32 ChatGPT prompts for data analysts to turn past projects into compelling case studies
Most data analysts have at least one really good project buried in a private Looker dashboard, a forgotten dbt repo, or a SQL snippet saved as final_query_v8_FINAL.sql. Almost none of them have a public case study to show for it. That’s the gap these 32 ChatGPT prompts for data analyst case study writing are designed to close.
In this guide, I’m going to give you the exact prompts I use, the sample output each one produces, and the 2026 hiring context that makes case studies matter more this year than last. You’ll get a comparison table, a 30-day sprint plan, a People Also Ask section, and a mistakes list - everything you need to ship 3 polished case studies before your next interview cycle.
Quick answer: The fastest way to turn an old project into a case study is to feed ChatGPT the context (the business question, the data you used, the tools, the actions you took) and then iterate through six blocks: Story, Method, Data, Visualization, Impact, and Writing. The 32 prompts below cover all six blocks in order. Open ChatGPT, paste prompt 1, and you can have a first draft of a case study in under an hour.
Why your SQL project is invisible (with a 2026 stat you can quote)
A data case study is a short public story that walks a stranger through one analytical project you shipped - the problem, the method, the data, the visualization, the result, and what changed because of it. The format is borrowed from product case studies and product design portfolios, but the bones are the same: a beginning, a middle, and a measurable end.
Here’s the cold water. The U.S. Bureau of Labor Statistics projects 34% employment growth for data scientists from 2024 to 2034 - “much faster than the average” - and the related “market research analyst and marketing specialist” category is projected to grow 7% over the same period (BLS Data Scientists, 2025; BLS Market Research Analysts, 2025). At the same time, the median annual wage for data scientists hit $112,590 in May 2024, and for market research analysts $76,950 (BLS Data Scientists, 2025; BLS Market Research Analysts, 2025). The talent market is bigger, the pay is higher, and the funnel is more crowded. Hiring managers are sorting through hundreds of applicants per role, and a GitHub link without a story is invisible next to a case study that reads in eight minutes.
Pull quote: “If AI can make the graph, what’s left for the human? The answer starts with judgment, context, and the responsibility to make data meaningful.” - Cole Nussbaumer Knaflic, founder of storytelling with data, May 2026.
That quote matters here because the case study is the layer AI can’t generate for you. ChatGPT can draft the first 80%. The remaining 20% - your real business judgment, your real metric movement, your real stakeholder moment - is what makes you hirable. The prompts below get you to that 80% in a hurry.
The 4-part data case study anatomy
Every strong data case study has the same four parts, in this order:
- Context and stakes. What was the business question, who needed the answer, and what was at risk if nobody answered it? This is the “so what” before the data.
- Method and data. What tools did you use (SQL, dbt, Tableau, Power BI, Looker, Mode, Hex, Snowflake, BigQuery, Python), what data did you touch, and how did you clean it? This is where analysts usually over-write. Keep it tight.
- Finding and visualization. What did the data say, and what chart made the answer obvious? A single hero image is worth more than ten bullet points.
- Action and impact. What did the stakeholder do, what metric moved, and what did you learn? This is the part most case studies skip. Don’t skip it.
The 32 prompts below are organized into six sections that match those four parts plus two helpers (visualization and writing). If you only have time to run a few, run prompts 1, 2, 6, 23, and 28 first - those five will give you a complete skeleton draft in one sitting.
SECTION 1: Story & context prompts (1–5)
Story and context is the section most analysts skip and hiring managers read first. If you nail this part, the rest of the case study writes itself.
Prompt 1 - The 6-sentence story spine
Purpose: Build a tight opening paragraph that frames the project in plain English, before any tool or chart is mentioned. Think of this as your case study’s “logline.”
Act as a portfolio coach for data analysts. I'm going to paste rough notes
about a past project below. Your job is to rewrite them as a 6-sentence
opening paragraph for a public case study. Use this exact structure:
Sentence 1: Who I was (role, team, company type).
Sentence 2: What the business question was, in plain English.
Sentence 3: Why it mattered, including a dollar or volume number if I have one.
Sentence 4: What I had to figure out (the messy part).
Sentence 5: What I built or did (the one-line method).
Sentence 6: What changed because of my work, in a measurable way.
Rules:
- Write for a hiring manager who has 90 seconds.
- Avoid the words "delve," "leverage," "robust," "tapestry," and "moreover."
- Use contractions.
- If a number is missing, leave a [BRACKET] placeholder instead of inventing.
- Keep the whole paragraph under 110 words.
Here are my notes:
[PASTE YOUR ROUGH NOTES, MESSY THOUGHTS, OR EVEN JUST BULLETS]Example output (using a synthetic project about a DTC coffee subscription):
I was a data analyst on the growth team at a 40-person DTC coffee subscription in 2025. The VP of Marketing asked, “Why did monthly churn jump from 4.1% to 6.8% between March and May?” The spike cost roughly $180K in annualized subscription revenue if it held. I had to figure out which customer cohort, channel, and product line was driving the bleed. I built a churn decomposition in SQL and dbt, then a cohort waterfall in Tableau. After we acted on the finding, 90-day churn dropped to 4.4% by the end of Q3.
Pro tip: If the prompt gives you a wall of text, ask ChatGPT to “shrink this to 80 words while keeping the metric.” Recursive brevity is a real skill.
Prompt 2 - The “before and after” anchor
Purpose: Create the single sentence that explains why this project existed in the first place. This becomes the callout at the top of your case study and the line you paste into LinkedIn.
Here's a 6-sentence opening paragraph for a data case study:
[PASTE THE OUTPUT FROM PROMPT 1]
Rewrite it as a single "before and after" sentence in this format:
"Before my work, [SPECIFIC BAD STATE with a number]. After my work,
[SPECIFIC GOOD STATE with a number], because [ONE-LINE CAUSE]."
Constraints:
- Total length under 35 words.
- Both states must include a concrete number, percentage, or dollar figure.
- If the second number is smaller (an improvement), make the direction obvious.
- Output three candidate versions so I can pick the strongest.Example output:
Before my work, monthly churn was 6.8% and bleeding $180K in annualized revenue. After my work, churn dropped to 4.4% because we paused the underperforming paid-social cohort and reworked the onboarding email. (28 words)
Pro tip: This single sentence is also the perfect cover-letter line and the first line of your resume bullet. Mine it for everything.
Prompt 3 - The stakeholder map
Purpose: Make sure you’ve named the humans involved. Hiring managers want to see you work with non-analysts, not just data.
For this data case study:
[PASTE PROJECT DESCRIPTION]
Build a 4-column stakeholder table with these columns:
1. Person or role (use a job title, not a real name unless I provide one)
2. What they cared about
3. What they pushed back on
4. How I got their buy-in
Include at least 4 rows: the requester, the data owner, the skeptic, and the executive sponsor. For each row, be specific - no "they wanted insights." Real motivations, real objections.Example output (truncated):
| Person or role | What they cared about | What they pushed back on | How I got their buy-in |
|---|---|---|---|
| VP of Marketing (requester) | Stopping the churn bleed | ”Don’t waste two weeks on a post-mortem” | Showed a 3-slide teaser in 48 hours |
| Data Engineering Lead (data owner) | Pipeline uptime | ”Don’t write new dbt models in the middle of a sprint” | Reframed it as a one-table audit, not a project |
| Head of Finance (skeptic) | Confidence in the number | ”Last time, the analyst was off by 30%“ | Hand-built a backtest against the prior quarter |
| CEO (executive sponsor) | Quarterly OKR impact | ”Will this delay the launch?” | Promised a 10-day turnaround, hit 8 |
Pro tip: A real stakeholder table is the single best signal to a hiring manager that you can actually function inside a company. It does more work than any chart.
Prompt 4 - The “what could have gone wrong” beat
Purpose: Add the risk and constraint layer. Most analyst case studies pretend the project was clean. Real ones show the mess.
For this data case study:
[PASTE PROJECT DESCRIPTION OR PROMPT 1 OUTPUT]
Write a 120-word section titled "What could have gone wrong" that
covers three risks:
1. A data risk (something wrong with the source data).
2. A stakeholder risk (someone who could have killed the project).
3. A time or technical risk (something that nearly blew the deadline).
For each risk, name it, explain the early warning sign, and tell me
what I did about it. Write in first person, present tense for the risks,
past tense for the action I took.Example output: The dbt model for monthly churn was a brand-new build and the production table was 4 days behind real time, so the first version of the cohort analysis was 11% off. I caught it by sanity-checking against a manual Stripe export. I flagged the pipeline lag to the data engineering lead on day 2; we agreed to backfill before finalizing the cohort cuts. The VP of Marketing was skeptical after the Q1 forecast was 30% off, so I built a backtest from the prior quarter and shared it as an appendix slide. We had 8 working days before the leadership review, and I spent 2 of them on data quality.
Pro tip: This section is also a great “lessons learned” callout for your portfolio index page.
Prompt 5 - The “so what in one breath” elevator pitch
Purpose: Get a one-sentence summary you can paste at the top of the case study, on your resume, and on LinkedIn.
I just finished a data case study draft. Here's the project:
[PASTE THE 6-SENTENCE SPINE FROM PROMPT 1]
Write 3 versions of a one-sentence pitch I can say out loud, in under
10 seconds, that captures the project, the action, and the result.
Rules:
- All three versions must include at least one number.
- Version 1: punchy and direct.
- Version 2: more conversational, suitable for an interview.
- Version 3: tag-led, suitable for a portfolio headline like
"How I cut churn 35% with one SQL query."Example output:
- I used a cohort waterfall to identify a $180K-a-year churn spike in our DTC subscription and helped cut it 35% in one quarter. (punchy)
- Last year I found that one paid-social cohort was driving 40% of our churn increase, and we paused it within 48 hours. (conversational)
- How I cut churn 35% with one Tableau cohort waterfall and a 48-hour pause. (tag-led)
Pro tip: Use version 3 as the title of your case study and version 1 as the meta description of your portfolio page.
SECTION 2: Methodology prompts (6–11)
Method is where you prove you actually did the work. The trick is to be specific without writing a textbook chapter. Each of these prompts gives you one tight, named section.
Prompt 6 - The tool stack summary
Purpose: Generate the “how I built this” sidebar. Hiring managers scan this to see if your stack matches the role.
For this data case study:
[PASTE THE PROJECT DESCRIPTION]
Build a 4-row summary of the tool stack I used, in this exact format:
| Step | Tool | Why I picked it | What I would change next time |
Cover at least these 4 steps: data storage, transformation, analysis,
and visualization. If I used ChatGPT, Claude, or another LLM, add a
5th row for "AI assistance." Keep each cell under 20 words.Example output:
| Step | Tool | Why I picked it | What I would change next time |
|---|---|---|---|
| Data storage | Snowflake | Already the company warehouse, no new infra | Would push for BigQuery for cost |
| Transformation | dbt | Version-controlled SQL, peer review built in | Would add dbt tests for null user_id |
| Analysis | Mode SQL + Jupyter | Mode for fast cohort cuts, Jupyter for backtest | Would consolidate in Hex |
| Visualization | Tableau | Stakeholders already licensed, easy to share | Would add a Looker view for self-serve |
| AI assistance | ChatGPT | Drafted the case study narrative from my notes | None - this section was a time-saver |
Pro tip: The “what I would change next time” column is your secret weapon. It signals senior judgment.
Prompt 7 - The method in 3 sentences
Purpose: Get a one-paragraph method summary that sits under your opening story and above your charts.
Here is the case study opening:
[PASTE PROMPT 1 OUTPUT]
Write a 3-sentence "Method" paragraph that:
- Names the primary analysis technique (e.g., cohort decomposition,
regression, A/B test read-out, time-series forecast).
- Names the data sources by type, not by exact table name.
- Names the single most important assumption behind the method.
Keep the total length under 80 words. If you use a technical term like
"cohort decomposition," define it in 6 words the first time.Example output: I used a cohort decomposition, which means I split churned customers into the month they started and tracked them forward. I pulled subscription events from Snowflake and joined them to ad-spend data in BigQuery via a shared user_id. The biggest assumption was that the cancellation date in our billing system matched the customer’s actual decision date - I validated it against a 5% survey sample.
Pro tip: The “single most important assumption” is the line that separates a senior analyst from a junior one. If you can name it, you can defend it in an interview.
Prompt 8 - The “this is not a textbook” honesty beat
Purpose: Add a short paragraph about what you skipped, what’s noisy in the data, and what a textbook would have done differently. This is gold for senior roles.
For this case study:
[PASTE METHOD PARAGRAPH FROM PROMPT 7]
Write a 100-word section called "Where I cut corners" that lists 2 or 3
things I deliberately did NOT do, and why each was the right call given
the deadline. Examples of what to mention: skipped a full causal model,
used a 5% survey instead of full census, did not build a streaming
dashboard, used a heuristic instead of ML, etc.
Be honest. Hiring managers trust analysts who know their shortcuts
were intentional.Example output: I did not run a formal causal model because the marketing team needed a decision in 8 days, not a 6-week DAG build. I used a 5% cancellation-reason survey instead of a full census to keep the turnaround tight, and I checked the result against the prior quarter’s churn backtest, which was within 4%. I also did not build a streaming dashboard - a static Tableau view was enough for a one-time read-out, and the team now refreshes it weekly in Looker.
Pro tip: Save this beat for case studies where you had a real deadline. Skip it for learning projects where the deadline was self-imposed.
Prompt 9 - The decision tree
Purpose: Turn the method into a visual that hiring managers can follow in 20 seconds. This works especially well for “I had three options and picked one” stories.
For this case study:
[PASTE METHOD PARAGRAPH]
Draw an ASCII decision tree with this structure:
- Root node: the business question.
- 3 branches: the 3 methods I considered.
- For each branch: one line on what it would have given me, and one
line on why I rejected it (or picked it).
- Leaf node: the method I chose and the one-line reason.
Use box-drawing characters or simple ASCII so it pastes cleanly into
Markdown. Keep it under 25 lines total.Example output:
Q: Why did churn jump 2.7 pts in 2 months?
├── Cohort decomposition → Pinpoints which start month is bleeding
│ (CHOSEN - fast, matches stakeholder mental model)
├── Multivariate regression → Would have given feature importance
│ (REJECTED - too slow, 6-week build, no peer review for stats)
└── Manual cancellation reasons → Qualitative
(REJECTED - only 5% survey response, noisy)Pro tip: ASCII decision trees screenshot well. Convert it to a Lucidchart or Excalidraw image before publishing and it becomes a hero visual.
Prompt 10 - The validation step
Purpose: Explain how you knew your answer was right. Hiring managers love this section because it shows you don’t trust your own output by default.
For this case study:
[PASTE METHOD PARAGRAPH]
Write a 120-word "How I validated the result" section covering:
1. A backtest against a known historical period.
2. A sanity check against a second data source.
3. A peer review or stakeholder challenge I survived.
If I didn't do one of these, say so honestly and tell me what I should
have done instead. Be specific about the numbers - exact percentages,
not vague "I checked it was right."Example output: I ran the same cohort cut on Q1 churn data, where the cause was already known to be a price change, and the model correctly flagged that cohort. The result also lined up with a 5% survey response rate where 38% of churned users cited “too many emails” as the top reason - matching the same time period. I sent the draft to the Head of Finance on day 6; she flagged a 4% gap between my number and the Stripe export, and we resolved it by switching from cancellation date to last-active date.
Pro tip: “Resolved a 4% gap by switching from cancellation date to last-active date” is the kind of sentence that gets you hired. It shows domain judgment.
Prompt 11 - The “what I’d do differently” reflection
Purpose: A short reflective beat that turns a project writeup into a learning artifact. Great for portfolios, less critical for resumes.
For this case study:
[PASTE EVERYTHING ABOVE]
Write a 150-word reflection titled "What I'd do differently" with
exactly 3 numbered points. Each point must include:
- The thing I would change.
- The trigger that would have made me change it earlier.
- One specific tool or technique I would add next time.
Avoid corporate language. Write like you're texting a friend who's
also an analyst.Example output:
- I would have started with the cancellation-reason survey, not ended with it. The data was sitting in Typeform the whole time and I would have caught the “too many emails” signal on day 2 instead of day 6.
- I would have set up a dbt test for null user_id in the subscription events table. We had a 0.4% null rate that almost cost me a day.
- I would have used Hex instead of switching between Mode and Jupyter. Half my backtest time was copy-pasting between the two.
Pro tip: Three short numbered points is the right length. Five feels like a confession.
SECTION 3: Data & query prompts (12–17)
Data is the part most analysts bury under buzzwords. These prompts help you name the dataset, the cleanup, and the SQL or Python in a way a non-analyst can scan.
Prompt 12 - The data source map
Purpose: Build the data provenance section. This is the boring-but-critical part of any case study that touches PII, sampling, or joins.
For this data case study:
[PASTE THE PROJECT DESCRIPTION]
List the data sources I used, in this exact format for each:
**Source name**
- Type: (e.g., production database, third-party API, public dataset)
- Approximate size: (rows, GB, or "small, under 1M rows")
- Owner team: (e.g., data engineering, marketing ops)
- Refresh cadence: (real-time, hourly, daily, weekly)
- Why I trusted it: (one sentence)
- One risk: (one sentence)
Cover at least 3 sources. If I used a public dataset, name it (Kaggle,
UCI, data.gov, etc.).Example output:
subscription_events
- Type: production database, Snowflake
- Approximate size: 12M rows, 4 GB
- Owner team: data engineering
- Refresh cadence: hourly
- Why I trusted it: it’s the source Stripe writes to for billing events
- One risk: cancellation events lagged by 2–6 hours during peak load
ad_spend_daily
- Type: third-party API export (Meta, Google)
- Approximate size: 90K rows, small
- Owner team: marketing ops
- Refresh cadence: daily
- Why I trusted it: reconciled against platform UI within 0.5% weekly
- One risk: Meta changed the API field names in April, broke my join
cancellation_survey
- Type: Typeform export
- Approximate size: 1.2K rows, small
- Owner team: growth
- Refresh cadence: continuous
- Why I trusted it: 5% response rate matches the industry benchmark
- One risk: respondents skew toward unhappy users
Pro tip: If your case study uses only one source, the “one risk” column forces you to be honest about the limitation.
Prompt 13 - The “cleaning is analysis” paragraph
Purpose: Make data cleaning visible without making it the whole case study. Hiring managers want to know you can do this, but they don’t want a war story.
For this data case study:
[PASTE PROJECT DESCRIPTION AND DATA SOURCES]
Write a 100-word "Cleaning" paragraph that names:
- The 2 or 3 biggest data quality issues I found.
- The single most important cleanup decision I made.
- One thing I would automate next time using dbt, Great Expectations,
or a similar tool.
Do not list every column I checked. Pick the 2 or 3 that actually
changed the result.Example output: The biggest data quality issues were a 0.4% null user_id rate in subscription_events (dropped 3 days of cohort data until I backfilled from Stripe), a duplicate-cancellation event in 0.2% of records (deduplicated by keeping the latest event), and a timezone mismatch between ad_spend_daily (UTC) and the subscription table (PST). I converted everything to UTC at the dbt staging layer, which meant every downstream model stopped needing its own conversion. Next time, I would codify the dedup logic as a dbt test, not a one-off SQL CASE WHEN.
Pro tip: “0.4% null user_id” is a specific number. “Some nulls” is not. Always trade in specifics.
Prompt 14 - The named SQL technique
Purpose: Show your technical chops in one tight code block with a 2-line explanation. This is where you earn SQL street cred.
For this data case study:
[PASTE METHOD PARAGRAPH]
Pick the single most important SQL technique I used in this project
(e.g., window function, self-join, CTE chain, QUALIFY clause, LAG,
pivot, cohort grid). Write a 1-paragraph explanation of why this
technique was the right call, then a 6-12 line SQL snippet that
demonstrates it.
Rules:
- Use ANSI SQL or BigQuery/Snowflake syntax, no MySQL-only features.
- The snippet must be runnable in isolation against a synthetic
table called `events` with columns: user_id, event_type,
event_timestamp, plan_type.
- Comment each line of the SQL.
- After the snippet, write 2 sentences on what the output would
look like in a real run.Example output: I used a LAG window function inside a CTE to compute the gap between each user’s last_active event and their cancellation event. Window functions let me keep the original row context while computing the gap, which a self-join would have made messier and slower.
WITH last_seen AS (
SELECT
user_id,
event_timestamp AS last_active_at,
plan_type,
ROW_NUMBER() OVER (
PARTITION BY user_id
ORDER BY event_timestamp DESC
) AS rn
FROM events
WHERE event_type IN ('active_session', 'cancellation')
),
gaps AS (
SELECT
user_id,
plan_type,
DATEDIFF(
'day',
MAX(CASE WHEN rn = 2 THEN last_active_at END),
MAX(CASE WHEN rn = 1 THEN last_active_at END)
) AS days_to_cancel
FROM last_seen
GROUP BY 1, 2
)
SELECT plan_type, AVG(days_to_cancel) AS avg_days
FROM gaps
GROUP BY 1
ORDER BY avg_days DESC;Expected output: a 3-row table showing average days between last active session and cancellation, broken down by plan type. In our case, the monthly plan had a 4-day median gap and the annual plan had a 21-day gap, which alone was a story.
Pro tip: Annotate the snippet with what it would return. Hiring managers read the comment block, not the SQL.
Prompt 15 - The “if I had to do it again” data model
Purpose: Show data modeling maturity. This is the section that separates analysts who know SQL from analysts who know data architecture.
For this data case study:
[PASTE METHOD PARAGRAPH AND DATA SOURCES]
Draw an ASCII star schema or entity-relationship diagram for the
data model I wish I'd had. Include:
- 1 fact table (e.g., fct_subscription_events).
- 3 to 4 dimension tables (e.g., dim_user, dim_plan, dim_channel).
- The grain of the fact table in one line at the top.
- One example dbt model name for each table.
Keep it under 30 lines. Use simple ASCII so it pastes into Markdown.Example output:
Grain: 1 row per user-event per day.
fct_subscription_events
├── user_id → dim_user
├── plan_id → dim_plan
├── channel_id → dim_channel (paid social, organic, referral, etc.)
└── event_date → dim_date
dim_user (SCD type 2, user attributes over time)
dim_plan (plan_type, billing_interval, monthly_price)
dim_channel (channel, campaign_id, attribution_model)
dim_date (calendar lookup)
Example dbt model names:
- stg_stripe__subscription_events.sql
- dim_user.sql
- fct_subscription_events.sqlPro tip: A star schema diagram in a portfolio signals that you understand analytics engineering, not just ad-hoc SQL. Worth a screenshot.
Prompt 16 - The “I almost got fooled” data caveat
Purpose: Add a single cautionary sentence that proves you can spot a bad number in the wild. This is a senior signal.
For this data case study:
[PASTE METHOD PARAGRAPH AND DATA SOURCES]
Write a 90-word "What almost fooled me" section that describes a
single moment when the data looked convincing but was wrong, and
the one check that saved me. End with a single-sentence lesson
that an entry-level analyst could apply on their next project.Example output: On day 4, the cohort chart showed a clean monotonic decline in churn, exactly the textbook shape. I almost presented it. The check that saved me was comparing the totals to a 5% sample of the raw Stripe CSV - the totals were off by 11% because of a timezone conversion bug. The lesson: when the data looks too clean, it usually is. Always spot-check against a row-level source before you build a chart on top of an aggregate.
Pro tip: The “too clean” line is a useful checklist item for any analyst - present it like a proverb.
Prompt 17 - The data glossary sidebar
Purpose: Build a small sidebar or footnote that makes the case study approachable to non-data readers. Useful when the audience is a hiring manager from a non-technical team.
For this data case study:
[PASTE PROJECT DESCRIPTION]
Generate a 6-row data glossary for the most important terms in
the case study. For each term, give:
- The plain-English definition (one sentence).
- An example from THIS project (not a generic example).
- A "say this instead" alternative for non-technical readers.
Cover at least: cohort, churn, MRR, attribution, retention, and one
project-specific term.Example output:
| Term | Plain English | Example from this project | Say this instead |
|---|---|---|---|
| Cohort | A group of users defined by when they started | The March 2025 signup cohort | ”people who joined in March” |
| Churn | The % of subscribers who cancel in a period | 6.8% in May, up from 4.1% | “people who canceled” |
| MRR | Monthly recurring revenue, the predictable slice | $310K in May, projected $285K in June | ”monthly subscription revenue” |
| Attribution | How we credit a signup to its source channel | 60% paid social, 25% organic, 15% referral | ”which channel drove the signup” |
| Retention | The % of a cohort still active at day N | 71% of the March cohort was active at day 90 | ”the percentage still with us” |
| Cancellation lag | Days between last-active event and cancel event | 4 days for monthly, 21 days for annual | project-specific |
Pro tip: This glossary doubles as the FAQ section of your portfolio page. Reuse it.
SECTION 4: Visualization & design prompts (18–22)
Visualization is the section that gets you the screenshot. These prompts help you pick the right chart, write the chart title, and design for a non-analyst reader.
Prompt 18 - The hero chart decision
Purpose: Pick the single chart that anchors the case study. Most case studies bury the answer in a dashboard. The hero chart is the one you’d print and tape to a wall.
For this data case study:
[PASTE METHOD AND IMPACT NOTES]
Recommend the single best hero chart for this case study. Tell me:
1. The chart type (e.g., cohort waterfall, before/after bar, slope chart,
annotated line chart, small multiples, treemap).
2. Why this chart type is right (1 sentence).
3. The exact title that should sit above the chart.
4. The one sentence of body text under the chart.
5. The single annotation I should add directly to the chart.
6. The color palette (1–2 colors max) and the data-ink ratio goal.
Then provide a 1-paragraph "If I had a second chart" recommendation.Example output:
- Chart type: cohort waterfall.
- Why: it shows the gap between the bad month and the recovery month in a single image, which is the only thing the reader needs to know.
- Title: “We cut monthly churn from 6.8% to 4.4% in 90 days by pausing one paid-social cohort.”
- Subtext: a one-line note that says “March to May 2025; the green bar is the post-action cohort.”
- Annotation: a callout on the worst cohort with the words “Paid social, Mar–May.”
- Color palette: a single accent red for the bad bar, single accent green for the recovered bar, gray for everything else. Data-ink ratio: drop gridlines, drop the second y-axis, drop the legend.
If I had a second chart, I’d add a small-multiples grid of the same waterfall broken down by plan type, with the same red/green coding, so the reader can see that the win was concentrated in the monthly plan.
Pro tip: The “title” is actually the most important sentence in the case study. A chart title that includes the action and the result turns a viewer into a believer in 3 seconds.
Prompt 19 - The chart title rewrite
Purpose: Replace a descriptive title with a takeaway title. This is the single highest-leverage edit you can make to a chart.
Here is a chart title from my case study:
[PASTE TITLE]
Here is the data the chart shows:
[PASTE 1-SENTENCE DESCRIPTION]
Rewrite the title in 3 styles:
1. The "newspaper headline" version (active verb, includes a number).
2. The "Twitter thread" version (one short sentence, ends in a colon).
3. The "academic" version (descriptive, no verbs).
For each version, write a 1-sentence subtitle that explains the
"so what" in plain English. The reader should be able to understand
the chart from the title and subtitle alone.Example output:
| Version | Title | Subtitle |
|---|---|---|
| Newspaper headline | Churn dropped 35% in 90 days after we paused paid social | The March 2025 paid-social cohort was 38% of the bleed |
| Twitter thread | We cut churn 35% by pausing one ad campaign: | The lift came from the monthly plan, not the annual plan |
| Academic | Monthly churn rate, March–August 2025, with paid-social intervention | The vertical line marks the May 14 pause of the underperforming cohort |
Pro tip: Always use the newspaper version as the main title. The other two are useful for academic or LinkedIn rewrites.
Prompt 20 - The “I removed this” anti-junk beat
Purpose: Demonstrate that you understand design restraint. This is a quiet but powerful senior signal.
For this data case study hero chart:
[PASTE CHART DESCRIPTION OR SCREENSHOT TEXT]
Write a 100-word "What I removed" section that lists 3 chart elements
I deliberately removed and why each removal made the chart clearer.
Cover at least:
- Gridlines (or the lack thereof).
- The second y-axis.
- The legend.
- 3D effects, if any.
- Pie chart (if applicable - explain why you avoided it).
This proves I can make a chart that an executive can read in 10 seconds.Example output: I removed the second y-axis because the absolute dollar line and the percentage line were on different scales, and the right-side axis was confusing two readers in dry-runs. I removed the legend - the chart now has a single label on each bar, so the legend was dead weight. I removed the gridlines and replaced them with a single faint horizontal line at the pre-action baseline. I almost used a pie chart for the channel mix breakdown and switched to a horizontal bar because the two small slices in the pie were unreadable.
Pro tip: This beat also pairs beautifully with a storytelling with data before-and-after screenshot of the chart’s evolution.
Prompt 21 - The accessibility pass
Purpose: Make sure the chart works for colorblind readers and screen-reader users. Two minutes of work; very high signal.
For this data case study hero chart:
[PASTE CHART DESCRIPTION]
Write a 60-word accessibility checklist covering:
1. Color choice (which colorblind-safe palette I used, e.g., Okabe-Ito).
2. Text contrast (WCAG AA or AAA on the chosen background).
3. Direct labels vs legend (and why direct labels are better for screen readers).
4. Alt text for the image, under 125 characters.
5. The one chart element I'd add a pattern or texture to, beyond color.
Be specific. Don't say "use accessible colors" - name the palette and
the contrast ratio.Example output:
- I’m using the Okabe-Ito 8-color palette, which is the standard colorblind-safe option, with a deep red (#D55E00) for “bad” and teal (#009E73) for “good.”
- Text is dark gray (#222) on a near-white background (#FAFAFA), which gives a contrast ratio of about 13:1, well above WCAG AA.
- Direct labels on every bar - no legend, since screen readers can’t tie a swatch in a legend to a bar in a chart.
- Alt text: “Bar chart showing monthly churn falling from 6.8% in May to 4.4% in August 2025, with the May bar in red and the August bar in green.”
- I’d add a subtle dot pattern to the green bar so colorblind readers can still distinguish it from a plain gray.
Pro tip: If you use Tableau or Power BI, the default color palette is usually not colorblind-safe. Override it. This is a 30-second fix.
Prompt 22 - The Tableau / Power BI / Looker setup note
Purpose: Show that you can publish a chart in the tool the team uses, not just export a PNG. This is the difference between “I can analyze data” and “I can ship a chart to a stakeholder.”
For this data case study hero chart:
[PASTE CHART DESCRIPTION AND TOOL USED]
Write an 80-word "How to reproduce this" section that tells another
analyst how to rebuild the chart in the same tool I used. Cover:
- Data source connection (one line).
- The single calculated field or measure I had to write (with the formula).
- The chart type in the tool's native menu name.
- The one filter or parameter that drives the highlight.
- A link to download the workbook if I have one hosted publicly
(Tableau Public, Power BI public, Looker public, etc.).
Use the actual menu names - "Marks card" not "the formatting tab."Example output: Connect to the Snowflake fct_subscription_events view. The only calculated field is churn_rate = COUNT(DISTINCT CASE WHEN event_type='cancellation' THEN user_id END) / COUNT(DISTINCT user_id). Use a “Bar Chart” mark with cohort_month on Columns and churn_rate on Rows. Add a highlight_cohort parameter as a filter, with the value “2025-03” to drive the red bar. The workbook is published on Tableau Public at [link].
Pro tip: Publishing a public Tableau Public or Power BI Public link is the single highest-trust signal in a case study. It lets the reader click through and check your work.
SECTION 5: Impact & outcome prompts (23–27)
Impact is the section most analysts skip. Don’t. It’s also where AI-assisted first drafts are weakest, because the metric movement is your specific business result. Use these prompts to write the section that gets you the interview.
Prompt 23 - The quantified impact beat
Purpose: Force yourself to put a number on the win. If you can’t, the project may not be a case study yet - and that’s okay.
For this data case study:
[PASTE THE WHOLE DRAFT SO FAR]
Write a 100-word "Impact" section that covers:
1. The primary metric that moved, with the before-and-after number.
2. The secondary metric that moved, even if smaller.
3. The dollar value of the impact, if calculable.
4. The time window over which the impact was measured.
5. One counterfactual: what would have happened if the work hadn't been done?
If I can't quantify an impact, write the section anyway but mark the
quantification as [UNKNOWN - NEED TO CONFIRM]. Do not invent numbers.Example output: Monthly churn dropped from 6.8% in May to 4.4% in August, a 35% relative reduction. The secondary metric, 90-day retention for the March cohort, recovered from 61% to 71%. The annualized revenue impact was approximately $180K, calculated by multiplying the recovered subscription count by the average monthly plan price. The time window was 90 days post-action. The counterfactual: based on the prior 3 months’ trend, churn would have likely crossed 8% by Q4, which would have triggered a board-level conversation about pausing all paid acquisition.
Pro tip: If you can’t quantify the impact, the case study is still worth publishing - but be honest about it. Hiring managers trust analysts who can say “I think it was X, but I’d need a longer measurement window to confirm.”
Prompt 24 - The “before vs after” sentence
Purpose: Get a single line you can use in the case study title, the resume bullet, and the LinkedIn post. This is the one line that matters most.
For this data case study:
[PASTE THE IMPACT SECTION FROM PROMPT 23]
Write 3 versions of a "before vs after" sentence for the headline,
in this exact format:
"By [ACTION], I helped [TEAM/ORG] move [METRIC] from [BEFORE] to
[AFTER] in [TIME WINDOW]."
Constraints:
- All three versions must include a primary metric, a number, and a
time window.
- Version 1: business-style, suitable for a resume.
- Version 2: first-person, suitable for a portfolio headline.
- Version 3: tweet-length, under 200 characters.Example output:
- By pausing a single underperforming paid-social cohort, I helped the growth team reduce monthly churn from 6.8% to 4.4% in 90 days. (resume)
- I helped our growth team cut monthly churn 35% by pausing one paid-social cohort in 48 hours. (portfolio headline)
- I cut monthly churn 35% in 90 days by pausing one paid-social cohort. (tweet, 65 chars)
Pro tip: Resume bullets should start with an action verb (“Built,” “Led,” “Identified,” “Cut”) and end with a number. Always.
Prompt 25 - The stakeholder reaction quote
Purpose: Add a single sentence that shows the human reaction to your work. This is the part AI can’t generate, because it requires a real human’s words.
For this data case study:
[PASTE THE IMPACT SECTION]
Write 3 versions of a one-sentence stakeholder reaction that I could
have plausibly received after presenting the analysis. Each version
should be:
- In the stakeholder's voice, not mine.
- Specific to the metric that moved, not generic praise.
- Under 25 words.
If I have a real quote from a real person, paste it in [BRACKETS]
and write the synthesized versions around it.Example output:
- “This is the first time I’ve seen the cohort breakdown - we’re pausing the underperforming paid-social campaigns today.” (VP of Marketing)
- “Your backtest against Q1 convinced me. We were about to cut the entire growth budget.” (Head of Finance)
- “Best 8 days of analysis we’ve gotten this year. Can you do the same thing for retention next quarter?” (CEO)
Pro tip: If you have a real quote from a real stakeholder, use it verbatim and tag the person with their permission. Real quotes beat synthesized ones every time.
Prompt 26 - The “still standing 6 months later” check
Purpose: Add credibility by explaining whether the result held. A case study that doesn’t say this reads as a one-time win.
For this data case study:
[PASTE THE IMPACT SECTION]
Write a 90-word "6 months later" section that covers:
1. Whether the metric held, slipped, or improved.
2. One reason for the result.
3. One thing I would monitor going forward.
4. One place where the original recommendation broke down.
If I don't have 6 months of data, write the section anyway for
"30 days later" or "next quarter" and be honest about the time window.Example output: Six months later, monthly churn was at 4.6% - a small regression from the 4.4% low but still well below the 6.8% peak. The reason: we relaunched paid social in July at a smaller budget, which reintroduced about 0.2 points of churn. The thing I’d watch is the “too many emails” cancellation-reason survey, which ticked up in October and is a leading indicator. The original recommendation broke down when we tried to apply the same cohort cut to the annual plan, where the dynamics were different and the result was noise.
Pro tip: “The original recommendation broke down when…” is one of the most credible sentences in any case study. It signals that you kept thinking about the work after the deck shipped.
Prompt 27 - The lessons learned trio
Purpose: Compress the case study into three lessons a peer could apply. Useful for the closing section and for blog syndication.
For this data case study:
[PASTE EVERYTHING ABOVE]
Write 3 numbered lessons learned. Each lesson must:
- Be under 25 words.
- Start with a verb.
- Be applicable to a different project, not just this one.
- Be honest, including a failure or surprise if relevant.
After the 3 lessons, write a single 1-sentence "If I had to do it
all over again" line that captures the entire case study's spirit.Example output:
- Trust the cancellation date, not the last-active date - billing systems reflect decisions, not behavior.
- Backtest before you present - 2 hours of backtesting saved me from a 30% error in the board deck.
- Stakeholders trust small charts more than dashboards - one chart, one number, one decision.
If I had to do it all over again, I would have run the cancellation-reason survey on day 1, not day 6.
Pro tip: These three lessons are also a great “What I write about” social-media post series. Each one can become a 200-word LinkedIn post on its own.
SECTION 6: Writing & packaging prompts (28–32)
Writing and packaging is the section that turns a draft into a published case study. These prompts handle the readme, the title, the meta description, the LinkedIn post, and the email pitch.
Prompt 28 - The case study readme
Purpose: Generate a tight 150-word opening readme that you can paste at the top of your GitHub repo, Notion page, or Substack post. It should make the project skim-readable in 30 seconds.
For this data case study:
[PASTE EVERYTHING ABOVE]
Write a 150-word case study readme with this structure:
- Sentence 1: project name and one-line tagline.
- Sentence 2: the business question.
- Sentence 3: the data and tools.
- Sentence 4: the key finding.
- Sentence 5: the impact.
- Sentence 6: a link to the hero chart and a link to the code.
- Sentence 7: a one-line "what you'd hire me to do again."
Tone: confident, plain, third person for the first 6 sentences,
first person for the last sentence. Use contractions.Example output: Churn Post-Mortem: How a $40K/mo DTC Coffee Co Cut Subscriber Loss 35% in 90 Days. The VP of Marketing asked why monthly churn jumped 2.7 points between March and May 2025. Using Snowflake, dbt, and a cohort waterfall in Tableau, I found that a single paid-social cohort was driving 38% of the bleed. The team paused the cohort within 48 hours; 90-day churn dropped to 4.4%. Annualized impact: ~$180K. Hero chart · GitHub repo · Tableau Public. You should hire me to do this kind of work again.
Pro tip: The last sentence is the only first-person sentence. Use it. It signals confidence without sounding braggy.
Prompt 29 - The case study title
Purpose: Generate a title for the case study that’s specific, includes a number, and would make you click in a busy Slack channel. This is the title that goes on your portfolio, your Substack, and your LinkedIn post.
For this data case study:
[PASTE THE 6-SENTENCE SPINE FROM PROMPT 1]
Generate 5 title options. For each, give:
- The title.
- A 1-sentence rationale.
- A score from 1 to 10 on: specificity, action, and click-worthiness.
Rules:
- Every title must include at least one number (percentage, dollar,
count, or time window).
- Every title should fit in 80 characters.
- Avoid clickbait. No "you won't believe" or "the truth about."
- The 5 titles should be diverse in style: one technical, one narrative,
one punchy, one analytical, one question.Example output:
- How I cut monthly churn 35% with one Tableau cohort waterfall. (narrative, specificity 9, action 8, click 8)
- The 90-day churn fix that saved $180K in subscription revenue. (punchy, specificity 9, action 7, click 9)
- A cohort decomposition approach to DTC subscription churn. (analytical, specificity 7, action 5, click 6)
- Can a 5-day SQL audit cut churn 35%? Yes. (question, specificity 8, action 8, click 7)
- A reproducible churn waterfall in SQL, dbt, and Tableau. (technical, specificity 9, action 6, click 7)
Pro tip: Use the highest-scoring title as your case study headline. Use the second-highest as your LinkedIn post. Use the question-style as your Twitter hook.
Prompt 30 - The meta description and tags
Purpose: Get the SEO meta description, the 5 tags, and the 1-line elevator pitch for a portfolio page. This is what shows up in Google search and on your portfolio’s project card.
For this data case study:
[PASTE EVERYTHING ABOVE]
Write:
1. A meta description, 150–160 characters, that includes the primary
keyword "ChatGPT prompts for data analyst case study" once and
one concrete number. Do not keyword-stuff.
2. Five tags, in lowercase, hyphenated, suitable for a portfolio CMS.
3. A 1-line elevator pitch (under 200 characters) suitable for a
portfolio project card.
For each, write 1 sentence on why it works.Example output:
- Meta: 32 ChatGPT prompts for data analyst case study writing: turn old SQL, Tableau, and dbt projects into 8-minute portfolio stories that get interviews. (158 characters - primary keyword once, one concrete number, no stuffing)
- Tags:
churn-analysis,cohort-decomposition,dbt-modeling,tableau-waterfall,dtc-subscription - Elevator pitch: Cut monthly churn 35% in 90 days using a cohort waterfall in SQL, dbt, and Tableau. (66 characters)
Pro tip: Most portfolio CMSs (Notion, Substack, Ghost, Hex) have a “description” or “excerpt” field that uses the meta description. Paste it in verbatim.
Prompt 31 - The LinkedIn post
Purpose: Turn the case study into a LinkedIn post that gets engagement without sounding thirsty. Three-paragraph structure, ~150 words, with a single chart attached.
For this data case study:
[PASTE EVERYTHING ABOVE]
Write a LinkedIn post with this structure:
- Line 1: a one-line hook that includes a number and a verb.
- Line 2: a blank line.
- Paragraph 1: the project, in 2 sentences, no jargon.
- Line break.
- Paragraph 2: the surprising finding, in 2 sentences, specific.
- Line break.
- Paragraph 3: the lesson, in 1 sentence, applicable to peers.
- Line break.
- Final line: a call-to-action linking to the case study, with a
1-sentence teaser for what's in the link.
Rules:
- Total length 130–170 words.
- No hashtags except 2 at the very end.
- No "I'm excited to share" or "I was thrilled to."
- Write in first person, contractions on.Example output:
I cut monthly churn 35% in 90 days by pausing one paid-social cohort. (1 line hook with a number)
Last quarter, our churn jumped from 4.1% to 6.8% in 60 days. The VP of Marketing asked why, and the easy answer was “the market.” It wasn’t. A cohort decomposition in SQL and dbt showed that 38% of the bleed came from a single paid-social campaign we launched in March. We paused it in 48 hours. By August, churn was back to 4.4%.
The lesson: when churn jumps, look at the cohort grid before you blame the macro. Most spikes are specific, local, and fixable in a week.
Full case study with the SQL, the dbt model, and the Tableau workbook in the comments. (link)
#dataanalytics #casestudy
Pro tip: Post on Tuesday or Wednesday between 9 and 11am in your target market’s time zone. Engagement drops 40% on weekends.
Prompt 32 - The cold pitch email
Purpose: Generate the cold pitch email you send to a recruiter or hiring manager when you want the case study to do the talking. Short, specific, and respectful of their time.
For this data case study:
[PASTE EVERYTHING ABOVE]
Write a cold pitch email with these exact elements:
- Subject line: under 60 characters, specific, includes a number.
- Line 1: greeting and one-line context ("I'm a data analyst with
4 years of experience at DTC companies").
- Line 2: the project headline in 12 words or fewer, with a number.
- Line 3: the link to the case study.
- Line 4: one specific reason you're reaching out to this person or
company, in 1 sentence.
- Line 5: a low-friction ask, e.g., "Open to a 15-minute intro?"
- Sign-off: 1 line, no headshot, no quote.
Total email body length: under 90 words. No attachments.Example output:
Subject: A churn case study that cut revenue bleed 35% in 90 days
Hi [Name],
I’m a data analyst with 4 years of experience at DTC subscription companies.
I cut monthly churn 35% in 90 days by pausing a single underperforming paid-social cohort.
[Link to case study]
I’m reaching out because your team just opened a senior analyst role focused on retention, and the cohort decomposition approach in my case study maps directly to the work in that job description.
Open to a 15-minute intro next week?
Thanks, [Your name]
Pro tip: Send this email to one specific person, not a generic “careers@” inbox. Reply rates double when the subject line names a project and a number.
Comparison table: 32 prompts by section, goal, and output
This is the table I wish I’d had when I started. It maps every prompt to the case study section it serves, the kind of output it produces, and the 2026 tool it pairs with.
| # | Prompt name | Case study section | Output type | Best paired tool | Approx. time to run |
|---|---|---|---|---|---|
| 1 | The 6-sentence story spine | Opening | Paragraph | ChatGPT, Claude | 8 min |
| 2 | The “before and after” anchor | Opening | One sentence | ChatGPT | 3 min |
| 3 | The stakeholder map | Context | Table | ChatGPT + Notion | 12 min |
| 4 | The “what could have gone wrong” beat | Context | Paragraph | ChatGPT | 8 min |
| 5 | The “so what in one breath” elevator pitch | Opening | 3 sentences | ChatGPT | 5 min |
| 6 | The tool stack summary | Method | Table | ChatGPT | 6 min |
| 7 | The method in 3 sentences | Method | Paragraph | ChatGPT | 5 min |
| 8 | The “this is not a textbook” honesty beat | Method | Paragraph | ChatGPT | 6 min |
| 9 | The decision tree | Method | ASCII diagram | ChatGPT + Excalidraw | 8 min |
| 10 | The validation step | Method | Paragraph | ChatGPT | 6 min |
| 11 | The “what I’d do differently” reflection | Method | 3 bullets | ChatGPT | 6 min |
| 12 | The data source map | Data | Table | ChatGPT + Notion | 8 min |
| 13 | The “cleaning is analysis” paragraph | Data | Paragraph | ChatGPT | 6 min |
| 14 | The named SQL technique | Data | Code block | ChatGPT + Mode/Hex | 12 min |
| 15 | The “if I had to do it again” data model | Data | ASCII diagram | ChatGPT + dbdiagram.io | 10 min |
| 16 | The “I almost got fooled” data caveat | Data | Paragraph | ChatGPT | 5 min |
| 17 | The data glossary sidebar | Data | Table | ChatGPT | 6 min |
| 18 | The hero chart decision | Visualization | Recommendation | ChatGPT + Tableau | 10 min |
| 19 | The chart title rewrite | Visualization | 3 titles | ChatGPT | 4 min |
| 20 | The “I removed this” anti-junk beat | Visualization | Paragraph | ChatGPT | 5 min |
| 21 | The accessibility pass | Visualization | Checklist | ChatGPT | 5 min |
| 22 | The Tableau / Power BI / Looker setup note | Visualization | Instructions | ChatGPT | 6 min |
| 23 | The quantified impact beat | Impact | Paragraph | ChatGPT | 6 min |
| 24 | The “before vs after” sentence | Impact | 3 sentences | ChatGPT | 4 min |
| 25 | The stakeholder reaction quote | Impact | 3 quotes | ChatGPT | 5 min |
| 26 | The “still standing 6 months later” check | Impact | Paragraph | ChatGPT | 6 min |
| 27 | The lessons learned trio | Impact | 3 bullets | ChatGPT | 5 min |
| 28 | The case study readme | Writing | Paragraph | ChatGPT + Notion | 6 min |
| 29 | The case study title | Writing | 5 titles | ChatGPT | 5 min |
| 30 | The meta description and tags | Writing | Strings | ChatGPT | 4 min |
| 31 | The LinkedIn post | Writing | Post | ChatGPT + LinkedIn | 8 min |
| 32 | The cold pitch email | Writing | ChatGPT + Gmail | 6 min |
How to use the table: Start with row 1, then row 2, then row 18, then row 23, then row 28. That order - Story, Chart, Impact, Readme - gives you a publishable first draft in 35 minutes. Fill in the rest over the next day or two.
People Also Ask: FAQ for data analyst case study writing
1. What is a data analyst case study?
A data analyst case study is a short public writeup of one analytical project that walks a stranger through the business question, the method, the data, the visualization, the result, and the impact. It is the single best portfolio asset for a data analyst because it shows the entire workflow in one place, not just a chart or a SQL snippet.
2. How long should a data case study be?
Most strong case studies run 1,200 to 2,500 words, plus charts. A 600-word teaser with one hero chart beats a 5,000-word essay every time. If you can’t tell the story in 1,500 words, you don’t yet understand it well enough to publish.
3. What tools should I mention in a data case study?
Name the tools the reader would expect: SQL (or your dialect), your data warehouse (Snowflake, BigQuery, Redshift, Databricks), your transformation tool (dbt, Airflow, Spark), your analysis tool (Mode, Hex, Jupyter, Observable), and your visualization tool (Tableau, Power BI, Looker). If you used an LLM (ChatGPT, Claude, Cursor, Copilot), say so in the AI assistance line.
4. How do I quantify impact if the metric is soft?
Use proxies. If you improved a “customer experience” project, use the closest measurable outcome: reduction in support tickets, increase in NPS, decrease in time-to-resolution. If you can’t tie your work to a business metric, run a small before-and-after A/B test to create one. The willingness to invent a metric is itself a senior signal.
5. Should I publish case studies on my own site or on a platform like Substack, Notion, or Hex?
Yes to all three. The portfolio is the canonical version (your own domain). Substack is the distribution version (subscribers, email). Notion is the lightweight embeddable version (for recruiters who want a quick read). Hex or Observable is the technical version (for peer analysts who want the runnable code). Each platform has a different audience, and the case study only has to be re-skinned, not rewritten, to fit.
6. Can I publish a case study about confidential work?
Yes, but with care. Replace company names with role/industry descriptions (“a 40-person DTC subscription company” instead of “Blue Bottle Coffee”). Round dollar figures (”~$180K” instead of “$184,213”). Drop the names of internal tools, dashboards, or stakeholders unless you have explicit permission. Most case studies read better when they’re anonymized anyway, because the focus stays on the work.
7. How do I handle a case study where the project failed?
Publish it. The format is “the question, the method, the finding, the action, the result” - and the result can be “we did not move the metric.” Failure case studies are scarcer and more credible than success ones. Just be specific about what you learned, and be specific about what the next iteration would do differently.
8. What’s the difference between a case study and a blog post?
A case study is a portfolio asset, designed to be skim-read in 8 minutes by a hiring manager. A blog post is a thought-leadership asset, designed to be read in 12 minutes by a peer or future customer. The case study wins the interview; the blog post wins the inbound lead. Both are worth writing, but they serve different jobs.
9. How do I know if my case study is good enough to publish?
The 8-minute test. Hand it to a peer analyst and ask them to read it once, then explain the project back to you in their own words. If they can summarize the question, the action, and the result in 30 seconds, it’s good. If they ask “so what did you actually do?” or “what was the number?”, rewrite those two sections and try again.
10. Do ChatGPT prompts actually help, or do they generate slop?
They help if you use them as a thinking partner, not a ghostwriter. The prompts in this article are designed to extract context from you, not to make up context. If you paste empty prompts, you’ll get empty slop. If you paste your real notes, your real numbers, and your real stakeholders, the output will read like you wrote it on a good day. The 20% that ChatGPT can’t do - the judgment, the metric, the stakeholder quote - is exactly the part hiring managers are looking for.
A 30-day “3 case studies” sprint
Here’s the plan I recommend. It assumes you have one project fully drafted and two more in your head, and you can give it about 5 hours per week.
Week 1: Project 1, draft mode (5 hours)
- Day 1 (1 hour): run prompts 1, 2, 5 to lock the story spine and the headline.
- Day 2 (1 hour): run prompts 6, 7, 9, 14 to lock the method and the SQL.
- Day 3 (1 hour): run prompts 18, 19, 20 to lock the hero chart and the title.
- Day 4 (1 hour): run prompts 23, 24, 25 to lock the impact section.
- Day 5 (1 hour): run prompts 28, 29, 30, 31 to write the readme, the LinkedIn post, and the meta description.
Week 2: Project 1, polish mode (5 hours)
- Days 6–8: rewrite the weak sections by hand. Run prompts 4, 8, 10, 11, 16, 26, 27 to fill in the missing beats.
- Days 9–10: get one peer review and one stakeholder review. Rewrite again.
Week 3: Project 2, draft + polish (5 hours)
- Repeat week 1 with a different project, using only the prompts that fit. Most analysts find that project 2 takes 30% less time than project 1 because the workflow is now muscle memory.
Week 4: Project 3, draft + polish + publish (5 hours)
- Repeat week 1 again. Add a publish step on day 5: post on LinkedIn, embed in your portfolio, send the cold pitch email to one specific person.
By the end of 30 days, you have 3 published case studies, a refreshed LinkedIn profile, and 3 cold pitch emails ready to send. The compounding effect is real: each case study makes the next one easier to write and easier to send.
Common mistakes to avoid
I’ve read hundreds of data case studies, written probably fifty, and coached analysts through the rest. The same five mistakes show up over and over. Here’s the list, ranked by how often I see them.
Mistake 1: Leading with tools, not stakes
The first sentence of a case study should never be “I used dbt and Snowflake to…” The first sentence should be “We were bleeding $180K a year and nobody knew why.” Tools show up later, in the method section. Stakes show up first, in the story section. If a reader can’t tell why your project mattered by the end of the first paragraph, the tools in the second paragraph won’t save it.
Mistake 2: Burying the result in a dashboard
A dashboard is a tool for ongoing monitoring. A case study is a tool for one-time storytelling. If your case study links to a 12-tab Tableau dashboard as the “main visual,” you’ve made a dashboard tour, not a case study. Pick one chart. Print it. Tape it to the wall. That chart is the case study.
Mistake 3: Forgetting to name a human
Hiring managers want to see that you can work with people who don’t think in SQL. If your case study has zero stakeholder names, zero pushback, and zero reactions from non-analysts, it reads like a school project, not a work project. Prompt 3 (the stakeholder map) and prompt 25 (the stakeholder reaction quote) are the two prompts that fix this. Use them.
Mistake 4: Skipping the impact section
I get it. Sometimes you can’t quantify the win. Sometimes the project is “I built this dashboard and people use it.” Sometimes the metric moved but you can’t isolate your contribution. Write the section anyway. Be honest about the uncertainty. The willingness to say “I think it was X, but I don’t have a clean attribution” is more credible than a confident made-up number.
Mistake 5: Publishing once and never updating
A case study is a living document. The 6-month check (prompt 26) is the single best reason to come back to a case study and add a “what happened next” beat. Hiring managers notice when your work is dated, and they notice even more when your work is current. Set a calendar reminder to revisit each case study 6 months after publication.
Final word: your past projects are a goldmine
You probably have 3 to 5 projects sitting in private repos, private dashboards, or memory that are case-study-ready right now. The bottleneck is not the project. The bottleneck is the writing - and that’s exactly what these 32 ChatGPT prompts for data analyst case study writing are designed to fix.
Pick one project today. Run prompts 1, 2, 5, 18, 23, and 28. You’ll have a 600-word first draft by the end of the hour. Spend the rest of the week filling in the rest. By next Friday, you’ll have a portfolio asset that you’ll be quoting in interviews for the next five years.
If you want a head start, the storytelling with data library has real before-and-after chart case studies that show what “good” looks like at the chart level. Pair that with the BLS data scientist job outlook page for the demand context, and you’re set.
Now go publish something.