The supplier SDS arrives in German and sits in an inbox until someone translates it.
The Safety Data Sheet of a chemical product is mandatory under REACH and CLP — and in the real plant it arrives in German, Italian or French, gets forwarded to quality, and the ERP product record still carries the old version. iLEAN Connect intercepts the email, recontextualises the 16 sections into English and links the SDS to the product in the ERP. The manager signs it off — the plant operates with the correct version.
Critical information about a chemical does not come in through a sensor — it comes in through an email in German.
In a mid-sized chemical distributor or plant, the real SDS flow looks like this:
- The supplier sends the email with the SDS attached, in their own language — German, Italian, French, English, Dutch. For them this is correct: it is the original certified document.
- The email lands in purchasing or in a shared mailbox. Someone forwards it to quality or compliance. Someone downloads it, saves it in a network folder by supplier, and pings whoever has to update the product record.
- The English translation arrives later — or arrives incomplete, or never arrives. The product record in the ERP still carries an SDS two versions old. The plant operator checks whatever is at hand, and whatever is at hand is the old one.
Multiply that by hundreds of suppliers and you understand why section 8 (exposure controls) or section 11 (toxicology) of your internal database is never up to date. It is not carelessness: it is data islands in different languages with nobody to operate them. Gap one (data in heterogeneous formats) and gap two (no agent to process it) are exactly this problem.
iLEAN does not change the email flow — it recontextualises what arrives and deposits it in the ERP.
The problem is not missing information: the SDS is there in the email, complete and certified. It is information living in a German PDF that never reaches the ERP and that, when it does, arrives late and badly. iLEAN Connect acts as the filler that seals the crack between the supplier inbox and the product record, without asking you to change ERP and without asking the supplier to change anything.
Connect captures the email. The agent extracts the 16 sections, recontextualises them into English while respecting the standardised H and P statements, and links the SDS to the ERP product. The manager signs it off — never the other way round.
The iLEAN pieces applied to the multilingual SDS parser:
- Connect — the ear on email. Shared mailbox or automatic forwarding from the purchasing mailbox. It detects SDS attachments in PDF, DOCX or scan, identifies the source language and archives the original email as chain of evidence. The supplier keeps sending exactly what they always sent.
- Agents — the recontextualisation. An anchored task par excellence: the 16 sections of REACH Annex II follow a known structure, and H and P statements are standardised across every official EU language. The agent extracts each field, quotes it in the source language as evidence and returns the English version with the correct technical term. It cross-checks name, CAS and product code against the ERP record and links them.
- The three rings. Extraction and translation live in ring 3 (where the AI has autonomy). Ring 2 verifies: does the product exist in the ERP? Is this SDS version later than the previous one? Does the CAS match? Only then does ring 1 replace the SDS in the product record — and never without a signature from the chemical safety or quality manager. The previous version stays archived for audit.
Manual forwarding to quality vs. a parser that links to the ERP at second zero
| Aspect | Classic SDS-by-email flow | With iLEAN Connect + parser agent |
|---|---|---|
| Input language | The supplier's; translation dependent | DE, IT, FR, ES, NL, PT → EN recontextualised |
| Time to reach the ERP | Days or weeks after forwarding | Second zero as a proposal; minutes to signature |
| ERP version vs. supplier version | Frequently out of sync | Detects version, flags obsolete ones, asks for signature |
| H and P statements | Rewritten; risk of losing literal wording | Quoted literally (standardised EU code) |
| Traceability for a REACH/CLP inspection | Loose emails scattered across folders | Chain: original email + extract + signature |
| Load on the quality manager | Translate, transcribe, link | Validate the match and sign — not transcribe |
Impact estimate for your plant — to be validated with your own numbers.
The block below is an estimate to be validated with the actual data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.
- Chemical distributor or plant with 40-200 European suppliers, SDS received in 3-6 languages, a typical ERP (SAP, Dynamics, Sage, Odoo). Pilot on the 20 suppliers that concentrate the bulk of the catalogue — the Pareto from the book.
- Shared mailbox + parser agent + ERP integration. First value expected within a few weeks: the first foreign-language SDS arriving already linked to the right product and signed off by quality.
- Indicative payback between 4 and 9 months. Hard levers: quality hours given back (translating and transcribing is what eats the most time), lower risk in a REACH/CLP inspection with an outdated record, avoiding the incident of an operator consulting an old SDS.
- Expected reduction in SDS update time in the ERP of ≥30% in the first quarter, conservatively — the ceiling is set by the quality of the PDF the supplier sends.
And the fair doubt of the chemical safety manager
What if the AI mistranslates an H statement and leaves a hazard out of the record? — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely recontextualises a data point from one format into another (German PDF → ERP field in English), the best models brought the error rate below 1.5% [1]. And even then, the critical part is not decided alone: H and P statements are quoted in their standardised EU code (H315, P280…), never rewritten, and the chemical safety manager signs off before the change enters the ERP. The previous version stays archived for audit.
[1] OpenAI paper “Why Language Models Hallucinate”, 2025 — on the reliability of AI in anchored tasks.
What people ask about the multilingual chemical SDS parser
What is an SDS and why does it arrive in different languages?
The Safety Data Sheet (Sicherheitsdatenblatt in German, Scheda Dati di Sicurezza in Italian, Fiche de Données de Sécurité in French) is the document a chemical supplier is legally obliged to send to the customer under the REACH Regulation and CLP. It has 16 mandatory sections (identification, hazards, composition, first aid, handling, and so on). The law requires the manufacturer to supply it in the language of the member state where the product is placed on the market, but in practice the German supplier first sends the original SDS in German and the translation arrives weeks later — or never arrives complete.
How does iLEAN Connect intercept the supplier email without changing the flow?
Connect sets up a shared mailbox (sds@yourcompany, suppliers@yourcompany, or forwarding from the existing purchasing mailbox) and captures every incoming email with attachments. It does not force you to change the supplier address or to train the supplier. It automatically detects whether an SDS is attached (PDF, DOCX, even scanned), identifies the source language and passes it to the extraction pipeline. The original email stays archived as chain of evidence.
How does it extract the 16 sections and translate them without losing technical data?
A language model tuned for SDS recognises the 16 sections in the source language (REACH Annex II standard) and extracts what matters: chemical name, CAS number, EC number, GHS pictograms, H and P statements, limit values. The recontextualisation into English preserves the technical terminology (it does not turn Gefahrenhinweis into a loose hazard notice — it uses hazard statement H with the original code). H and P statements are quoted literally because they have a standardised form in every official EU language.
How does it link the SDS to the product in the ERP?
The agent cross-checks the product name in the SDS, the CAS number, the manufacturer code and the order code against the product record in the ERP (SAP, Dynamics, Odoo, Sage). If the match is clear, it attaches the SDS to the product, updates the hazard fields, the H and P statements, and records the SDS version and date. If the match is ambiguous, it escalates to the quality manager for validation. It never matches blindly — the three-ring pattern applies here too.
What if the SDS contains a critical value the AI extracted wrong?
SDS extraction is an anchored task: every field comes from a specific section of the original PDF, not from free generation. The best models brought the error rate on this kind of task below 1.5%. Even so, the iLEAN flow does not close on its own: the SDS stays in proposed state with every field linked to the original line of the PDF, and the chemical safety or quality manager signs it off after review. The previous version stays archived for REACH/CLP audit.
Tell us your case and within 48h we will send you the estimated ROI of this parser for your chemical suppliers.
We work on the real data of your plant, not on ours. Diagnostic with no commitment.
Request estimated ROI in 48h See iLEAN Connect