The Subrogation Money Insurers Leave on the Table

Subrogation is the least glamorous word in insurance and one of the most profitable. It's the process of recovering money from the party actually responsible for a loss, after you've paid your customer. Done well, it flows straight to the bottom line — recovered money is nearly pure margin.

Most insurers leave a lot of it on the table. Not because they don't try, but because the opportunities are identified too late, by the wrong people, from incomplete information. It's a data problem, and it's an unusually clear one.

Why subrogation is a timing game you're losing

Recovery potential decays fast. The evidence that establishes another party's liability — the scene, the witnesses, the vehicle, the maintenance records, the CCTV — degrades or disappears within days. The other party's insurer gets less cooperative once they've closed their own file.

So the single biggest lever in subrogation is identifying the opportunity early, ideally at first notice of loss. And that's exactly where the current process fails.

In most carriers, subrogation potential is spotted late — by a subrogation specialist reviewing files after settlement, or by an adjuster who happened to notice and happened to flag it. By then the trail is cold on a chunk of recoverable claims.

Why the opportunities get missed

The signal is in the narrative, not a field. Whether a third party is liable is usually described in the FNOL text — "the other driver ran the light", "the contractor left the valve open", "a falling branch from the neighbour's tree". It's rarely a structured checkbox. So it's invisible to any system that only reads fields, and it depends entirely on a human reading and reacting.

The adjuster isn't incentivised on it. Their job is to settle the customer's claim fairly and fast. Subrogation is someone else's department and next quarter's problem. Flagging it is unrewarded extra work.

The context needed to judge liability is scattered. Prior claims, policy details, the other party's information — spread across systems, so even a diligent adjuster can't easily assemble the picture at speed.

Where AI genuinely fits

This is one of the cleanest AI use cases in insurance, because the task is well-defined: read the FNOL, decide whether a third party is plausibly liable, and flag it — at intake, automatically, on every claim.

1. Read the narrative. Language models are good at exactly this: extracting "was another party involved and potentially at fault?" from free-text descriptions. This alone recovers the claims that used to slip past because nobody read them as subrogation candidates.

2. Score recovery potential. Not every liable third party is worth pursuing — recovery cost matters. Combine the liability signal with claim size, party identifiability, and historical recovery rates for similar cases to rank opportunities.

3. Trigger preservation immediately. The moment a claim is flagged, kick off evidence preservation — request the CCTV, document the scene, notify the other insurer — while it still exists. The flag is worthless if it doesn't trigger fast action.

4. Route to the right people. Get flagged claims to subrogation specialists at FNOL, not post-settlement.

The honest constraints

It flags candidates, not certainties. Liability is a legal judgment. The AI's job is to make sure a human specialist sees every plausible case early — not to decide recovery. False positives here are cheap (a specialist glances and dismisses); false negatives are the expensive ones you're trying to eliminate.

It needs the narrative data accessible. If your FNOL text is locked in a system the model can't read, or arrives as a scanned form, you're back to the document problem first.

Recovery still requires the operational muscle. AI finds and prioritises; people and process recover. If your subrogation team is under-resourced, you'll flag opportunities you can't pursue. The data unlocks the potential; the organisation has to realise it.

Why this one is worth doing first

Among insurance AI use cases, subrogation has an unusually clean business case: recovered money is nearly pure margin, the task is well-scoped, and you can measure the lift precisely — recovery rate before versus after, on comparable claims.

Start narrow: run a language model over last year's FNOL narratives and flag the ones with third-party liability signals. Compare against what your team actually pursued. The gap between "flagged by the model" and "recovered by the team" is money you already had a right to and didn't collect — and it's usually a startling number.

Subrogation is the rare place where fixing the data doesn't just cut costs; it recovers cash you're legally owed and currently leaving behind.

We build the FNOL data pipelines and context that make early subrogation detection possible. More at IntelliBooks.