Automating DNSH evidence for ISR label renewal

For an ISR-labelled fund, renewal season used to mean one thing for the stewardship team: three weeks of pulling Do-No-Significant-Harm evidence by hand. Here's how that becomes an afternoon.

France's ISR label has become materially more demanding, and the Do No Significant Harm (DNSH) test sits at the centre of it. To keep the label, a fund has to show — holding by holding — that its investments don't significantly harm any environmental or social objective, evidenced against the relevant Principal Adverse Impact indicators. The standard is reasonable. The way most teams meet it is not.

The typical workflow is a scavenger hunt. An analyst opens the holdings list, then cross-references adverse-impact data from one provider, controversy screens from another, and engagement notes from a shared drive — copying each result into a spreadsheet that becomes the "evidence pack." For a multi-fund range, that's weeks of work, and every renewal starts from scratch.

The automation we built collapses that. It takes your holdings, maps each position to the DNSH indicators the label requires, and pulls the supporting evidence — adverse-impact data, controversy flags, and your own engagement and voting history — into a single structured view. Where the evidence is complete, it's assembled. Where it isn't, the gap is flagged explicitly, so the team spends its time on the handful of positions that actually need a human decision rather than re-checking the hundreds that don't.

Crucially, engagement is treated as first-class evidence. If a holding trips a DNSH indicator but your team has an active, documented engagement addressing exactly that issue, that context is attached automatically — which is often the difference between a defensible position and a forced divestment.

The output is a dated, sourced evidence pack that maps one-to-one to the label's requirements, ready for review and sign-off. What took three weeks of manual pulling becomes an afternoon of checking and approving.

It does not remove the human step, and it shouldn't. The label is a commitment, and a person still owns the judgement on the edge cases. But there's no reason that judgement should be buried under a fortnight of copy-paste.