The analyst workflow

Good data analysis follows a repeatable process. The analyst workflow takes you from a question to an answer someone acts on — and knowing the steps keeps you from flailing in a spreadsheet. This lesson covers the analyst workflow. Let's map the process. Let's learn the workflow.

The steps from question to decision

THE DATA-ANALYSIS WORKFLOW — a repeatable process every analysis follows (in Excel, SQL, Python,
or a BI tool alike). The thinking is the same regardless of tool:

1. ASK — define the QUESTION / goal. What are you trying to answer or decide? A clear, specific
   question drives everything (asking-the-right-question lesson). Vague question -> useless
   analysis.

2. COLLECT / GET THE DATA — gather the data that can answer it (databases, spreadsheets, APIs,
   surveys, exports). Know where it lives + get the RIGHT data (collecting lesson).

3. CLEAN / PREPARE — fix the data so it's trustworthy: handle missing values, remove duplicates,
   fix types + formats, structure it tidily. THE BIGGEST step (~80% of the work) + the least
   glamorous. Garbage in -> garbage out (cleaning lessons).

4. EXPLORE (EDA) — get to know the data: summary statistics, distributions, counts, quick charts.
   Understand its shape, spread, + oddities BEFORE deep analysis (summarising lesson).

5. ANALYSE — answer the question: filter, group + compare, find trends + relationships,
   calculate metrics. The core "finding the answer" step (grouping/trends lessons).

6. VISUALISE — turn findings into clear CHARTS that reveal the pattern at a glance (choosing-the-
   right-chart lesson).

7. COMMUNICATE — tell the STORY: state the insight + recommendation in plain language to the people
   who'll act on it (story + presenting lessons). The finish line.

IT'S ITERATIVE, NOT LINEAR — you loop back constantly: exploring raises new questions; analysis
reveals dirty data to clean; findings prompt new angles. Expect to go back + forth, not march
straight through.

The data-analysis workflow — a repeatable process every analysis follows (in Excel, SQL, Python, or a BI tool alike — the thinking is the same regardless of tool): (1) ask (define the question / goal — what are you trying to answer or decide? — a clear, specific question drives everything; vague question → useless analysis); (2) collect / get the data (gather the data that can answer it — databases, spreadsheets, APIs, surveys, exports — know where it lives and get the right data); (3) clean / prepare (fix the data so it's trustworthy: handle missing values, remove duplicates, fix types + formats, structure it tidily — the biggest step, ~80% of the work, and the least glamorous — garbage in → garbage out); (4) explore (EDA) (get to know the data: summary statistics, distributions, counts, quick charts — understand its shape, spread, and oddities before deep analysis); (5) analyse (answer the question: filter, group + compare, find trends + relationships, calculate metrics — the core "finding the answer" step); (6) visualise (turn findings into clear charts that reveal the pattern at a glance); and (7) communicate (tell the story: state the insight + recommendation in plain language to the people who'll act on it — the finish line). It's iterative, not linear — you loop back constantly: exploring raises new questions; analysis reveals dirty data to clean; findings prompt new angles — expect to go back and forth, not march straight through.

Doing the workflow well

GETTING THE MOST FROM THE WORKFLOW:
- START WITH THE QUESTION, ALWAYS — every step serves answering it. Without a clear question you'll
  wander + produce charts nobody needs. (Step 1 is the most important.)
- SPEND THE TIME ON CLEANING/PREP — beginners rush to "analysis"; pros know ~80% is getting +
  cleaning the data. Clean data = trustworthy answers.
- EXPLORE BEFORE YOU CONCLUDE — always look at the data (summary stats, distributions) before
  deep analysis. It catches errors + reveals what's really there.
- KEEP IT REPRODUCIBLE + DOCUMENTED — note your steps, assumptions, + data sources. So you (+
  others) can re-run, trust, + audit the analysis. (Code/queries beat manual clicking here.)
- END WITH THE DECISION — the workflow isn't done at a chart; it's done when you've COMMUNICATED an
  insight someone can ACT on. Tie the finding to a recommendation.
- ITERATE — treat it as a loop: question -> data -> clean -> explore -> analyse -> new questions ->
  refine. Analysis is exploratory.

A SIMPLE MENTAL CHECKLIST FOR ANY ANALYSIS:
  What's the QUESTION? -> What DATA answers it? -> Is the data CLEAN + trustworthy? -> What does
  EXPLORING show? -> What's the ANSWER (analyse)? -> What CHART shows it clearly? -> What's the
  INSIGHT + recommendation (communicate)?

THE PRINCIPLE: follow the workflow — ASK -> COLLECT -> CLEAN -> EXPLORE -> ANALYSE -> VISUALISE ->
COMMUNICATE — iterating as you go. Start with a clear QUESTION, invest in CLEANING (~80%), EXPLORE
before concluding, keep it reproducible, + finish with a COMMUNICATED decision. This process turns
messy data into confident answers. Follow it every time.

Getting the most from the workflow: start with the question, always (every step serves answering it — without a clear question you'll wander and produce charts nobody needs; step 1 is the most important); spend the time on cleaning/prep (beginners rush to "analysis"; pros know ~80% is getting and cleaning the data — clean data = trustworthy answers); explore before you conclude (always look at the data — summary stats, distributions — before deep analysis — it catches errors and reveals what's really there); keep it reproducible + documented (note your steps, assumptions, and data sources so you and others can re-run, trust, and audit — code/queries beat manual clicking here); end with the decision (the workflow isn't done at a chart; it's done when you've communicated an insight someone can act on — tie the finding to a recommendation); and iterate (treat it as a loop: question → data → clean → explore → analyse → new questions → refine). A simple mental checklist: what's the question? → what data answers it? → is the data clean and trustworthy? → what does exploring show? → what's the answer (analyse)? → what chart shows it clearly? → what's the insight + recommendation (communicate)? The principle: follow the workflow — ask → collect → clean → explore → analyse → visualise → communicate — iterating as you go; start with a clear question, invest in cleaning (~80%), explore before concluding, keep it reproducible, and finish with a communicated decision — this process turns messy data into confident answers.

The mistake beginners make

The first mistake is skipping the question — diving into data/charts with no clear question, then wandering and producing useless output; always start with the question. The second mistake is rushing past cleaning — jumping to analysis on dirty data (garbage in, garbage out); invest ~80% in getting and cleaning the data. The third mistake is not exploring first — concluding before looking at the data's shape/errors; explore (EDA) first. And stopping at the chart — producing a visual but no communicated insight/decision; finish with the story and recommendation. And treating it as linear — expecting a straight march when analysis loops back; iterate. Start with the question, clean thoroughly, explore first, communicate the decision, and iterate.

Your turn

---

Your turn

1. Learn the 7 steps: ASK (define the question) -> COLLECT (get the data) -> CLEAN (fix it) -> EXPLORE (get to know it) -> ANALYSE (answer it) -> VISUALISE (chart it) -> COMMUNICATE (tell the story).
2. Always start with the question: make step 1 (a clear, specific question) the foundation of every analysis - every other step serves answering it.
3. Invest in cleaning: expect ~80% of the work to be collecting + cleaning + preparing the data (missing values, duplicates, types, structure) - clean data = trustworthy answers.
4. Explore before concluding: always look at the data (summary stats, distributions, quick charts) before deep analysis, to catch errors and see what's really there.
5. Finish with a decision: end the workflow not at a chart but at a COMMUNICATED insight + recommendation someone can act on - and treat the whole thing as an iterative loop, not a straight line.

Key points

• The ANALYST WORKFLOW (same in Excel/SQL/Python/BI): 1) ASK (define the QUESTION — the most important step), 2) COLLECT (get the right data), 3) CLEAN/PREPARE (missing/duplicates/types/tidy — ~80% of the work; garbage in -> garbage out), 4) EXPLORE/EDA (summary stats/distributions before concluding), 5) ANALYSE (filter/group/compare/trends — find the answer), 6) VISUALISE (clear charts), 7) COMMUNICATE (the insight + recommendation — the finish line).
• It's ITERATIVE, not linear — you loop back constantly (exploring raises questions; analysis reveals dirty data; findings prompt new angles).
• Do it well: START with a clear question (else you wander), invest in CLEANING (~80%), EXPLORE before concluding, keep it REPRODUCIBLE + documented (code/queries > manual clicks), END with a communicated DECISION (not just a chart), and iterate.
• The checklist: What's the QUESTION? -> What DATA answers it? -> Is it CLEAN? -> What does EXPLORING show? -> What's the ANSWER? -> What CHART shows it? -> What's the INSIGHT + recommendation?
• The mistakes: skipping the question (wander), rushing past cleaning (garbage in/out), not exploring first, stopping at the chart (communicate a decision), and treating it as linear (iterate).