The RFQ is decision-readybefore anyone opens it
A broker’s e-mail with a dozen attachments becomes a structured case: extracted data, a risk score, a coverage recommendation and a list of missing documents. The underwriter steps in at the end — and decides.
An RFQ doesn’t arrive as a form. It arrives as a mess.
In commercial insurance, a request for quotation is correspondence: an FW: chain, a dozen files in mixed formats, scans with no text layer, encodings from two decades ago.
40 minutes before work even starts
Someone reads the e-mail, opens the attachments and re-types the same data into the system. Only then does the actual underwriting begin.
A bottleneck at volume
At dozens of enquiries a week, manual triage delays the answer to the broker — and in risk placement, response time is often what decides.
Material that breaks tools
97% of enquiries are FW: chains, one in three PDFs is a scan without a text layer, ZIP and 7z archives, iso-8859-2 encodings. Generic OCR/RPA tools don’t survive this.
Four pillars: from inbox to decision
LynxFlow plugs into the mailbox (IMAP, webhook or upload) and takes every enquiry down the same track — all the way to a decision-ready case.
Enquiry analysis
Attachment classification, OCR on scans, extraction of ~130 fields per product schema, tax-ID checksum validation with official registry data pulled in, and a completeness checklist.
Underwriting
A risk profile from factors (industry codes, loss history, registry data), per-tenant risk appetite and rules — a 0–100 score and a suggestion: accept, refer or decline. With reasons.
Risk coverage
A clause catalogue built from real slips, product-by-product and insurer-by-insurer coverage comparison, a recommendation with gaps and sums insured.
Client service
A ready-to-send draft e-mail about missing documents, a case timeline, SLAs with deadlines, and a “talk to your data” assistant — plain-language questions about cases and decisions.
LynxFlow prepares the decision — it doesn’t make it. The system does the triage and the data completeness; the underwriter has the final word. Human-in-the-loop isn’t a limitation, it’s a compliance requirement we take seriously.
What you get as standard
Reads everything that arrives
PDF, XLS, DOC, ZIP/7z archives, scans without a text layer, legacy encodings — classification and OCR hardened on real correspondence, not just clean files.
Data from official registries
Checksum validation of company identifiers, with the official name, address and industry codes pulled from registries overriding whatever the e-mail said.
Triage with reasons
A risk engine: accept / refer / decline with explanations — driven by rules and a risk appetite configured for your organisation.
Coverage comparison
A clause catalogue, coverage mapping across products and insurers, and search over policy wordings. Calibrated on your documents during onboarding.
Complete file or a list of gaps
The system knows what each product requires — it lists the missing documents and drafts the e-mail to the broker itself.
Multi-tenant from day one
One installation serves an insurer, a broker and an MGA — each sees only its own product catalogue and its own data.
Three perspectives, one platform
The same case looks different from each side of the market — LynxFlow knows all three.
Insurer
“Our products for this risk”
The corporate lines team: faster triage of broker volume, shorter response times, consistent analysis quality.
Broker
“Where to place this risk”
Products from multiple insurers and a market view: coverage and terms compared, instead of assembling enquiries to several carriers by hand.
MGA / agency
“The carriers we have agreements with”
The same engine on a narrower scope: a subset of carriers, your own catalogue, your own rules.
Numbers from the benchmark and the real corpus
data extraction accuracy (completeness 0.93)
of real-corpus cases processed without a single pipeline error
product field fill-rate on the real corpus (95% for client data)
automated tests guarding every change
Proven on material that breaks tools
Our real corpus is no laboratory: 97% of enquiries are FW: chains, more than half the headers are iso-8859-2, one in three PDFs is a scan without a text layer, the median e-mail is 1.7 MB. Tested on the real market, at a major Polish insurer — F1 of 93–100% on the key fields.
Auditable by design — through AI Watch Tower
LynxFlow’s entire LLM layer runs through AI Watch Tower: a full trace of every call, costs per cost centre, PII masked in logs and a versioned prompt registry. For a supervised insurer, that’s auditability in the sense the AI Act and DORA require.
Explore AI Watch Tower