Replacing a manual review with evidence-derived acceptance gates

Writing low-income housing as a commercial insurance property required a lengthy evaluation by an underwriter and then a senior underwriter — of the building, the client base, and a range of decisions that were made from consistent inputs but varied with the individual doing the review.

Underwriting resisted automating acceptance, holding that these risks required professional, case-by-case judgement.

We automated the approval decision by deriving acceptance gates for each individual building feature — using information the front-line agent already collects at the point of sale.

We tested the underwriters' own stated requirement empirically: the additional information they asked for had no impact on claims payouts, measured against the existing claims categories.

• The entire lag time associated with the manual review was removed.
• The category was proven profitable — the sentiment attached to low income was not a risk signal.
• Underwriting capacity was released, and category profitability improved, because the manual decisions had been negatively impacting profitability in that category.

Resistance was defended as professional judgement, not job protection — so the claim was testable, and it was falsified. The data showed the manual process itself was the problem: lag was suppressing a profitable line while the organisation believed it was exercising caution.