A carrier I spoke with uses satellite imagery to detect hail damage, drones to scan rooftops, and an automated triage engine that routes claims by severity in seconds. Then an adjuster opens a Word document, copies last week’s denial letter, swaps out the policyholder’s name, and hits send.

That letter, riddled with stale policy citations and leftover facts from a different claim, sits in front of a plaintiff attorney whose AI already flagged three errors before the adjuster finished lunch.

This is the last mile problem. Carriers have poured billions into modernizing every stage of the claims lifecycle except the final, most consequential step: the letter that communicates the coverage decision. And in 2026, that gap is where lawsuits are born.

The Last Mile Nobody Talks About

The term “last mile” originates in logistics. It describes the final leg of delivery, the shortest stretch, yet the most expensive and failure-prone. In claims, the parallel is exact.

Carriers have invested heavily in first notice of loss (FNOL) automation, aerial estimating, telematics, and predictive triage. These technologies work. They speed up intake, reduce fraud, and improve accuracy at the front of the pipeline. But when the adjuster sits down to draft the settlement letter, denial, or reservation of rights (RoR), the workflow reverts to something recognizable from 2005.

Workflow analysts have identified this exact breakdown. FitGap’s diagnostic on claims closure describes adjusters hunting for scattered evidence, photos, receipts, vendor notes, across fragmented inboxes and shared drives right at the moment they need to compose a legally binding document. 

McKinsey has noted a persistent “last-mile gap” in enterprise data strategies: organizations possess sophisticated analytics but fail to translate them into clear, compliant end-user communication. In insurance, this means a carrier can deploy AI to price a roof replacement down to the shingle, yet still rely on a human to manually transcribe the coverage rationale into a letter.

The Copy-and-Paste Monster

Why do adjusters draft letters by using old letters in template libraries? It’s not laziness. It’s system design.

When document logic and templates are hard-coded into legacy core systems, a simple change to a regulatory disclosure becomes a multi-month IT project. The systems can’t adapt dynamically to the nuances of a specific claim. So adjusters maintain personal folders of old letters, possibly on top of a larger template library used by the department. They copy, paste, and edit. 

I’ve seen this firsthand at the carriers we work with. When you measure productivity by volume, people choose speed over compliance. I don’t blame them. The problem isn’t the adjuster. It’s the system built around them.

The consequences of copy-and-paste are predictable. A similar but different policy gets cited. An endorsement from a prior claim stays in the letter. During a catastrophe surge, when Department of Insurance regulations compress response timelines, the reliance on recycled templates spikes. Template drift accelerates. Errors multiply at exactly the moment when claim volume, and exposure, is highest.

The Other Side Has AI. You Don’t.

While carriers still write letters by hand, the plaintiff bar has industrialized. The era when a policyholder might overlook a vague denial or a misquoted exclusion is over.

EvenUp, an AI platform built for personal injury firms, recently raised $150 million in Series E funding, pushing its valuation past $2 billion. The platform processes roughly 10,000 cases per week across more than 2,000 law firms. Its AI ingests hundreds of thousands of historical cases, medical records, and insurance policy data to auto-assemble demand packages and compute settlement ranges from vast legal databases. Law firms using these tools report doubling their demand output and hitting policy-limit settlements more frequently, without adding headcount.

These tools don’t just speed up plaintiff workflows. They specifically hunt for inconsistencies in carrier communications. A vague denial, a missing exclusion reference, a date that doesn’t match the file are all the kinds of inputs that a $2 billion algorithm converts into leverage.

The threat extends beyond law firms. Consumer-facing platforms are arming policyholders directly. In health insurance, Counterforce Health offers free AI that ingests a denial letter and the patient’s policy, then generates a comprehensive appeal in minutes. Their reported success rate for overturning denials sits between 70% and 80%. The AI specifically targets administrative errors, incorrect coding, and vague reasoning in the carrier’s original letter.

The underlying mechanics are identical whether the algorithm is parsing a medical billing code or a homeowner’s roof damage estimate. It ingests the carrier’s letter, locates the missing citations, and outputs a structured dispute. The manually drafted letter is the specific trigger that activates the consumer AI.

P&C is next. The tools don’t care which policy sits on the desk.

The 4x Cost of Getting It Wrong

These aren’t abstract risks. The financial penalty for a flawed letter is measurable.

A CLARA Analytics study of over 50,000 workers’ compensation claims found that attorney involvement increased total claim costs by nearly 400% and prolonged claim duration by 195%. This pattern holds across broader P&C lines: commercial auto, general liability, homeowners. Once a claim tips into litigation, the financial trajectory changes completely.

The primary job of the initial claims letter is litigation avoidance. When an adjuster sends a vague or poorly reasoned denial, the policyholder feels dismissed. They call an attorney. And a claim that could have closed for its indemnity value now carries a 4x cost multiplier in defense fees, extended medical treatments, and prolonged indemnity periods.

This is happening against a macroeconomic backdrop that makes every mistake more expensive. A 2025 study by the Insurance Information Institute and the Casualty Actuarial Society found that legal system abuse added between $231.6 billion and $281.2 billion to liability insurance losses over the past decade. Claim frequency is falling. Claim severity is skyrocketing. The Swiss Re Institute confirmed that litigation costs drove U.S. liability claims up by 57% over ten years, peaking at 7% annual growth in 2023, independent of standard economic inflation.

A manually drafted, legally imprecise claims letter hands the plaintiff’s attorney a weapon in a judicial system already warped by social inflation and nuclear verdicts. The letter is where the exposure starts.

The Letter That Gets You Sued

The title of this article is not hyperbole. Courts routinely use poorly drafted correspondence as the foundational evidence in two of the most damaging legal actions a carrier can face: equitable estoppel and bad faith.

Estoppel: the boilerplate trap. When a carrier issues a reservation of rights letter, it must clearly identify the specific policy provisions at issue and explain how the facts of the claim interact with those provisions. Courts are unforgiving on this point. In Harleysville Group Insurance v. Heritage Communities, the court found a carrier’s RoR legally ineffective because it used a “cut-and-paste method” to list generic policy excerpts without tying them to the specific factual allegations. The carrier was estopped from denying coverage, forced to pay claims the policy was designed to exclude.

This is not an edge case. Courts in Pennsylvania, North Carolina, and Missouri have all ruled that vague, boilerplate RoR letters can invalidate a carrier’s coverage defenses and even support bad faith recovery. The exact operational workaround that adjusters rely on, copying and pasting template language, is what courts punish.

Bad faith: when errors become evidence. Every insurance contract carries an implied covenant of good faith and fair dealing. When a denial letter provides shifting reasons, relies on vague jargon, or fails to cite the specific policy provision, plaintiff attorneys present that letter to a jury as proof of malice or willful disregard.

State regulations enforce this directly. Missouri’s Code of State Regulations prohibits denying a claim unless the denial explicitly references the exact policy provision at issue. Washington state law similarly mandates that a bare “not covered” statement is a regulatory violation.

In the 2025 federal case Hersh v. State Farm Fire & Cas. Co., plaintiffs survived a motion to dismiss by alleging fraud and bad faith based specifically on the contents of the adjuster’s letter. The allegation: the adjuster incorrectly stated that water damage occurred before the policy’s effective date. A manual error—misreading a date—reframed by the plaintiff as a deliberate attempt to defraud the consumer.

This is the risk profile of every manually drafted letter leaving your office today.

Why ChatGPT Won’t Save You

Carriers recognize the problem. Many are exploring AI solutions. But there is a critical distinction between deploying a generic large language model (LLM) and implementing purpose-built, explainable AI for claims correspondence.

The NAIC’s Model Bulletin on Artificial Intelligence, adopted in late 2023 and now being implemented by major state Departments of Insurance including New York, California, and Pennsylvania, imposes strict requirements on AI used for adverse claim actions. The framework mandates full explainability: carriers must document how the AI functions, which data it uses, and the logical chain from input to output. It requires continuous testing for unfair discrimination. And it demands human-in-the-loop oversight for high-stakes decisions like coverage denials.

A generic LLM fails every one of these tests. Standard models are prone to hallucination, such as inventing policy language, fabricating legal precedents, or generating plausible-sounding exclusions that don’t exist in the actual policy document. If a hallucinated exclusion appears in a denial letter, the carrier faces bad faith exposure and regulatory fines simultaneously.

What the NAIC framework demands is Explainable AI (xAI): a system that ingests the actual policy document, extracts verbatim language without alteration, and provides an auditable reasoning trail for every citation. The system must also feature compliance guardrails. If the AI cannot locate valid exclusionary language in the specific policy, it must refuse to generate a denial rather than invent one.

At Voltaire, this is exactly how we’ve built it. When an adjuster requests a denial letter but the policy provides no clear basis, our system returns “No relevant policy language was found.” It won’t fabricate a reason. That guardrail isn’t a convenience feature. It’s a regulatory requirement and a legal shield.

Fixing the Last Mile

The independent data paints a sequential picture. Legacy systems force adjusters into manual workarounds. Plaintiff AI detects the resulting errors at scale. Those errors push claims into litigation, triggering a 4x cost multiplier in an environment already inflated by social inflation and nuclear verdicts. Courts use the flawed letter itself as the primary exhibit.

The fix starts with the letter.

At Voltaire, we produce accurate, policy-cited claims correspondence in as little as 30 seconds. The AI connects directly to the carrier’s claims system, pulls the relevant policy, and generates a letter grounded in the exact language and facts of the loss. It requires minimal effort from adjusters, which is why 85%+ of adjusters at our clients’ carrier claims departments actually use it. That adoption rate is the real proof point. The best compliance tool in the world is worthless if nobody opens it.

One senior claims counsel at a publicly traded carrier told us they gave Voltaire five letters they considered perfect and asked the system to replicate them. It corrected mistakes they had missed.

During catastrophe surges, we onboard hundreds of independent adjusters and eliminate the day-one learning curve. A CAT adjuster unfamiliar with the carrier’s policy language can produce a letter that passes QA on their first attempt. When you’re paying overtime and premium rates for temporary resources, having them productive from hour one directly impacts loss adjustment expense and settlement quality.

The broader strategic value cascades from there. Supervisory review becomes focused on validating facts and coverage stances, not redlining grammar and fixing section references. Adjusters are freed from the administrative burden of policy-language lookup, reducing burnout and improving retention. Every letter leaving the office is objective, auditable, and defensible, removing the systemic risks of estoppel and bad faith at their source.

The Window Is Closing

P&C carriers are starting to use purpose-built AI for claims correspondence. Plaintiffs and consumers are not waiting either.

Five years from now, I expect boards and regulators will ask carriers why they haven’t adopted AI correspondence. The question won’t be “should we adopt?” It will be “how fast can we catch up?”

Carriers that fix the last mile now gain a measurable advantage: fewer disputes, faster cycle times, lower LAE, and adjusters who spend their energy on claim resolution instead of wordsmithing letters. Carriers that wait will keep feeding imprecise correspondence into a system designed to punish every error.

You automated the drone. You automated the triage engine. You automated the estimate. Now automate the letter. It’s the one that gets you sued.

Yo Sub Kwon is the Founder and CEO of Voltaire, an AI-powered claims correspondence platform serving property and casualty insurers. Learn more at voltaire.claims.