Claims AI should shorten the path from an open file to a completed claim communication. When a tool produces a quick answer that takes longer to correct, the adjuster inherits a second assignment: repairing the automation while keeping the claim moving. Recent employee research gives claims leaders a reason to look closely at that work.
Adjuster complaints can reveal work that claims AI leaves unfinished
In research published August 27, Glassdoor found that 98% of claims-adjuster comments about AI were critical in reviews submitted between June 2025 and May 2026. The examples included frustration with mandatory use and time spent correcting AI-generated errors.
The finding describes people who chose to mention AI in an employer review. Glassdoor’s methodology classifies sentiment by whether those mentions appear in the review’s pros or cons. The public report does not disclose the claims-adjuster sample size. It cannot tell us how all adjusters feel or establish whether a particular tool improves productivity.
But the complaints point to something claims leaders can investigate directly.
In WIRED’s August 31 reporting, adjuster Ahmad Jackson described an employer’s AI intake rollout that left staff rerouting misclassified claims. He also encountered erroneous summaries and faced upset claimants or attorneys when he passed those errors along. His account illustrates how work can travel beyond the step being automated.
A manager hearing similar feedback should ask the adjuster to show the affected files. Which information needed correcting? Where did the correction occur? Who else had to act before the file could move forward? Those answers can reveal a problem that a usage dashboard will never explain.
One missed detail can create several rounds of correspondence
Consider a hypothetical property claim. The policyholder has supplied photographs and a contractor’s estimate through separate channels. An automated summary captures the photographs but misses the estimate. A request for information is drafted from that summary and asks the policyholder to provide the estimate again.
The request goes out before anyone catches the omission. The policyholder calls to explain that the document was already submitted. The adjuster retrieves it, reconciles the summary with the file, documents the correction, and prepares a follow-up explaining what information is still needed. A supervisor may then need to review the revised communication.
The original request might register as a letter generated quickly. The correction could be recorded elsewhere as a phone call, a file note, and another letter. If the evaluation stops at generation, it misses the work the error created.
The location of the fix matters too. Catching the omission before sending still consumes time. Catching it after sending adds a policyholder interaction and an explanation for the repeated request. That is why a small drafting error can have a larger effect on adjuster workload than its word count suggests.
This example also shows why the response must reach beyond editing one letter. Correcting the summary’s access to submitted documents could prevent the repeated request. Asking adjusters to keep repairing the same omission leaves the underlying cause in place.
Measure the work required to finish the claim communication
Claims automation needs an evaluation that follows the output into the next person’s work. For correspondence, use completion of an accurate, reviewed communication as the endpoint, then check whether that communication required avoidable follow-up.
Start with a small, representative set of letters from the assisted workflow. Include straightforward cases and cases with multiple documents or unresolved questions. Follow each through drafting, review, dispatch, and any related correction. Compare it with similar work completed under the previous process.
For each communication, record three things:
- What work disappeared? Policy searches, copying language, formatting, or assembling a first draft may take less time.
- What work was added? Identify corrections, repeated requests, extra handoffs, and follow-up contacts caused by the output or the way it entered the process.
- Who performed that work? Include supervisors and reviewers, as well as the adjuster using the tool.
Classify edits by reason. A new estimate arriving after drafting is different from an estimate that was available but omitted. A reviewer refining an explanation is different from a reviewer repeatedly repairing incorrect claim facts. That distinction helps the team find problems it can actually fix.
Keep labor time and elapsed time separate. A letter can wait two days for five minutes of review. Removing that wait improves communication speed; reducing the review effort creates capacity. Both matter, and they require different changes.
This extends the claims AI ROI measurement approach to work that may be missed when teams report results separately. A reported gain should account for the people who receive, check, and act on the output.
Give adjuster feedback the power to change the workflow
In a September 2 interview with Insurance Business, Crawford’s chief AI, data, and cloud officer, David Wright, described testing workflow prototypes with adjusters and claims specialists. Teams repeatedly check accuracy, identify problems, and refine tools before wider use.
That approach gives employee feedback an operational consequence. For a correspondence team, the next step could be changing which documents feed a draft, making supporting policy language easier to inspect, or narrowing the letter types included in a rollout.
Assign recurring problems to someone who can change the process or product. Then return to the affected adjusters with the revised version and the files that previously caused trouble. Check whether the correction work actually declined.
Training should address unfamiliarity. Product and workflow changes should address recurring defects. A team needs a way to tell which problem it has, and adjusters need evidence that reporting a defect produces a response.
Reduce the drafting and review work your team has to finish
For claims correspondence, the opportunity is concrete: less time finding policy language, assembling the letter, formatting it, and resolving avoidable reviewer questions.
Voltaire supports that work with policy-supported drafts and review context that helps teams evaluate the letter. Correspondence is prepared for review and approval within the carrier’s claims process.
The same standard applies to Voltaire as to any claims technology: demonstrate that the team can complete the work with less effort. Bring a letter that routinely needs extra attention to the evaluation, and follow it all the way through review.
Request a demo to see how Voltaire handles correspondence drafting and review.
