Walkthrough · No. 03

CSR Insights: A domain application on a general engine

How a decade of CSR filings from 45,000+ companies became an application anyone can question in plain language: a Dataful dataset plus GoPie.

CSR Insights AI-generated dashboard charts: top states by total CSR expenditure and CSR spending by sector.
Fig. 1 A dashboard generated in CSR Insights from real filings. Factly product

§ 01 Problem

The problem

Larger Indian companies must spend 2 per cent of profit on CSR and report it. The filings are public, but “who funds health projects in Hyderabad, and how much?” meant downloading, cleaning and joining tables by hand.

§ 02 Information and data

What made it hard

  1. Company identity drift: identifiers change on listing, names vary, mergers need a successor.

  2. District names as written by companies versus official Local Government Directory codes.

  3. Projects tagged “PAN India” or with no location, which must not be mistaken for a state.

  4. The ministry’s own caveat that portal data is not guaranteed complete.

§ 03 Workflow and design

The decisions that shaped it

  • D1

    Standardise once, in Dataful

    Documented rules for listing changes, entity names, mergers and district codes, applied once.

  • D2

    Reuse the engine

    GoPie provides the agent, dashboards and reports; CSR Insights adds the dataset and the interface.

  • D3

    Three guided modes, not a blank chat

    Chat with data, generate a dashboard, generate a report, each with example prompts.

  • D4

    Publish the data journey

    An About the data page explains the sources and the standardisation.

  • D5

    Re-skin for the next domain

    BRSR Insights applies the pattern to sustainability disclosures, adding the filings so answers cite the PDF.

§ 04 Shipped, not planned

What it actually does

  • Plain-language questions over CSR filings from 45,000+ companies, 2014-15 to 2024-25, SQL shown
  • Dashboards from a prompt: KPI tiles, rankings and charts by state, sector, company and year
  • Narrative reports with charts, tables and executive summary, frozen and exportable

§ 05 Input → steps → output

Walkthrough

  1. Input

    Dataful’s standardised CSR master dataset, built from company disclosures on the National CSR Portal.

  2. Step 1: Choose a mode

    Chat with the data, generate a dashboard or generate a report, each with example prompts.

    CSR Insights home page: "Turn CSR data into insights" with chat, dashboard and report modes.
    Fig. 2 Three guided modes.
  3. Step 2: Ask and see the query

    “What was the CSR expenditure in each state of Northeast India in 2024-25?” The agent lists the states, SQL visible.

    CSR Insights chat on CSR spending in Northeast Indian states, showing the agent step trace, the generated SQL and the answer list.
    Fig. 3 Question, steps, SQL, answer.
  4. Step 3: Generate a dashboard

    One prompt produces KPI tiles, the top states and spending by sector, refinable by chat.

    CSR Insights AI-generated dashboard charts: top states by total CSR expenditure and CSR spending by sector.
    Fig. 4 The dashboard.
  5. Step 4: Generate a report, then review it

    A state-wise report for 2024-25: summary, charts, tables, frozen for review.

    CSR Insights report "State-wise CSR Spending Report for 2024-25": the snapshot header with its generation date, Complete and Edited badges, and an Export PDF button.
    Fig. 5 A frozen report, ready for review.
  6. Step 5: See the pattern repeat

    BRSR Insights runs the same engine on sustainability disclosures and cites the exact PDF.

    BRSR Insights chat answer with superscript citations and a "Sources cited" list of the company BRSR filing PDFs it drew on.
    Fig. 6 BRSR Insights: the adjacent application, with document citations.
  7. Output

    A cross-year, cross-state question answered in one session, with the query visible and the data documented.

§ 06 Evaluation

How we check it

  • The standardisation rules are documented, so users can judge what changed and why.

  • Source caveats are reproduced, not softened.

  • Report narratives are reviewed by a person before use.

§ 07 Evidence

The figures

  • 45,000+[1]

    companies covered

  • 2014-15 to 2024-25[1]

    financial years of filings

  • Top 1,000[2]

    listed companies in the BRSR source data

§ 08 Limitations and lessons

Limitations and lessons

read this part first

Limitations

  • The source data carries the ministry’s own caveats on completeness; CSR Insights shows them.
  • A generated dashboard can present “Pan India” as a state; review is part of the workflow.
  • Some BRSR filings disagree between PDF and XML; Dataful documents the resolution.

Lessons from operation

  1. A domain application is mostly data work and framing; the engine was already built.
  2. Standardise upstream. Entity reconciliation done once made every answer better.
  3. Guided modes with example prompts beat a blank chat box.

Sources (6)

  1. [1] CSR Insights: about the data · csr.dataful.in (opens external site)
  2. [2] Dataful: BRSR dataset page · dataful.in (opens external site)
  3. [3] CSR Insights: home · csr.dataful.in (opens external site)
  4. [4] BRSR Insights: home · brsr.dataful.in (opens external site)
  5. [5] Dataful: home · dataful.in (opens external site)
  6. [6] GoPie: home · gopie.ai (opens external site)

Sources reviewed 15 November 2026.

Discuss a project

Have a domain that deserves its own application?

Tell us about the data, the audience and the questions they ask.