Get Your Accuracy Number
For insurance brokers & insurers

Cover terms captured once. Not typed twice.

Sums insured, deductibles, sub-limits and clause references pulled straight from the schedule, with a confidence score on every field. Proven on 500 of your own schedules in 14 days.

Fixed pilot fee. Below the field accuracy we agree together, you pay nothing. No per-schedule pricing, no annual licence.

What rekeying cover terms really costs

The schedule is already the source of truth. It just isn't machine readable.

Every renewal, every mid-term adjustment, every new binder: someone reads a schedule and retypes the numbers into a system that will be used to advise a client. Work out how many times that happened this year.

8–15 min

per schedule to key sums insured, deductibles, sub-limits and clause references, longer where the wording sits in an appendix.

1–4%

keying error rate on limits and deductibles. A misplaced digit in a sub-limit is not visible until a claim tests it.

3 systems

the same figures typically get typed into: the broker system, the client presentation and the renewal spreadsheet.

1 person

senior enough to read a schedule correctly, doing work that does not need their judgement.

Do the math with your numbers

Three fields. This is what rekeying cover terms costs you per year.

Per month€3,600
Per year€43,200
The pilot that ends it€4,000 once

Excludes the errors-and-omissions exposure created when a mistyped limit reaches a client summary, which is the real reason this work stays with senior people.

The mechanism

How a schedule becomes structured cover data

A schedule is not a form. The same cover appears as a table on one insurer's paper, as prose in another's, and as a clause reference in a third's. Extraction has to understand the cover, not the layout.

Locate the cover sections

Property, liability, business interruption, extensions and exclusions are found wherever they sit, including in appendices and endorsement pages attached after the main schedule.

Normalise the terms

Limits, sub-limits, deductibles and indemnity periods are read with their basis: per occurrence or aggregate, each and every claim or annual. A number without its basis is not captured, because a number without its basis is a liability.

Score and flag

Every field carries a confidence score. Anything ambiguous, including cover that is only referenced by a clause code, is flagged for a human rather than being silently normalised into something tidy.

What this looks like on a real schedule

A commercial combined schedule on the left. The structured cover data on the right. Illustrative example with fictional data.

Schedule of Insurance · Period 01.09.2026 to 31.08.2027
Policy MI-CC-449012
Insured: Calder Foods Ltd
Issued: 22.08.2026
SectionCoverLimitExcess
ABuildings, reinstatement basis4,600,0002,500
BStock, incl. deterioration980,0005,000
CBusiness interruption, 24 months2,200,00072 hrs
DPublic liability, per occurrence5,000,0001,000
Extensions per clause set MI-14
Deterioration sub-limit 150,000
Premium 18,940.00
Subject to the terms of clause set MI-14, supplied separately · refrigeration warranty applies to Section B
Extracted & normalised
Buildings sum insured4,600,000 reinstatement basis99.6%
Stock sum insured980,00099.4%
BI indemnity period24 months, limit 2,200,00099.2%
BI time excess72 hours time-based, not monetary98.6%
PL limit5,000,000 per occurrence99.5%
Deterioration sub-limit150,000 sits under Section B97.9%
ExtensionsClause set MI-14 referenced but not attached43%
⚠ The extensions are named by clause code only and the wording is not in this document → captured as a reference and flagged, never guessed from the clause code. Your team is told exactly what is missing.
Structured record written
MI-CC-449012 · Bldgs 4.6m · Stock 980k · BI 2.2m/24m · PL 5m · extensions pending wording

The offer

The Policy Data Intake Pilot

One fixed fee. Fourteen days. Your real schedules from your real insurers. A field-level accuracy number that tells you, and us, whether the full build is worth doing.

Not a demonstration on a clean specimen schedule. The pilot ends with your own documents parsed and a field-by-field score you can challenge line by line.

  • 500 of your real schedules processedAcross the insurers you actually place with, including the ones whose schedules run to forty pages.
  • Field-level accuracy reportLimits, deductibles, indemnity periods and clause references scored separately, because they fail independently.
  • Basis-of-cover analysisWhere the document states a number without stating its basis, and how often that happens per insurer.
  • Working write path into your systemA test batch written end to end, so you see the structured output land in your own environment.
  • Fixed-price quote for the full buildScoped from measured results on your schedules. Valid 90 days.
  • ROI calculation on your own volumesYour schedule count, your handler cost, your rekeying error rate. If the numbers do not justify the build, that conclusion goes in the report.
€4,000 fixed

Against a problem costing most clients €40,000–70,000 a year.
Credited in full against the build if you proceed within 90 days.

Get Your Accuracy Number

Your side of the pilot: about two hours, total

45 minkickoff call to agree the accuracy threshold and pick the sample
10 minforward an email folder or grant read access to the document inbox
60 minresults session: the accuracy report, the exceptions, the quote

No software to install, no workshop series, no project team on your side. Your people keep working; we work on copies.

The Accuracy-or-Free Guarantee

Before we start, we agree the accuracy threshold that makes this worth deploying. Together, in the engagement letter. If the measured result on your documents comes in below it, the pilot is free. We can offer this because we scope honestly: if we don't think your document mix can hit the number, we decline the pilot and tell you why.

Pilots run one at a time per engineer, so they start in the order requests arrive. Current lead time: about two weeks

Your actual alternatives

What else you could do with this problem

Do nothingHire another technicianVertical SaaS productBraynex pilot → build
Captures limits with their basis✕ whatever gets typed✓ if trained well✕ number only, basis lost✓ basis captured or flagged
Cost over 3 yearsrising with renewals€120–170k+per-document pricingknown, fixed, quoted
Handles a forty-page schedule✕ slowly✓ slowly✕ first pages only✓ including appendices
Proof before commitment-✕ demo on their specimen✓ measured on yours
Flags cover it cannot see-✓ usually✕ silent omission✓ named and scored

Why believe us

We build document systems for businesses that cannot afford to be wrong

Medical / Legal

MedChrono

AI that converts medical records into legally defensible timelines. Accuracy-critical, regulated, every extraction traceable to its source page.

Financial advisory

Advisory documentation

Client meeting recordings become compliant advisory documents automatically, adopted across the practice for every client meeting.

Vision AI at volume

Sightline

Production vision-AI moderation for marketplace listings: real ML in production at volume, not an API wrapper.

Swiss entity (Zug) · EU-region hosting · GDPR & revised FADP compliant · named senior engineers you meet before signing

Fair questions

The things you should be skeptical about

What happens when cover is only referenced by a clause code?

It gets captured as a reference and flagged as incomplete, which is what happened in the worked example above. Guessing what a clause code contains would be the single most dangerous thing a system like this could do, so it does not. If you supply the clause library, we can resolve them; if not, the gap is visible.

Is this good enough to feed a client-facing summary?

Not on its own, and we would advise against wiring it that way at first. The realistic pattern is that extraction removes the typing and a human confirms the fields that carry advice risk. The accuracy report tells you which fields are safe to trust and which are not, which is a better basis for that decision than a vendor assurance.

Why not just ask insurers for structured data?

Worth doing, and where it exists you should use it. In practice the structured feeds cover your largest insurers and the schedules keep arriving from everyone else. This is for the everyone-else, which is usually where the manual time actually sits.

What if we'd rather buy an off-the-shelf product?

Sometimes you should. If the pilot shows a market product handles your insurer mix, the report says so and we part as friends. The fixed fee buys the answer, not a predetermined recommendation. A build gets recommended only where the evidence earns it.

Where does our data go?

EU-region hosting, a GDPR-compliant processing agreement, and a contract with our Swiss entity in Zug. Nothing trains a public model. Your own compliance lead can review the setup before the pilot starts, and the pilot sample can be narrowed to a subset you are comfortable sharing.

Why should a company our size get your senior engineers?

Because we are structured for exactly this engagement size. The large consultancies staff €60k projects with juniors or decline them; we deliver them with the people you actually meet. That is the trade we have built the company on.

Start here

Claim a pilot slot

Tell us how many schedules you process and which insurers dominate. An engineer replies within one business day and tells you plainly whether your insurer mix is a good fit or a poor one.

FS

Florian Shabani

Lead Engineer · runs your pilot

The people on this page are the people on your project. No handover to a delivery team you never met: the engineer who scopes your pilot is the engineer who builds it.

One pilot at a time per engineer · current lead time: ~2 weeks

No newsletter, no drip sequence. One reply from an engineer, within one business day.