Summary
"AI claims processing" is used as one phrase for three jobs. Most automation does the first well and stops. The value, and the risk, sit in handling and adjudication.
- Three jobs, three risk profiles. Processing is administrative; handling is the human layer; adjudication is the policy judgement.
- STP is earned, not switched on. Automate approvals only after shadow-mode agreement on real claims; never automate the decline.
- UK rules allow automation with safeguards. Articles 22A–22D (since February 2026) plus the FCA Consumer Duty: explain decisions, keep a person reachable, design out foreseeable harm.
- Six vendor tests. Citations to the page, shadow mode, overrides as rules, thresholds you control, no auto-declines, and your cloud with no training on client data.
Processing, handling, adjudication: three words, three jobs
Claims processing is the administrative path: read the file, extract the facts, check cover and amounts against a tariff (the schedule of approved rates or fee bands for a claim type), chase what is missing, and route the claim. No policy judgement is required, yet it often consumes the time of people hired for judgement.
Claims handling is the human layer: ownership, review, claimant contact, escalation and the audit trail. Regulators look here when things go wrong, because this is where the customer meets the insurer.
Claims adjudication is the decision itself: does this policy respond to this loss, in what amount, under which conditions. It carries legal and financial consequence, and "the AI decided" is not an acceptable sentence.
| Job | Definition | Automation risk | Safe automation rule |
|---|---|---|---|
| Processing | Complete and route the claim record | Low | Automate freely; log every request and check |
| Handling | Human review, correction, escalation | Medium | Tool the handler; keep a clear audit trail |
| Adjudication | What the policy says, and whether to pay | High | Cited recommendations; approvals only inside thresholds; never automate the decline |
A vendor who uses the three words interchangeably has not said where the decline line sits.
What straight-through processing means in insurance claims
Straight-through processing, STP, is a claim that moves from notification to payment with no human touch at any step. The term comes from securities settlement and keeps its original meaning: not faster review, but no review.
Industry STP figures vary widely and mostly come from vendors: around a third for standard personal lines in 2025–26 surveys, and 70% or more in vendor case studies on narrow types such as travel or motor glass. The spread is about design, not technology. A claim type with a defined tariff, a verifiable claimant and a bounded payout can be automated almost entirely; a claim that needs policy interpretation cannot, and should not.
Three conditions make STP safe:
- Defined band: claim types, amount ceilings, and referral triggers set in advance.
- Approvals only: an automated yes inside the rules is a service improvement; an automated no is a regulatory event.
- Measured agreement: shadow mode on real claims first, then ongoing sampling that handlers re-decide.
Our underwriting analysis works through the same logic for applications, with the arithmetic for the threshold at which automation pays.
Claims automation versus agents
"Claims automation" usually means rules engines, RPA and document extraction. Each follows a script and stops at the first exception. Most handler time sits in that exception share.
An agentic system reasons across documents, requests what is missing and cites its reasons. It flags conflicts, applies the policy pack by clause number, and can turn a handler correction into a rule. Agents can still be confident when wrong, so safety sits in thresholds, shadow mode, checkable citations and a human on every decline.
How AI claims processing works: six steps
Work the claim in six steps. Automate the reading and the routine approvals; keep every decline and every edge case with a handler. The pipeline below is the one we have run in production for a UK healthcare trust; the full account is in the AI claims processing case study.
UK surveys put claims handling at the top of the AI agenda and show how little of it is yet fully automated. In RDT-commissioned surveys of insurers and MGAs reported by Insurance Times (fieldwork late 2024 and early 2025), 67% named claims handling as a top target for human-led AI, yet only 9% reported fully automated claims processes. Specialist vendors publish higher straight-through rates on narrow lanes: Tractable, for example, states that 70% of claims reviewed in its Claim Review product can proceed without human involvement, alongside an 8-day cycle-time reduction from FNOL triage and a 50% cut in estimate-writing time with estimatics pre-population (Tractable, insurers). Those figures are vendor-reported and apply to the lanes they serve; they are not industry averages.
- Intake and digitisation. Ingest PDFs, emails, scans and portal uploads into a structured claim record. No judgement yet; the goal is a complete, searchable file.
- Validation and enrichment. Check membership, cover and tariff schedules. Request missing documents automatically and log every chase.
- Risk and anomaly scoring. Score tariff variance, anomalies and complexity with reasons a handler can read. Thresholds stay with the firm.
- Cited recommendation. Draft the coverage position and amount against the policy pack, with clause and page citations. Approvals may settle inside pre-agreed bands; every potential decline is referred.
- Routing and escalation. Send referred claims to the right handler with the evidence bundle, risk profile and draft assessment already attached.
- Audit and learning. Log every step for information governance. Promote handler overrides to rules after senior approval, then re-measure agreement in shadow mode.
| Step | Typical automation | Human required |
|---|---|---|
| 1–3 Intake to scoring | High | Exception review only |
| 4 Cited recommendation | Approvals inside thresholds | Every decline; every edge case |
| 5 Routing | Assignment and pack assembly | Handler decision on referred claims |
| 6 Audit and learning | Logging and proposed rules | Senior approval of new rules |
The agent's job is that no claim reaches a handler cold, and no routine approval waits behind one that needs judgement.
Healthcare and medical claims
Health claims are harder because documents are clinical, tariffs are detailed, and the data is special category under UK GDPR. They are also high volume: routine, well-documented claims dominate handler cost. Keep automated settlement inside narrow bands; route clinical judgements, coverage questions and every potential decline to a person, with cited sources attached.
What UK rules allow
Two frameworks govern AI claims decisions in the UK, and both changed recently.
The Data (Use and Access) Act 2025 replaced Article 22 of UK GDPR with Articles 22A to 22D, in force since 5 February 2026. Solely automated decisions with legal or similarly significant effects are permitted with safeguards: notice, representations, human intervention and the right to contest. Where special category data is involved, including health, the permission is narrower. See the ICO guidance on automated decision-making; its 2026 consultation on the new provisions is worth checking in final form.
The FCA Consumer Duty requires good outcomes and the avoidance of foreseeable harm. The FCA has repeatedly singled out claims handling. Efficiency gains must not make claims slower, more opaque or more burdensome, and firms must evidence that automated processes do not produce unfair outcomes.
Put together, the rules do not forbid AI claims processing or even automated settlement. They require that the customer can reach a person, that the firm can explain each decision, and that the firm has checked the system does not harm particular groups. A design that automates approvals inside thresholds, refers every decline to a handler and keeps cited reasons for every recommendation meets all three without a legal argument.
The EU AI Act adds a separate layer for firms with EU reach. Claims handling is not named as high-risk there, as covered in Does the EU AI Act apply to UK insurers?
Judging AI claims management software
Six tests separate a claims engine from document extraction with a chatbot on top. Put each to any vendor, including us.
- Citations to the page. Every recommendation should link each reason to the document and page it came from, so a handler can check it in seconds rather than re-read the file.
- Shadow mode before trust. The engine should run alongside your team on real claims, with agreement measured, before anything is automated. Ask for the agreement figure and how it was computed.
- Overrides become rules. When a handler corrects the engine, the correction should be capturable as a rule, with senior approval, so the system learns your book rather than a generic one.
- Thresholds you control. Claim types, amount ceilings and referral triggers should be yours to set and change, not configuration the vendor owns.
- No automated declines. The engine should be unable to decline a claim without a person, as a property of the design rather than a setting.
- Your cloud, your data. Deployment in your own AWS, Azure or Databricks environment, and a contractual commitment that your claims data is not used to train models.
Our own product, Claim It Quick, is built around exactly these six, and the product film shows each one on anonymised real claims.
Putting this into practice
Start with a fortnight of measurement, not a procurement. Pick two claim types (one clean and high-volume, one that regularly bounces) and record where handler time goes: reading, chasing documents, reconciling sources, deciding, writing up. That split shows how much an agent can take without touching a decision.
Then run the engine in shadow mode for a month. Agreement points to a candidate automated band; disagreement points to a rule to add or a boundary to draw. Only then talk about straight-through processing, and only for approvals.
To run that exercise with us, book a demo and tell us which claims you handle.
Frequently asked questions
What is AI claims processing?
Using language models and agents to take a claim from arrival to a validated record: reading documents, extracting facts, checking cover and tariffs, and requesting what is missing. It is the first stage of a claim's life, and the one most automation stops at.
What is the difference between claims processing and claims adjudication?
Processing is administrative: getting the claim complete and routed. Adjudication is judgement: deciding what the policy says about this claim and whether it is paid. Automation of processing is low risk and high volume. Automation of adjudication needs citations, thresholds, shadow-mode evidence and a human on every decline.
What is straight-through processing in insurance claims?
A claim settled with no human touch at any step, from notification to payment. It is only safe inside explicit thresholds, only for approvals, and only after the engine's recommendations have been measured against handler decisions on real claims.
Can AI decline an insurance claim in the UK?
Articles 22A to 22D of UK GDPR (from the Data (Use and Access) Act 2025, in force since February 2026) allow solely automated decisions with significant effects only with safeguards, and more narrowly for special category data. The FCA Consumer Duty adds a duty to avoid foreseeable harm. In practice, declines should stay with a handler, with the engine's reasons on file.
How is agentic AI different from claims automation or RPA?
Agentic means agent-style AI that reasons across documents, requests what is missing and cites its reasons. Rules engines and robotic process automation follow a script and stop at the first exception. The gain is not a faster happy path; it is that exceptions, where handler time actually goes, get handled.
When is straight-through processing safe?
When three conditions hold: the automated band (claim types, amount ceilings, referral triggers) is defined in advance; automation applies to approvals only; and agreement with handler decisions has been measured in shadow mode on real claims, with ongoing sampling afterwards. Declines, referrals and fraud flags stay with a person.
What should we look for in AI claims management software?
Citations to the page for every recommendation, a shadow mode that measures agreement with your handlers before anything is automated, logging of overrides so corrections become rules, thresholds you control, deployment in your own cloud, and a contractual commitment that your claims data is not used to train models.