For Kimo Winterbottom & the Bridger team · from Sean Svendesen & Matt Soldo
Rampart AI is our working name for a supervised team of AI agents that would watch every demand, score every claim, and draft every document across the litigation lifecycle — while your adjusters and licensed counsel keep every decision that matters. It doesn’t exist yet. This page is what we’d build, why we think it pays for itself, and the questions only you can answer.
the visible number
the real exposure
aimed at you today
AI-built, human-governed
The stakes
This is outside-in research — public cases, industry data, vendor claims. You and your team live this book every day, so you’ll know immediately where we’ve got it wrong. That correction is exactly what we’re asking for.
Your defense costs are paid outside policy limits — a litigated file routinely costs more to defend than the policy could ever pay. The real exposure is what happens when a time-limited settlement demand is missed or fumbled: courts have turned a $100K policy into a $9.2M bad-faith judgment. On a 30/60 book, every serious claim is a potential setup.
Plaintiff firms use tools like EvenUp and Supio to generate litigation-grade demand packages at near-zero cost — with vendor-claimed 30%+ higher settlements and 69% more policy-limits settlements. There is no defense-side equivalent at scale. Bridger faces this asymmetry on every demand it receives, today.
AI that turns a lawsuit into a complete draft answer in minutes is already in production at a Fortune 50 carrier. Triage, record review, deposition tools — all proven individually. What no one sells is an integrated system tuned to a minimum-limits CA/TX book, with the audit trail and human controls a carrier needs. That is what we build.
Drawn to scale — at this scale, a defended file and the policy itself nearly vanish. That’s the point: avoiding a single excess-verdict setup pays for this entire program, which is why the system’s highest-value job is never missing a demand, not drafting cheaper briefs.
The asymmetry
AI-drafted demands now make small claims worth lawyering, so demand volume and completeness keep rising — while your response capacity is flat. Industry analysts call it the defense intelligence gap. Rampart AI closes it: every inbound package is deconstructed line by line — billed versus paid, duplicate charges, treatment buildup, causation gaps — and answered with a cited rebuttal in hours.
The system
Each agent does one job and does it relentlessly — around the clock, on every file at once. An orchestrator keeps each claim’s state and dispatches work. Here they are laid out along the life of a claim — tap any agent; a ▶ means it links to a live demo you can run.
The life of a claim
Intake
the claim arrives
Triage
day one
SIU & adjusters set the track
Pre-suit
the danger zone
adjuster + counsel decide
Litigation
if suit is filed
counsel signs & files
Resolution
settle or defend
the adjuster decides
Always on, every stage
Never miss a time-limited demand
Reads every inbound attorney letter and answers one question: is this a time-limited policy-limits demand — explicit or disguised? It validates the demand against California and Texas statutory rules, starts a countdown clock, and escalates with a ready-to-sign acceptance package.
A mishandled demand is how a small policy becomes a seven-figure judgment — in Harvey v. GEICO, a $100K policy became $9.2M in liability. On a 30/60 book, the same trap is set on every serious claim. This agent can only recommend acceptance or escalate — it is structurally incapable of rejecting a demand.
Human gate: Adjuster + counsel decide; the Sentinel can never reject
▶ Watch this agent work· live demoLive demos
One pattern repeats across the team: read everything, draft everything, decide nothing. Six simulations — fictional files, real mechanics. In the pilot, every one of these runs against your closed files instead.
A routine-looking letter arrives on a Friday afternoon. Buried in paragraph three is a time-limited policy-limits demand — the kind that, mishandled, becomes a seven-figure bad-faith judgment. The flagship Demand Sentinel reads it the way it reads every inbound attorney communication.
Moreno & Vásquez, Attorneys at Law · Fresno, CA
Re: Our Client R. Aguilar — Your Insured M. Delgado — Loss of 02/14/2026
Dear Claims Department:
As you know, this office represents Ms. Aguilar for injuries sustained when your insured rear-ended her vehicle at a signal on Blackstone Avenue. Liability is not reasonably in dispute.
Ms. Aguilar has completed a cervical fusion at C5-C6. Her medical specials exceed $147,000, as reflected in the records and bills referenced herein.
In the interest of efficiency, my client will execute a full release of all claims against your insured in exchange for your available bodily-injury policy limits, provided payment is received at this office within twenty-one (21) days of the date of this letter, after which this offer is withdrawn and we will proceed accordingly.
Govern yourselves accordingly.
Very truly yours,
D. Moreno, Esq.
Awaiting inbound correspondence…
The Sentinel reads the letter on the left the way it reads every inbound attorney communication.
The proposed first step · Phase 0
Before any statement of work, we propose a shadow pilot: the three agents run as one supervised pipeline against your historical claims. The system observes and scores — it sends nothing, decides nothing, touches no live file. Then we lay it side by side with what actually happened. Zero live-fire risk; real evidence on your own data.
Caught in time
A clean-liability file the pilot flags for early tender 18 days before the demand was due. In reality it sat in a queue and tried to a ~$210K excess judgment.
Caught before payment
Three 'separate' minor-impact claims the fraud flag ties to one clinic-and-attorney cluster at intake — before a dollar goes out the door.
Correctly left alone
A claim the pilot recommends handling routinely — and that's exactly what happened. A triage tool that flags everything is useless; this is the precision side.
The human-dashboard view
A working console, not a slideshow: the pre-suit worklist, each claim’s recommended track and score, the calendared deadlines with their dual-key check, the fraud and coverage flags, the full audit trail, and the backtest scorecard. Every recommendation ends at a human gate — Rampart AI recommends, your licensed adjuster decides.
Open the working console →BR-…04417
BR-…07732
BR-…09105
BR-2025-04417 · R. Aguilar
Illustrative files and figures are fictional, for demonstration only. In the pilot these run against Bridger’s real closed pre-suit files, against pre-agreed outcome labels, with strict temporal discipline. Any dollar figure is a modeled estimate, never a measured result.
The economics
Every figure below is an industry average or a vendor claim — not one of them is Bridger data, because we don’t have Bridger data. You do. Set the sliders to what your book actually looks like and see what the math says. Drag every assumption down to pessimistic and notice what happens to the bottom line.
Annual fee savings
$905,200
45% off the $2M fee line — before any indemnity effect
At these settings, modeled litigation defense alone exceeds the $2M baseline, so it is capped there.
Not in this number
Indemnity savings from demand deconstruction and earlier tenders — likely larger than the fee savings.
Also not in it
Tail risk: one avoided bad-faith setup pays for the program many times over.
Safeguards
The system never
Every settlement or tender decision goes to a licensed adjuster. Everything filed with a court or sent to opposing counsel is reviewed, edited, and signed by licensed counsel — that's Sean's side of the house. These aren't policies — they're how the system is physically constructed. It drafts and recommends; your people decide and sign.
Courts have already compelled insurers to hand over their AI's claim evaluations. So every agent output is written to professional standards from day one — factual, documented, and something you'd be comfortable seeing projected in a courtroom.
More than 1,500 court decisions now involve fabricated AI citations. Our verifier — a fully independent AI sharing nothing with the agents that write the drafts — re-derives every citation against primary sources and blocks anything unverified from leaving the system. The author never grades its own work.
Enterprise AI contracts that forbid the AI from ever training on your data. A written governance program — model inventory, bias testing, human-override tracking, audit logs of every prompt and output — built to answer the questions CDI and TDI market-conduct exams will ask. Cheap to build now, expensive to retrofit under exam.
The timeline
We build this system the same way it works: with supervised AI agents doing the heavy lifting and senior engineers holding the gates. The component technologies are already proven in production at carrier scale — the work is orchestration and tuning to your book, and that is exactly what AI-accelerated engineering compresses best. You’re looking at the proof: this site, including the working Sentinel simulation above, was built that way in days. And every phase gates on measured accuracy, not the calendar — if Phase 0 says the Sentinel isn’t catching every demand yet, Phase 1 waits until it does.
Shadow pilot on closed files
You give us 50–100 closed litigated files. We run the system against history: every demand the Sentinel would have caught and how many days earlier; every file triage would have flagged before it blew up; drafted answers blind-graded by attorneys against what was actually filed; valuation calls scored against actual outcomes.
What you get
A dashboard of your own files, scored — the ROI math in this pitch recomputed with your data instead of vendor claims. Zero live risk. Zero regulatory exposure.
Honest fine print: speed comes from how we build — never from skipping a gate. Every document still passes through licensed counsel; every settlement decision still belongs to your adjusters.
The questions
We can research case law and benchmark vendors all day. What we can’t know is what your book actually does. Five questions decide whether the economics on this page are real — and only people inside Bridger can answer them:
How many files go to litigation each year, and what does the average one cost to defend? (We guessed 200 and $13K — the calculator above is waiting for the real numbers.)
What share of serious claims settle pre-suit versus in litigation — and does that split look different in California than in Texas?
Has Bridger ever faced a disputed §999 or Stowers demand, or an excess-verdict scare? What did the file look like the week the demand arrived?
On the Texas book, who actually holds settlement authority — Bridger, or Redpoint?
If the answer-to-discovery pipeline cost a third of what it does today, would you litigate files you currently settle?
The shadow pilot replaces every industry figure on this page with your files, scored against what actually happened — every demand caught or missed, every valuation called right or wrong. Here is everything it takes to start:
The fuel for the shadow pilot. Closed files only — no live claims, no regulatory exposure, nothing leaves enterprise infrastructure.
One senior adjuster who knows where the bodies are buried, a few hours a week, to ground-truth what the system finds.
Automation savings only reach Bridger if fees are restructured to pass them through. Flat-fee volume is the cleanest path — worth aligning on early.
Who’s asking
Attorney and entrepreneur. The licensed-counsel half of this proposal: every document that would ever reach a court or opposing counsel goes through his review and carries his signature, not an AI’s.
Engineer and product builder. The system half: the agents, the audit trail, the human gates — including this site and the working Sentinel simulation above, built the same way Rampart AI would be.
Neither of us will pretend to know the non-standard auto business better than the people in your building. We know what this class of system can do; you know whether Bridger needs it. The pilot is how we both find out, cheaply.
The plaintiff bar didn’t wait for permission to bring AI to your claims desk. The only question is how long the asymmetry lasts.
Sean Svendesen & Matt Soldo · matt@soldo.org