Evaluation · Insurance
Document processing and evaluation for insurance
A claim file is usually a disordered pile of paperwork: invoices, reports, photographs, statements. Zero-Shot builds systems that read those documents into fields, list what is missing and flag inconsistencies. The decision stays with the adjuster; the system only makes the file readable.
What we build
- 01Field extraction from invoices, statements, survey reports and policy documents
- 02Missing-document detection based on what each claim type requires
- 03Cross-document inconsistency flags for dates, amounts, plates and identity fields
- 04Per-document-type accuracy reporting and a review queue for low-confidence fields
- 05A file summary that speeds up the adjuster's reading — a summary, not a decision
How it runs
01
Inventory the document types
Which documents arrive for which claim type. Accuracy targets are set separately for each type.
02
Per-field targets
Thresholds are high for critical fields like amounts and dates, lower for free text. No single average accuracy is used.
03
Extraction first, summary later
Only field extraction goes live at first. The file summary is added once extraction clears its threshold.
04
Adjuster feedback
Every field an adjuster corrects goes back into the set, and corrected fields are prioritised for improvement.
What we measure
- Field extraction accuracy per document type and per field
- Miss rate on missing-document detection
- File preparation time, from arrival to reaching the adjuster
- Share of fields requiring adjuster correction
- False alarm rate on inconsistency flags
When this is the wrong fit
The system makes no claim decisions, issues no payment approvals and interprets no policies. Those stay with the adjuster and with regulation. We do not take on builds intended to decide claims automatically — neither the regulatory position nor the outcomes support it.
Common questions
Does the system approve claims automatically?
No. It reads documents, lists what is missing and flags inconsistencies. Approval, rejection and amount stay with the adjuster. The boundary is architectural: the system has no write access to decision fields.
Do you assess damage from photographs?
Visual damage assessment is a separate piece of work, measured separately. It is possible, but accuracy varies sharply by damage type and it is never used as a decision on its own. Here, images enrich the file the adjuster reads.
How is personal data protected?
The system can be deployed inside your infrastructure so documents are processed without leaving it. Which fields are stored, for how long and who can see them is written down before setup, and access is logged.
Other services
What we've written on this
- Where to start with enterprise AIHow to choose a first AI project: which tasks qualify, in what order to proceed, and what to expect in the first eighty days.
- Writing prompts for productionWhy a prompt that works in a demo breaks in production. Treating prompts like code: structure, versioning, boundaries and measuring every change.
- Why AI makes things up, and how to reduce itWhy language models state false things confidently. The real cause of hallucination, which countermeasures work, and which do not.
Last updated: 2026-09-05
Let's talk about this one
Tell us what you want to build and within two days we'll tell you who takes it and how we'd measure it.