Purpose
A spreadsheet on its own does not tell a story. This SOP takes a raw CSV export, whether it is LinkedIn analytics, sales data, ecommerce orders, or support tickets, and turns it into an interactive dashboard artifact inside Claude. The dashboard is built around the actual questions the data can answer, not a generic chart for every column.
Who this is for
Anyone who has a CSV export and needs a client-ready or internal analytics view without manually building charts in a spreadsheet tool.
Tools needed
- A Claude account with the Artifacts feature available
- A CSV export of the spreadsheet to analyze
- The master dashboard prompt included below
01
Prepare the spreadsheet
A clean export saves back and forth later.
- a.Confirm the column headers sit in the first row.
- b.Remove unrelated tabs, notes, or summary rows that are not part of the actual dataset.
- c.Export the sheet as a single CSV file.
02
Open Claude with artifacts on
Start a new conversation and confirm Artifacts is enabled, since this is what renders the dashboard as an interactive view rather than plain text.
03
Attach the CSV
Upload the CSV file into the conversation before sending any instructions, so Claude has the actual data to inspect rather than a description of it.
04
Paste the master prompt
Copy the full master prompt from the resource section further down this page and send it in the same message as the attached CSV.
Tip
Use the Copy button on the prompt block so nothing gets altered by accident while pasting.
05
Let Claude analyze and build
Claude inspects every column, works out what the data represents, decides the dashboard story, and generates the interactive artifact in one pass. This step needs no input, just time to run.
06
Review against the checklist
Before accepting the dashboard, check it against what the prompt promised.
- a.Do the KPI cards at the top match real totals from the sheet.
- b.Do the charts respond to hover and, where relevant, to clicks and filters.
- c.Do the insight cards use exact numbers rather than vague statements.
- d.Does the raw data table let you search, sort, and scroll through the source rows.
07
Request refinements
If something is off, name the exact section and the exact change needed instead of asking for a full rebuild.
- a.Point to the specific chart, KPI, or tab.
- b.State what is wrong or missing.
- c.Let Claude update just that part of the artifact.
08
Share the artifact
Once the dashboard is approved, share the artifact link directly or hand it off to whoever requested the analysis.
The master prompt
Paste this in full alongside the attached CSV in Step 4. It tells Claude how to read the data, what questions to answer, how to design the dashboard, and what rules to follow so nothing is invented.
Prompt
You are a senior data analyst, dashboard designer, and frontend engineer.
I am going to give you a CSV file.
Your job is to transform the CSV into a polished, useful, interactive analytics dashboard artifact.
Do NOT simply visualize every column.
First understand the dataset, determine what the data represents, identify the most useful business questions the data can answer, and then design the dashboard around those insights.
The final result should feel like a professionally designed analytics product, not a generic AI-generated dashboard.
==================================================
1. FIRST: UNDERSTAND THE CSV
==================================================
Before designing anything, inspect the full CSV and determine:
- what each row represents
- what each column represents
- which columns are:
- dates
- categories
- dimensions
- IDs
- URLs
- text
- counts
- percentages
- currency
- durations
- calculated metrics
- which columns are useful for analysis
- which columns should not be visualized directly
- whether any metrics can be derived from existing fields
Detect relationships between fields.
Examples:
If the CSV contains:
Impressions
Clicks
Conversions
derive:
CTR = Clicks / Impressions
Conversion Rate = Conversions / Clicks
If it contains:
Revenue
Orders
derive:
Average Order Value = Revenue / Orders
If it contains:
Engagements
Impressions
derive:
Engagement Rate = Engagements / Impressions
Do not invent data.
Only calculate metrics that can legitimately be derived from the supplied CSV.
Handle:
- blank cells
- null values
- duplicate rows
- inconsistent capitalization
- numbers stored as strings
- currency symbols
- percentage values
- date formats
- very long text fields
- URLs
Never silently make up missing values.
==================================================
2. IDENTIFY THE DASHBOARD STORY
==================================================
Determine the 4 to 8 most useful questions someone would want this dataset to answer.
For example:
- What is happening overall?
- What is changing over time?
- Which items are performing best?
- Which categories contribute most?
- Where are the outliers?
- Which segments underperform?
- What patterns appear in the data?
- What requires attention?
Use these questions to decide the dashboard structure.
Prioritize useful insights over the number of charts.
The dashboard should communicate:
OVERVIEW
-> TRENDS
-> BREAKDOWNS
-> TOP/BOTTOM PERFORMERS
-> ACTIONABLE INSIGHTS
when those sections make sense for the dataset.
Do not force this exact structure if another structure is more appropriate.
==================================================
3. DASHBOARD LAYOUT
==================================================
Build a responsive dashboard with a premium dark analytics-product aesthetic.
Overall visual direction:
- dark charcoal/navy background
- subtle borders
- slightly lighter cards
- restrained use of accent colors
- excellent spacing
- compact typography
- rounded cards
- subtle shadows
- high information density without clutter
- sophisticated SaaS analytics interface
- similar in feel to Linear, Stripe, Vercel, modern BI tools, and premium analytics products
- NOT a colorful generic Bootstrap dashboard
- NOT excessive gradients
- NOT glassmorphism everywhere
- NOT huge headings
- NOT giant empty cards
- NOT "AI slop"
The dashboard should look especially good when recorded vertically on a monitor for a social-media reel.
Use the available screen width efficiently.
Suggested page structure:
HEADER
Dataset / report title
Short contextual subtitle
Date range if detectable
Row count where useful
GLOBAL CONTROLS
Relevant filters based on the CSV, such as:
- date range
- category
- location
- status
- campaign
- person
- product
- channel
Only add filters that actually make sense for the dataset.
Do not create fake filters.
==================================================
4. KPI CARDS
==================================================
At the top, show approximately 3 to 6 primary KPIs.
Choose them intelligently from the CSV.
Examples could include:
Total Revenue
Total Orders
Average Order Value
Total Impressions
Total Engagements
Engagement Rate
Customers
Conversion Rate
Average Duration
Open Opportunities
Completion Rate
Each KPI card should contain:
- small uppercase or subdued label
- large primary number
- optional secondary context beneath it
- subtle accent
- useful formatting
Examples:
36.1K
1,508
0.97%
$182.4K
12m 34s
Use:
K
M
B
where appropriate.
Do not show meaningless KPIs just to fill space.
==================================================
5. AUTOMATIC CHART SELECTION
==================================================
Choose chart types based on the data.
TIME SERIES
Use line or area charts.
Examples:
- revenue over time
- impressions over time
- leads generated per day
- order volume per month
CATEGORY COMPARISON
Use horizontal or vertical bar charts.
Examples:
- revenue by product
- leads by source
- customers by country
RANKINGS
Use ranked horizontal bars or lists.
Examples:
- top posts
- top campaigns
- top products
- top customers
RELATIONSHIPS
Use scatter plots.
Examples:
- impressions vs engagement
- spend vs conversions
- price vs sales
DISTRIBUTIONS
Use histograms / ranges only when useful.
PART-TO-WHOLE
Use donut/pie charts sparingly and only when they genuinely improve comprehension.
Avoid excessive pie charts.
Do not choose chart types randomly.
==================================================
6. EVERY CHART MUST BE INTERACTIVE
==================================================
Every visual should have useful interaction.
At minimum:
- hover tooltips
- highlighted data points
- highlighted bars
- hover states
- readable exact values
- labels where useful
Where appropriate also provide:
- click-to-select
- click-to-filter
- crosshair on time-series charts
- active states
- linked filtering between visuals
For example:
Click "London" in a location chart
-> update the rest of the dashboard to show London data.
Click a post in a scatter plot
-> highlight the corresponding item in the ranking list.
Hover over a line
-> show:
date
metric value
secondary metric if useful
Interactions should feel intentional and polished.
==================================================
7. TABS / VIEWS
==================================================
If the dataset has several distinct analytical dimensions, create tabs.
Possible examples:
Overview
Performance
Audience
or
Overview
Products
Customers
Regions
or
Overview
Campaigns
Channels
Conversions
Do not mechanically use these labels.
Create tabs based on the actual CSV.
Each tab should answer a different set of questions.
Avoid making the user scroll through 20 unrelated charts on one page.
==================================================
8. TOP / BOTTOM PERFORMERS
==================================================
Where applicable, automatically identify:
- top performers
- bottom performers
- highest growth
- highest volume
- strongest conversion
- largest outlier
- strongest category
Show these using clean ranked lists or bar visualizations.
Do not claim that something is "good" or "bad" unless the data supports the comparison.
Use wording such as:
Highest
Lowest
Largest
Smallest
Fastest growth
Most engagement
==================================================
9. INSIGHT CARDS
==================================================
After analyzing the data, add a small number of data-backed insight cards.
For example:
AUDIENCE INSIGHT
51% of the audience is Director level or above.
CONTENT INSIGHT
Three posts generated 62% of total impressions.
GROWTH INSIGHT
Friday generated the highest average daily lead volume.
SALES INSIGHT
Enterprise opportunities convert 1.8x more frequently than SMB opportunities.
Only generate an insight when it can be calculated from the CSV.
Do not create vague AI commentary such as:
"Engagement seems healthy."
Be specific.
Use exact numbers.
==================================================
10. TABLE / RAW DATA VIEW
==================================================
Provide a clean way to inspect the underlying records.
The table should support, where feasible:
- searching
- sorting
- scrolling
- pagination or row limiting
- filtering
- readable formatting
Do not display extremely long URLs in full.
Use:
View
or truncate them visually while keeping them clickable.
Allow users to inspect the source records behind the charts.
==================================================
11. DATA FORMATTING
==================================================
Format data intelligently.
Examples:
36100
-> 36.1K
0.0097 stored as ratio
-> 0.97%
182400
-> 182.4K
currency:
-> $182,400
Dates:
-> Jun 29
-> Jun 2026
depending on the context.
Do not convert identifiers into numerical metrics.
Do not sum IDs.
Do not average arbitrary numeric columns unless averaging makes analytical sense.
==================================================
12. DATE HANDLING
==================================================
If the CSV contains a date or timestamp column:
detect the earliest and latest dates.
Use the actual dataset period in the dashboard subtitle.
Example:
LinkedIn Analytics
Apr 1 to Jun 29, 2026
Allow date filtering when useful.
If several date columns exist, determine which one is most relevant to the main analysis.
==================================================
13. RESPONSIVE DESIGN
==================================================
Desktop layout should feel like a professional analytics dashboard.
It must also remain usable at:
1440px
1280px
1024px
768px
mobile
On smaller screens:
- stack cards
- preserve chart readability
- keep navigation usable
- prevent horizontal overflow
- collapse grids intelligently
Do not simply shrink everything.
==================================================
14. TECHNICAL REQUIREMENTS
==================================================
Build the entire artifact as a working interactive frontend.
If the artifact environment supports React, use React.
Otherwise create a self-contained HTML/CSS/JavaScript application.
Prefer a SINGLE ARTIFACT / FILE when possible.
Do not require a backend.
Do not require an API.
Do not fetch external data.
The CSV supplied in this conversation is the data source.
Embed or parse the CSV data inside the artifact.
Use a robust charting solution available in the environment.
If no charting library is available, use SVG or Canvas.
Do not leave charts as static mockups.
They must actually render from the CSV data.
==================================================
15. INTERACTION QUALITY
==================================================
Make the interface feel like a real product.
Include subtle:
- hover transitions
- selected states
- chart animations
- tooltip animations
- tab transitions
- button feedback
Keep animations approximately 150 to 300ms.
Do not use excessive bouncing, glowing, spinning, or decorative animation.
Analytics should remain the focus.
==================================================
16. DATA INTEGRITY RULES
==================================================
These rules are mandatory:
DO NOT invent data.
DO NOT fabricate missing dates.
DO NOT create fake benchmarks.
DO NOT manufacture percentage changes.
DO NOT claim causal relationships from correlation.
DO NOT create fake comparison periods.
DO NOT generate random values to make charts look better.
If something cannot be calculated, omit it.
If data quality is poor, surface that clearly.
==================================================
17. AUTOMATIC ADAPTATION
==================================================
The dashboard must adapt to the supplied CSV.
Do NOT assume this is:
LinkedIn data
marketing data
sales data
financial data
customer data
Determine that from the CSV itself.
For example:
A LinkedIn CSV might produce:
Overview
Posts
Audience
A sales CSV might produce:
Overview
Pipeline
Reps
Deals
An ecommerce CSV might produce:
Overview
Revenue
Products
Customers
A support CSV might produce:
Overview
Tickets
Agents
Issues
A project CSV might produce:
Overview
Projects
Status
Team
The structure should emerge from the dataset.
==================================================
18. VISUAL PRIORITY
==================================================
The first screen should immediately answer:
"What is happening?"
Someone should be able to understand the most important information within approximately 5 seconds.
The visual hierarchy should be:
1. primary KPIs
2. dominant trend
3. important breakdowns
4. rankings
5. deeper analysis
Do not give every chart equal visual weight.
==================================================
19. BEFORE BUILDING
==================================================
Internally inspect the dataset first.
Determine:
Dataset type:
Row meaning:
Primary date field:
Main measures:
Main dimensions:
Derived metrics:
Useful filters:
Potential data-quality issues:
Best dashboard sections:
Then build the artifact.
Do NOT display a long planning document to me before building.
Go directly into creating the dashboard once the analysis is complete.
==================================================
20. FINAL QUALITY CHECK
==================================================
Before finishing, verify:
- all metrics come from the CSV
- totals are calculated correctly
- percentages use the correct denominator
- filters actually work
- tabs actually work
- every chart is interactive
- tooltips display meaningful information
- labels are readable
- charts resize correctly
- no chart is empty because of a parsing error
- URLs are handled cleanly
- date ranges are correct
- derived metrics are mathematically correct
- responsive behavior works
- no fake data has been introduced
The result should look like something a founder, marketing lead, sales director, analyst, or executive could genuinely use.
Most importantly:
TURN THE CSV INTO A STORY.
Do not merely turn columns into charts.
At a glance
| Piece | What it does | Comes from |
|---|---|---|
| CSV | The raw source data being analyzed | Your spreadsheet export |
| Master prompt | Tells Claude how to analyze and design the dashboard | Resource block above |
| Dashboard artifact | The finished interactive analytics view | Generated by Claude in Step 5 |
FAQs
Does this work with any spreadsheet, or only certain kinds of data?
It works with any CSV. The master prompt tells Claude to inspect the actual columns first and build the dashboard structure around what the data represents, instead of assuming it is marketing, sales, or financial data in advance.
Do I need to tell Claude what charts to use?
No. The prompt instructs Claude to choose chart types automatically based on the data, such as line charts for trends, bar charts for comparisons, and scatter plots for relationships.
What if my spreadsheet has messy or inconsistent data?
The prompt tells Claude to handle blanks, duplicates, inconsistent capitalization, currency symbols, and mixed date formats, and to surface data quality issues rather than silently inventing values.
Troubleshooting
Warning
If the CSV is attached but the dashboard still looks generic, confirm the full master prompt was sent in the same message as the file, not as a separate follow-up.
If a chart renders empty, the underlying column is likely stored as text instead of a number, or a date format Claude could not parse. Flag the exact column and ask for it to be re-parsed rather than rebuilding the whole dashboard.
If the KPIs look wrong, check that the source sheet did not have duplicate rows or a hidden summary row that got exported along with the data.
| v1.0 | Initial SOP built from the master dashboard prompt | September 2026 |
Want a dashboard built for you?
Send over the spreadsheet and what you need to know from it, and the dashboard gets built end to end.
Book a Call →Build Your First AI Agent
This is exactly the kind of workflow you will use once you start building your own dashboards, agents, and automations from the ground up.
Build My First Agent →