Contents
Agentic claims AI is the hottest thing in insurance right now. Touchless settlement, autonomous handling, FNOL-to-payout with no human in the path: it is on every conference agenda and in every core-vendor roadmap for 2026. The race is on to let the machine decide the claim.
We took the opposite bet. SwiftCase lets AI run the operation, and keeps a human owning the money. That sounds like the cautious choice. I would argue it is the only version a regulated insurer can actually deploy at scale, and I want to explain why.
Deciding a claim and running a claim are different jobs
Most of what gets called claims AI today is one of two things.
The first is a point tool that automates a slice: damage assessment from photos, document extraction, fraud scoring, a smarter estimate. These are genuinely good, and they make one step faster. But a faster step is not a faster claim. The case still has to be routed, chased, communicated, approved, settled and evidenced, and the point tool does none of that.
The second is a bot that auto-settles the simple cases. On a clean, low-value, personal-lines book, that can work, and it demos beautifully. But most of the claims that keep a board awake are not clean, low-value or simple. They are non-standard, disputed, multi-party, or regulated, and "the AI paid it in two seconds" is precisely the sentence nobody wants to read in a complaint file.
Running a claim is a third job, and it is the one almost nobody builds, because it is the hard one. It means being the operating system the whole claim moves through: every channel in, every step to settlement, the audit trail, the approvals, the escalations, the human-in-the-loop on every decision that involves money. That is what we built, and it is what "AI runs it, a human decides" actually requires.
Why "the AI decided it" is a liability, not a feature
In a regulated claim, the moment a machine generates a value, negotiates an offer, or settles a case, you have created a record with no named human owner. When that case becomes a complaint, and some of them always will, it can land at the Financial Ombudsman, and the Ombudsman asks for the file.
A file that says "the AI agreed the value" does not hold up. There is no accountable person, no evidence the customer was treated fairly, no defensible record of what was disclosed. Under Consumer Duty and SM&CR, that is not a technology gap, it is an accountability gap, and it is the firm's problem, not the vendor's.
So the design principle is simple and absolute. The AI never generates or negotiates a value. A named engineer sets it. The AI presents it with comparable evidence shown in both directions, captures the customer's response as structured data, and escalates any dispute back to that engineer. Every action is attributed to a named human in Timeline, our immutable audit logger. We wrote about this design in detail in why our AI never agrees a total-loss value, and about the philosophy behind it in human-in-the-loop AI for claims.
Fast where it is safe, human where it matters, audited everywhere
Keeping a human on the decision does not mean keeping a human on everything. That would defeat the point. The discipline is knowing which is which.
The AI runs the high-volume, low-judgement work at machine speed and machine scale: it reads the mess and opens the case, triages it, verifies identity, looks up status, updates fields, chases suppliers, and drafts communications. The human owns the decisions that carry money or judgement: the total-loss value, the settlement, the complaint outcome, the approval to send. And everything, AI action and human decision alike, is written to an audit trail that answers to a regulator by design.
Fast where it is safe. Human where it matters. Audited everywhere. That is a sentence a Chief Risk Officer can sign, which is more than can be said for "touchless".
On top of your core, across every channel, live now
There is one more reason this version is the deployable one. It does not ask you to rip anything out or wait for a roadmap.
The platform runs on top of whatever core system you already have. The agent layer works across voice, SMS, WhatsApp, web chat and email from a single definition, so you are not standing up five inconsistent bots. And it is provider-agnostic, failing over between model providers mid-conversation, so a single model outage does not drop a customer at 2am. We cover that in AI resilience.
That matters in 2026 specifically. A core-vendor agentic feature is built for the average of a thousand insurers, bolted to one core, usually claims-only and single-channel, and you are in the deployment queue. A platform that sits on top of your core can pilot this quarter, on the workflow you choose, across the channels your customers actually use.
This is in production, not on a slide
We are not describing a research project. More than 11.8 million cases have run through the platform since 2015, across 40,000 users and seven industries, with claims the deepest. A UK specialist motor insurer runs more than 600 of its people on it; you can read the specialist motor insurer case study. And at Laird Assessors, the agent layer handles out-of-hours intake, supplier chasing and total-loss conversations today, with the engineer-sets-value design described above.
UK-hosted. Cyber Essentials certified. Consumer Duty by design. None of which is exciting on a slide, and all of which is the reason a regulated operation can actually run on it.
The bet, stated plainly
Everyone is racing to let AI decide the claim. We built the platform that runs it, and keeps a human deciding. If the market is right and touchless settlement turns out to be safe and defensible at scale across complex, regulated books, we are one configuration change away from more automation, because the operating system is already there. If the market is wrong, and a lot of regulated insurers are quietly betting it is, then the operations that kept a human on the money will be the ones still standing.
Either way, the right place to start is a 30-day pilot on your noisiest workflow, measured against your own numbers. We brought this argument to the ILC ClaimsTech 2026 final, and gathered the proof in one place: the ClaimsTech hub.
Further reading:
- Why our AI never agrees a total-loss value: the human-in-the-loop design
- The AI that talks is the easy 20%: why the platform, not the conversation, is the project
- One platform, both sides of the claim: the thesis behind how we built this
- Human-in-the-loop AI for claims: the design philosophy, end to end
- The ILC ClaimsTech 2026 hub: the platform behind the pitch