A Human in the Loop Is Only a Control If the Loop Works

A lawsuit served on June 15 reached one commercial carrier’s inbox on August 20, 66 days after service. The answer period was thirty days. On a primary policy, the carrier would already have missed the deadline to answer and has now been exposed to a default judgment.

That example comes from a 2026 pilot with a commercial carrier on general liability and commercial auto claims. A person in the loop cannot catch what arrives after the window has closed.

Having a human review is the right model currently for claims decisions. Regulators expect the record to match: the NAIC’s model bulletin on insurers’ use of AI, adopted by about two dozen states as of early 2025, calls for a written AI program with governance and documentation that insurers can produce on request.

That puts the weight on the loop itself. A person in the loop is a safeguard on paper. Whether it holds up in practice comes down to three questions.

Three questions that tell you if your loop holds

1. Did they see it in time? A policy-limits demand. A letter of representation. An attachment that quietly changes severity. These don’t wait for triage. If AI only sees them once the claim is already in the system, your adjuster can end up signing off on a decision whose window has already closed.

2. Can they check it quickly and accurately? If the recommendation doesn’t point to the exact sentence, page or attachment, the adjuster has to reread the whole file or take it on faith. On a busy desk, faith wins. In one anonymized claim, an 11-page scanned demand letter arrived on day 23. The spoliation risk finding pointed to page 9, punitive damages to page 5 and attorney representation to page 1, so the reviewer could go straight to each one.

3. Can you prove it later? What arrived? When? What did the AI surface? Who saw it? What did they decide? When a regulator or opposing counsel asks, that record is your oversight. Without it, the human in the loop is just a name in a log.

How to test your own loop

Take the closed litigated claims from the last 12 to 24 months. For each one, find when the first warning arrived and when the team acted. Then find the days from the first warning to the reserve change. Add one more interval for the desk: the hours from receipt to recognition, by communication type, and again for your last catastrophe week.

Those intervals show whether the person in your loop had enough time to matter, and they come from your own closed files.

Where Isomer fits

Isomer™ works at the point of receipt. It reads each claim communication as it arrives, cites every finding to its source sentence or page, and writes the finding back to the claim file, showing what arrived, when and what was surfaced. It is core system agnostic, and the adjuster makes every call.

In the pilot behind the opening example, Isomer raised 17 alerts on 12 claims in the first three days, and six fatality claims surfaced from routine mail. By the second week, one in four new claims carried a signal.

A four-week assessment on closed litigated claims measures the intervals above in your own book: when the warning arrived, when the team acted, and the days of reserve lag.

The question to take to your team

Most carriers chose humans. Now they have to back that person up.

An adjuster’s approval is only as strong as the record behind it.

Sources: NAIC, Model Bulletin on the Use of Artificial Intelligence Systems by Insurers. Isomer pilot results, 2026, commercial carrier, general liability and commercial auto. Isomer anonymized claim example.