?
Not always. Some roles use SQL or Python heavily; others rely more on spreadsheets, BI tools, established reports and meetings. Coding is a tool for answering the analytical question, not the purpose of the job.
Do Data Analysts spend all day making dashboards?
No. A dashboard is one possible output. Requirements, metric definitions, data retrieval, quality checks, analysis, reconciliation and communication can all happen before a dashboard is ready to trust.
Do Data Analysts have meetings?
Often, yes. Meetings can clarify a request, agree definitions, explain a surprising result, present findings or decide which evidence should be collected next.
What is an ad-hoc analysis request?
It is a question requested for a particular immediate need rather than an established scheduled report—for example, “Why did refunds increase during the last three days?”
What happens when a Data Analyst finds incorrect data?
The analyst investigates the source of the problem, identifies which results were affected, corrects or escalates the issue appropriately, reruns the affected calculations and communicates any necessary correction.
Is Data Analysis mostly mathematics?
Mathematics and statistics matter, but everyday work also requires understanding business questions, retrieving data, checking quality, defining measures, using appropriate software and communicating findings.
Does AI replace Data Analysts?
AI can assist with queries, report creation, summaries and exploration, but the analyst still has to verify definitions, source data, filters, calculations, privacy, context and conclusions. [4]
So, what does a Data Analyst actually do at work?
The shortest accurate answer is:
Sometimes that means SQL. Sometimes Excel. Sometimes Power BI or Python. Sometimes it means a meeting. Sometimes it means spending two hours finding why two totals disagree. The tools change; the responsibility to understand, check, analyse, verify and explain does not.
And sometimes the most professional conclusion is: “The data shows where the problem is, but it does not yet prove why.”
Build practical analytical skills
Learn Data Science & Analytics at Newton Institute of Technology
NIT’s Data Science & Analytics programme develops practical skills in Excel, statistics, Python, SQL, data cleaning, visualisation and introductory machine learning. The current programme has six months of institutional training followed by a compulsory three-month industrial attachment, with physical and online study options. [5]
Study optionsPhysical · evening · online
LocationMigori Town, Kenya
Call / WhatsApp+254 713 584 858
Sources and references
- Microsoft Learn — Study guide for Exam PL-300: Microsoft Power BI Data Analyst. Current role framework and skills measured as of 20 April 2026, including stakeholder requirements, data preparation, modelling, analysis, visualisation, data quality, refresh and governance. Microsoft Learn.
- PostgreSQL Global Development Group — PostgreSQL 18 Tutorial. Primary technical documentation for relational database concepts, SQL queries, joins and aggregate functions. PostgreSQL documentation.
- Kenya Law — Data Protection Act, 2019, section 25. Primary statutory source for principles including lawful and transparent processing, purpose limitation, data minimisation, accuracy, retention and safeguards for transfers. Kenya Law.
- Microsoft Learn — Summarize a report with Copilot in Power BI. Current documentation for report-grounded Copilot summaries, questions, best practices, limitations and report-design considerations. Microsoft Learn.
- Newton Institute of Technology — Data Science and Analytics Course in Kenya. Current institutional source for programme content, six-month institutional training, compulsory three-month attachment, study options and KSh 60,000 school fee. View the programme.