We turn submission emails, ACORD forms, schedules and loss runs into structured underwriting data — inside Acturis, Open GI or whatever you already run. Every field is reviewable. Every change is logged.
Fields appear here once the submission is processed.
The 92%-confidence field is flagged, not silently accepted. That distinction is the whole design.
A submission arrives as an email with five attachments in three formats, none consistent between brokers. Someone opens each one, decides whether it's in appetite, and re-types the key fields so an underwriter can price it. Then chases the four fields the broker left out.
Twenty to forty minutes per submission, from inbox to system of record. Multiply by weekly volume and it is usually the largest hidden cost in the operation.
In specialty lines, the first credible quote often wins. Slow triage loses business you were always going to write.
More capacity means more admin hires. That link is what automation is supposed to break — and rarely does, because most projects stop at the demo.
Nobody should sign a five-figure automation project on a stranger's word. So we start with something small enough to approve this week, and priced so that stopping after it costs you very little.
You send 20–50 anonymised submissions. We map how intake works today, test extraction on your real documents, and give you a written report: the field taxonomy, measured accuracy per field, the integration route into your system, and a fixed price for a pilot. If the numbers don't justify automating, the report says so.
One narrow workflow, live end to end: classification, extraction, a review queue your team actually uses, export into your management system, and a full audit log. Measured against accuracy and handling-time targets agreed before we start.
More document types, more users, exception handling, monitoring, security controls and the integration work that makes it survive renewal season.
New broker formats appear constantly. We add document types, tune accuracy, monitor drift and stay on the hook when something breaks at 9am on a renewal day.
You already own a broker management system. It stores the data properly. The gap is everything that happens before the data gets there — and that gap is integration work, workflow design and exception handling, not a new product to buy.
Saying this up front saves us both a wasted call.
You operate in a regulated environment and you're being asked to hand documents to an external party. These are the standing terms — they apply from the first sample document, not from contract signature.
Discovery runs on redacted or anonymised documents. We don't need real insured names to measure extraction accuracy.
Storage and processing stay in UK or EU regions. Data residency is confirmed in writing before any document is sent.
Your documents are never used to train or fine-tune models, and are never shared with third parties beyond the named processors in the agreement.
Documents are deleted at the end of each engagement, with written confirmation. Retention beyond that only happens if you ask for it.
We sign yours. If you don't have one to hand, we provide a GDPR-aligned DPA covering roles, sub-processors, transfers and breach notification.
Every extracted value records its source document, page, confidence score, and who approved or amended it. Nothing enters your system unattributed.
Thilak Kumar A — software engineer, Chennai, India. I build the systems I sell. You will not be handed to a delivery team you've never spoken to, and there is no sales layer between you and the person writing the code.
My background is the two halves this problem needs: production AI document automation, and years of keeping ageing enterprise .NET systems alive. The second half is the one that matters most — extracting fields from a PDF is the easy part; getting that data safely into a system built fifteen years ago is where these projects usually fail.
It stores the data properly — that's its job. What it doesn't do is read an inconsistent broker email with four attachments and decide what goes where. We automate getting the data in. We integrate with your system; we never propose replacing it.
It depends entirely on your document mix, which is why we measure it on your documents during discovery rather than quoting a marketing number. Clean structured forms extract very reliably; scanned handwriting and inconsistent spreadsheets are harder. Every field carries a confidence score, and anything below your chosen threshold goes to a person. The goal is a large cut in handling time with full oversight, not removing humans.
It gets caught in the review queue before it reaches your system — that's what the queue is for. Low-confidence fields are flagged, the source page is shown next to the value, and the reviewer amends it in one click. Every correction is logged, and recurring errors feed back into improving extraction.
Calls happen in UK afternoons, which is our evening. Written updates go out weekly whether or not there's news. Contracts, invoicing and data handling follow UK norms. The cost base is the reason a discovery is £3,000 and a pilot starts at £8,000 — that's the advantage being passed to you, not a corner being cut.
The discovery is the trial — run on your real documents, with a measured accuracy report you keep whatever you decide next. A free demo on sample data proves nothing about your workflow, and any supplier willing to do weeks of unpaid work is telling you something about how they'll price the rest.
You do. Source code, documentation, prompts and configuration transfer to you on final payment. No per-seat licensing and no dependency that makes leaving expensive.
Discovery delivers a written report in two weeks. A pilot puts a working workflow in front of your team within four to eight weeks of starting. If someone promises you a production intake system in a fortnight, they haven't seen your loss runs.
Tell us how submissions arrive, who touches them and where the re-keying happens. If there's a case for automating it we'll say so, and if there isn't we'll say that too. No deck.