Why it matters

The review meeting stalls on one question.

A district officer points at the slide: "Where does 62% come from?" What happens next decides whether your evidence gets used.

Today

Three spreadsheets and a promise

10:02The analyst opens the master sheet, the cleaned export, then the pivot.
10:09The denominator differs from last month's deck. Nobody knows why.
10:15"We'll send it after." The decision slips a month.

With a Claude dashboard

One click, in the room

10:02The officer clicks the number on the shared dashboard.
10:02Source, formula and the rows behind it appear.
10:04They filter to their region and agree which blocks need support.

01 · See it

Anatomy of a Claude dashboard.

Tap the numbered dots to learn each part, and what to say to Claude to change it.

Learning dashboard · ExampleRefreshSourcesShare

Learning and delivery

Across all districts, 62% of students reached grade level, up 31 points.

Region: All ▾Round: Latest ▾From: 2026-04-01

OutcomesDeliveryData quality

At grade level62.4%

Change+31.2

Assessed9,842

BlockLatestChange

Ajnala61.3%+29.8

Khanna66.0%+34.1

02 · Build it

Write the prompt that builds it right the first time.

Good dashboards start from a decision, not a dataset. Fill in the five parts and copy the prompt into the chat beside a new dashboard. The fields start with an example.

Your prompt

03 · Feed it

Wherever your data lives, there is a way in.

Survey apps, case management tools, forms, spreadsheets, databases, even PDF reports. Pick the tool you use to see the best route, the steps and what to say to Claude.

Five routes, from simplest to most automated

1 · Export and attach

Works with any tool

Download a CSV, Excel or JSON export and attach it. Up to 8 MB per file; Excel sheets save as CSV first.

Best for survey rounds, one-off analyses, any tool without a connector

Build from the attached export

2 · Connect live

Always current

Connect the app in Claude, such as Google Drive or a data warehouse. A query runs each time someone opens, refreshes or filters.

Best for data that changes weekly or daily

Use the responses sheet in my Google Drive

3 · Bridge through a sheet

Live data from tools without a connector

Let your tool sync into a Google Sheet or a database, using its built-in integration or an automation tool. Then connect that live.

Best for field data that keeps arriving

Use the synced Kobo sheet in Drive

4 · Saved and scheduled

A file kept current for everyone

Claude re-runs a query on a schedule and saves the rows, so readers without access still see the numbers.

Best for weekly reports from restricted systems

Refresh this every Monday at 9

5 · Paste, extract or fetch

Data that isn't a dataset yet

Paste a table into chat, attach a PDF report or screenshot of a table, or ask Claude to find a public dataset. Claude turns it into a clean file first.

Best for annual reports, government portals, quick numbers

Pull the tables from this PDF into a dataset

Connectors available to you depend on your Claude plan and your organisation's settings. Any tool can always use route 1.

04 · Shape it

Every feature, and exactly what to say.

Six families cover everything a dashboard can do. Pick one to see its features.

Recipes

Twelve prompts, ready to send.

Highlighted words are yours to replace.

Patterns

Pick the shape by who reads it, and how often.

Tap a pattern to see who it serves and the prompt that builds it.

Tell the story Report progress Diagnose Run the programme

← Read once or at period endWatched week after week →

Top: external audience · Bottom: internal team

05 · Fix it

When something looks wrong.

Most problems have one of six causes. Open the one you see.

A source says it didn't load

Usually: the reader hasn't connected the app the data lives in, declined access, or lacks permission to the table.

Fix: the reader follows the connect prompt on the page. If it still fails, ask Claude, who re-runs the source and names who can grant access.

The attendance source didn't load for Priya. What's stopping it?

A number looks wrong

Usually: a definition. What counts as "active" or "reached" in the formula differs from what you expected.

Fix: click the number, open its formula in Sources, and tell Claude what it should count.

Change should only count people measured in both rounds

The data looks old

Usually: an attached file or a saved query that hasn't been refreshed. Sources shows when each source last updated.

Fix: attach the new file, switch to a live source, or set a schedule.

Replace the survey file with this new export

A filter changes some charts but not others

Usually: those charts read a source that doesn't use the filter.

Fix: ask for the filter to apply everywhere. A saved query can't follow a filter, so that source must stay live.

Make the region filter apply to every chart and table

It looks cramped beside the chat

Usually: too many tiles in a row, or wide tables, at about 400 px wide.

Fix: ask for fewer headline figures and a layout that stacks when narrow.

Keep four headline tiles and make it read well at phone width

Someone I shared it with sees nothing

Usually: they have no Claude account, the dashboard is still private, or the link was shared to a different account.

Fix: check the Share setting, and ask them to sign in with the email you shared to. A free account is enough.

06 · Share it

Viewing is free. The person asking pays.

Every viewer needs a Claude account, and a free one is enough. AI features draw on the asker's own plan, never the owner's.

Pro or Max planShare as "only you" or "anyone with the link". Good for sample or aggregated data.

Team or EnterpriseAdd named people or groups with view or edit access, plus admin controls.

A public audiencePoint a branded address at the link, or publish a frozen snapshot on your own site.

Protect

The finer the data, the tighter the door.

Private

Person-level rows

Names, IDs and records per person stay in your secure system.

Team

Site or team totals

Small groups suppressed. Shared with named access.

Public

Region or national totals

Vetted headline figures for a public snapshot.

Finer detail, fewer peopleCoarser totals, wider audience

Strip names, phone numbers, national ID numbers and exact addresses

Suppress groups smaller than 10 people

Check your MoU or donor data terms first

Label sample data on the page, every time

No chat box inside the dashboard

Do this: route questions through comments, and give frequent explorers a companion "Ask the data" page.

It shows data, it doesn't collect it

Do this: collect in your survey tool or a sheet, and let the dashboard read from there.

No maps

Do this: rank regions in bars or a sortable table, which compare better anyway.

07 · Master it

Three levels to the fullest use.

Most teams stop at level one. The real value sits in levels two and three.

Level 1 · Builder

A dashboard that answers

Built from one attached file
Four headline tiles and two charts
One filter
Every figure checked in Sources

You've arrived when you trust every number

Level 2 · Practitioner

A dashboard people use

Live data from a connected app
Tabs and deep links per audience
Click-to-drill and date ranges
Named calculations with plain definitions
Comments routed to Claude

You've arrived when the team asks questions on it

Level 3 · Evidence leader

A system partners rely on

Scheduled refresh with stale warnings
A data quality tab during fieldwork
Scorecard and Play mode for partner reviews
A scroll story for the annual report
Access reviewed each quarter
A public snapshot of vetted totals

You've arrived when the meeting opens the dashboard first

Cheat sheet

Twenty things to say to your dashboard.

Layout

Split this into Outcomes and Delivery tabs
Lead with the latest figure
Keep four tiles in the top row
Write a one-line takeaway under the title

Charts

Swap this to a line chart
Add a 60% target line
Show first to latest round as a slope chart
Label lines at their ends, no legend

Filters

Add a region dropdown
Add a date range, default last 90 days
Clicking a bar filters everything
Add a Reset button

Data

Replace the file with this new export
Use the live table instead
Define each metric on the page
Add a gender split to every chart

Present

Add a Play button by round
Make filter changes glide
Add a scroll story tab
Make a scorecard for the funder

Before you share

Ready to launch?

0%

Your ticks are saved in this browser only.

Questions people ask

Quick answers.

Do I need to know code or SQL?

No. You describe what you want and Claude writes the queries, formulas and page. Editors can see and edit the queries in Sources if they want to.

How is this different from Power BI or Looker Studio?

You build by conversation instead of drag and drop, and every reader can see the formula and rows behind a number, not only the author.

Can readers download the data?

Yes. Any source's rows download as CSV from Sources, so aggregate and de-identify before you attach a file.

Can I turn it into a PDF?

Not from inside the dashboard. For a printable report, ask Claude for a document or a slide deck from the same data.

Who can change it?

Only people with edit access. Viewers can read, filter and comment.

What if my data is in Excel?

Export the sheet as CSV and attach it, or keep it in Google Drive and connect Drive for live updates.

About the author

Shashidhar S A

Founder, HubForge Global · MERL & AI Innovator

En-Intrapreneur Driving evaluation ecosystems through People, Process & Product Co-Chair, IOCE Comms & Storytelling Co-Lead, EvalYouth Global Champion Engagement

A monitoring, evaluation and learning practitioner with about a decade of work across India, from grassroots field work to state-level government partnerships and MEL systems design. He wrote this playbook so that any team, in any field, can build evidence that people check, trust and use.

Connect on LinkedIn Visit HubForge

Claude DashboardPlaybook