ESG Data Technology — How Cleartraced Works

Our Technology

The infrastructure behind
financial institutions’ ESG data sourcing.

Our data pipeline sources, extracts and structures ESG data from public company information. Our AI does the repetitive work, and human experts review every datapoint before delivery.

Annual reports

Sustainability reports

Public filings

Policies & codes

Company websites

01 · Sources

Everything a company officially publishes.

AI extracts and structures the data

Any company, any country, any language

Extracting from text, tables and graphs.

Unified units and currencies

Confidence score

Anomaly detection

Every figure traced to its page and line

Human in the loop ✓ Verified

Our ESG data experts review and approve every datapoint before it is delivered.

Continuous monitoring

Our platform tracks new or updated disclosures, keeping the data constantly current.

02 · Extraction & Quality

AI structures the data, analysts sign off, monitoring keeps it current.

‹ › ⚲ app.cleartraced.com

Cleartraced’s ESG Data Platform

Platform

Search and compare

API

Integrate into your systems

File export

CSV, XLSX, JSON

03 · Platform

Data is available in our platform, as file exports, or via API.

FINDING COMPANY INFORMATION

Everything a company officially publishes.

EXTRACTING, STRUCTURING & VERIFYING DATA

AI structures the data. ESG data experts sign it off. Agent monitoring keeps data current.

PLATFORM DELIVERY

Delivered via Cleartraced’s platform, offering file export or API integration, whatever fits your infrastructure.

THE CLEARTRACED DATA PIPELINE

From scattered company information to structured, auditable ESG data.

Adding a company

We only need three inputs to run our process: the company’s country, the reporting year, and its name or identifier. New companies can be requested in whichever way is easiest for you: via email, a form within our data platform, an API request, or a file over SFTP, always handled with total privacy and confidentiality.

Request · 3 inputs

Name or identifier

Volvo Cars ABor LEI · ISIN · Ticker

Country

SwedenAny jurisdiction globally

Reporting year

2025Historical data available

Finding public information

Our system scans each company’s official webpages, downloads its annual, sustainability and regulatory reports, and combs the full site for further ESG disclosures such as human-rights policies and codes of conduct, tying every document to the right entity.

SOURCING

Found: Volvo_Annual_Report_2023.pdf

Found: Volvo_Sustainability_Report_2023.pdf

Found: Volvo_Code_of_Conduct.pdf

Found: Human_Rights_Policy.pdf

Found: Supplier_Code_of_Conduct.pdf

Found: Whistleblowing_Policy.pdf

Found: Official webpage (English)

Found: Official webpage (native language)

Process reports

Our system reads PDFs, scans complex tables and charts and converts them into structured, machine-readable text, handling complex layouts and any language.

TEXT

CHART

TABLE

IMAGE

Structured JSON

{
  "text":  "…",
  "table": […],
  "chart": {…},
  "image": "…"
}

Extract & structure

Structured AI prompts extract each datapoint in context, with high precision, and harmonise it to standard units.

As reported (PDF)

Scope 1312,000,000 kg

Scope 286,000,000 kg

Scope 34,120,000,000 kg

RevenueSEK 247.9B

→→→→

Normalised

Scope 1312,000 tCO₂e

Scope 286,000 tCO₂e

Scope 34,120,000 tCO₂e

RevenueEUR 21.5B

Units converted (kg → tCO₂e) and currency harmonised (SEK → EUR)

Normalise & standardise

Every value is mapped into common taxonomies and regulatory frameworks before storage, enabling consistent cross-company comparison.

GHG_scope1 → ESRS E1-6 · EBA P1

energy_total → ESRS E1-5 · GRI 302

board_gender → ESRS G1 · GRI 405

Verification

A rule-based system scores every datapoint across five layers. Anything below the confidence threshold is routed to an ESG analyst; nothing is guessed or dropped.

Automated checksOngoing

Quality score

Statistical range check

YoY consistency

Unit verification

Source-match confidence

LLM cross-validation

Flagged for expert review → resolved

Expert human verificationOngoing

Cleartraced Citations review view on a laptop: extracted ESG datapoints with YoY comparison and confidence score linked to the source disclosure PDF

Store & deliver

Validated data is stored securely and delivered in real time via platform, API or file, each datapoint linked to its source and re-verified against the original PDF.

PlatformFileAPI

Cleartraced platform login screen shown on a laptop

THE TECHNOLOGY ADVANTAGE

What our technology enables that humans and legacy providers simply cannot.

ANY

company · any size

a few hundred

capped by headcount

Human teams can only cover so many names. Our pipeline has no ceiling: any company that discloses, whatever its size or market.

ANY

datapoint you need

fixed set

vendor’s menu only

No fixed list of metrics. If a company discloses it, we extract and structure it, down to the datapoints your analysis needs.

ANY

reporting language

English-first

translation gaps

AI reads disclosures natively in any language. No translation queue, no blind spots for non-English markets.

Always

current, automatically

12–24 mo

legacy vendor lag

We track each company’s release dates and monitor for new disclosures. The moment data is published, the pipeline picks it up.

FAQ

Common questions about our technology

How does Cleartraced find the public information and PDFs? +

Our pipeline automatically discovers and retrieves annual reports, sustainability disclosures, and regulatory filings from company IR pages, stock exchange databases, and regulatory repositories, across any language and jurisdiction. No manual sourcing is required.

What AI models power the extraction? +

We use several large language models, including OpenAI, Claude and Gemini, among others, choosing whichever delivers the best accuracy for each task, combined with custom-trained classification and normalisation layers. The models are continuously evaluated against a ground-truth dataset curated by our in-house ESG analysts.

How is data accuracy verified before delivery? +

Every extracted datapoint passes through a six-stage validation pipeline: document parsing, AI extraction, cross-field consistency checks, unit and boundary validation, statistical outlier detection, and a final human sign-off for low-confidence values. Each datapoint ships with its source citation and confidence score.

Can the pipeline handle non-English reports? +

Yes. Our models are trained on reports across 30+ languages including Mandarin, Japanese, German, French, Spanish, and Nordic languages. All output is standardised to English with consistent units regardless of source language.

How quickly is data updated after a company publishes a new report? +

Our monitoring system detects new disclosures within hours of publication. Extraction and validation typically complete within 24–48 hours, compared to weeks or months with manual workflows.

Built with cutting-edge technology