Why AI Document Extraction Matters for Battery Passport Compliance
Regulation (EU) 2023/1542 requires economic operators to populate a battery passport with a defined set of data before placing certain batteries on the EU market. The data fields are drawn from Annex XIII of that regulation, which covers everything from electrochemical performance to carbon footprint to hazardous substance declarations. The problem is that none of this information lives in one place. AI document extraction for battery passport compliance is the process of pulling those values out of the scattered source documents your supply chain already produces, and mapping them to the correct fields in a structured data template.
This article explains how that extraction process works in Traceable, what it handles reliably, and where human review remains essential.
Readiness Assessment
Find out if your operation is ready for EU DPP compliance.
Five questions, about two minutes. You get a score for each area and a numbered action plan for the gaps.
Score Your ReadinessWhere Battery Passport Data Actually Lives
Before any extraction can happen, it helps to understand the document landscape your team is working with. Battery passport data is typically spread across at least six document types:
- Test reports to UN 38.3 and IEC standards, covering safety and electrochemical performance
- Safety data sheets, which carry hazardous substance information
- Material certifications from cell and component suppliers
- Declarations of conformity, including CE declarations
- Carbon footprint life-cycle assessment (LCA) reports
- Supply chain audit reports covering due diligence obligations
Each document type uses its own format, terminology, and structure. A UN 38.3 test report from one laboratory looks different from one produced by another. An LCA report from an environmental consultancy follows a different layout than a safety data sheet. Extraction software must handle that variation without requiring a custom template for every supplier.
What Traceable’s AI Document Extraction Does
Traceable accepts PDF, PNG, and JPG files for AI document extraction. Other file formats are rejected at upload. Once a file is accepted, the system reads the document and proposes where each extracted value belongs in the product’s data template.
For battery categories, that template is built from Annex XIII of Regulation (EU) 2023/1542, which sets out the information the passport must carry. For document types the system recognises, the extraction call is scoped to the template fields that document type can evidence; an unrecognised document is asked about every field in the template. In every case the system produces a proposed mapping, and the operator must confirm each assignment on the Field Mapping Review screen before it is applied to the product record.
That distinction matters. A carbon footprint figure expressed in kilograms of CO2 equivalent is proposed for the carbon footprint field under Annex XIII, point 1(c), which references the carbon footprint declaration in Article 7(1) and (2). For a UN 38.3 test report, the system proposes an assignment based on what it reads, and a person must check whether that proposal is correct in every case.
What AI Extraction Handles Well
The extraction performs reliably on three categories of information:
- Stated numerical values. Capacity in ampere-hours, nominal voltage, rated energy, and a carbon footprint figure expressed as a single number are all read accurately when they appear clearly in the source document.
- Test results with pass or fail verdicts. When a test report states that a battery passed or failed a specific test, the system captures that verdict and maps it to the relevant field.
- REACH SVHC substance presence. When a REACH SVHC declaration lists substances of very high concern, the system identifies which substances are listed and maps that information to the hazardous substance fields. Annex XIII, point 1(a), brings Annex VI, Part A into the passport, which covers general battery model information accessible to the public, including substance data.
These are the cases where the source document contains a clear, discrete value and the extraction has a defined target field. The system is reading and relocating information, not interpreting or calculating it.
What AI Extraction Does Not Do
Understanding the limits of AI document extraction for battery passport data is as important as understanding what it does. There are four things the system cannot do, and operators need to plan for each of them.
It does not verify that a value is correct. If a test report contains an error, the system will extract that error faithfully. Extraction is not audit. The system reads what is on the page; it does not cross-check the figure against a standard or against another document.
It does not fill in data that does not exist in the source document. If a supplier has not provided a carbon footprint figure, no extraction process will produce one. The field will remain empty until the data exists somewhere in a document the system can read.
It does not replace a missing test report. Annex XIII, point 3, requires test report results to be accessible only to notified bodies, market surveillance authorities, and the Commission. If the test has not been conducted, there is no report to extract from, and the passport cannot carry that information.
It does not make compliance judgements. The system does not tell you whether your battery meets the performance thresholds set in Regulation (EU) 2023/1542 or its delegated acts. It maps values to fields; it does not evaluate those values against regulatory limits.
There are also specific data types the system does not currently extract. CAS numbers are not captured by the AI extraction process. Substance concentration and recycled content percentage fields are among those the extraction call asks about for the relevant document types; like every extracted value, the result still needs operator confirmation before it is saved. Those values must be entered manually. This is relevant because Annex XIII, point 1(b), requires information on hazardous substances present in the battery, and Annex XIII, point 1(e), requires recycled content information as contained in the documentation referred to in Article 8(1). Operators who rely solely on extraction will find those fields empty and must plan for manual data entry to cover them.
For a full picture of the EU Battery Regulation 2023/1542 data requirements that feed into the passport template, including which Annex XIII points apply to which battery categories, that reference page sets out the structure in detail.
The Operator’s Responsibility After Extraction
Reviewing extracted values before a passport is published is the operator’s responsibility. This is not a procedural formality. Article 77(4) of Regulation (EU) 2023/1542 requires that the information in the battery passport be accurate, complete and up to date. The economic operator who places the battery on the market carries that obligation. Extraction is a tool to reduce the time spent locating and transcribing data; it does not transfer the accuracy obligation away from the operator.
The practical review process in Traceable works as follows. After extraction, the product’s completeness score shows what share of its required fields are filled. A separate regulation-conformance score lists which mandatory Annex XIII-derived data points are covered and which are missing. These two scores give the operator a clear view of what still needs attention before the passport can be published.
Where fields are missing because a supplier has not yet provided the underlying data, the platform allows the operator to send a supplier data request scoped to the specific fields that are empty. That request goes to the relevant supplier and asks only for the information that is missing, rather than requiring a full document resubmission.
A Note on Data Point Counts
You may encounter references to the number of data points required for a battery passport. The Commission has published a guidance document on Digital Battery Passport data points by category (Version 2.0, 15 August 2026). That document is not binding and states explicitly that it does not extend the rights and obligations in the legislation. It identifies 71 numbered data points across all categories. For February 2027, it makes 47 of those mandatory for EV batteries and 50 for LMT batteries, counting data point 51 (rated capacity, which the guidance marks as the same as data point 11 but dynamic), and 32 for industrial batteries above 2 kWh. These numbers come from the guidance document, not from Regulation (EU) 2023/1542 itself. The regulation sets the information requirements through Annex XIII; the guidance document is a non-binding interpretation of how those requirements translate into discrete data fields.
When Traceable refers internally to attributes, that refers to how the platform decomposes those data points for data entry and extraction purposes, which produces a larger number than the guidance document’s count. The two figures measure different things and should not be compared directly.
Planning Your Document Collection Process
Given what extraction can and cannot do, the most effective approach is to treat document collection and extraction as two separate steps, each with its own checklist.
First, confirm that every required document exists. For each Annex XIII data point your battery category requires, identify which source document carries that information. If a document is missing, request it from the supplier before starting extraction. Extraction cannot create information that does not exist in a document.
Second, check that your documents are in an accepted format. PDF is the most reliable format for extraction accuracy. Scanned images in PNG or JPG are accepted but may produce lower accuracy if the scan quality is poor or if the document uses complex table layouts.
Third, after extraction, review every suggested mapping for document types that do not have a predefined mapping. Confirm that the extracted value is correct and that the proposed field assignment matches the Annex XIII point it is intended to satisfy.
Fourth, use the completeness and conformance scores to identify gaps, then address them either through manual entry or through supplier data requests for the specific missing fields.
This four-step sequence keeps the operator in control of accuracy while letting the extraction process handle the time-consuming work of reading and relocating data from long technical documents.
To see how Traceable could handle this for your products, register for a free trial.
