Platform12 min read

AI report assistant: cut LCA and EPD interpretation from hours to minutes

A built-in AI assistant drafts the EN 15804+A2 interpretation in seconds. Architecture, prompt design, audit logging and the human approval loop.

By Last updated

A dense results report distilled into a single summary card by a compact assistant

In an EPD report, the most time-consuming part is not the calculation — it is the interpretation. Reading the dominance analysis, identifying hotspots, writing in industry best-practice phrasing — that is 8-12 hours in a traditional flow. The AI report assistant built into clca cuts that part to minutes. This article covers the architecture, prompt design, audit logging for the verifier and the human approval loop.

What the assistant does — and does not do

The AI assistant is trained on EN 15804+A2 methodology, the modulary table, the 19 indicators and industry best-practice phrasing. It reads the LCA output — dominance analysis, sensitivity, modulary table, DQR scores — and drafts the EN 15804+A2 interpretation section in natural language. It identifies hotspots ("A1-A3 dominates GWP at 78%", "the clinker ratio is the leading driver"), standardises tone and ensures grammatical consistency.

Architecture: deterministic engine + AI language layer

A single clean summary card resting on a thick report
  • 1) Deterministic LCIA engine — ecoinvent v3.12, 23+ categories, 7 LCIA methods
  • 2) Structured output: dominance JSON, sensitivity matrix, modulary table
  • 3) AI prompt layer: packs the structured output into LCA-trained model context
  • 4) Generation: EN 15804+A2 interpretation draft (3-5 paragraphs)
  • 5) Human approval UI: sentence-level edit / approve
  • 6) Audit log: which prompt, which output, who approved — visible to the verifier

How verifiers see it

Accredited verifiers (ITB, IBU, EPD International, ECOnsult) get full access to the audit log when reviewing an AI-assisted EPD. Every AI generation is logged with: model used, prompt hash, generation time, generated text, user edits, final approval. The verifier compares the draft to the final text to confirm human editorship actually happened. The independence and transparency requirements of ISO 14025 and EN 15804+A2 interpretation are thereby preserved.

Typical savings

  • Manual interpretation (traditional): 8-12 hours
  • AI draft + human approval (clca): 15-25 minutes
  • Typical saving: 92-95%
  • Quality: verifier acceptance rate equals manual (industry best-practice phrasing)
  • Saving compounds for LCA-copied variants — 10 variants = 80-120 hours saved

Tags

  • AI LCA
  • AI EPD assistant
  • AI report assistant
  • LCA automation
  • EPD interpretation automation
  • EN 15804 interpretation