Renewals Before They Become Emergencies

The Company

Nordvik Insurance Brokers works out of Oslo and looks after commercial clients in marine and logistics — shipping companies, freight forwarders, port operators, that kind of thing. About 55 people. Their clients tend to carry complex coverage: hull, P&I, cargo, liability, often with co-insurers across different markets, some documentation in English, some in Norwegian, some in whatever language the overseas insurer prefers. The policies live across three separate legacy portals, none of which talk to each other.

The Pain

Ingrid handles roughly 40 commercial accounts. When a renewal comes around, her first job is just finding everything. She logs into one portal for the current schedule, a second for claims, and sometimes a third for older endorsements. Then she's cross-referencing a spreadsheet she built herself, checking whether anything's lapsed or changed, trying to remember what the client mentioned on the phone six months ago. By the time she has a clear picture of where the client actually stands, she's already behind. The renewal meeting is in two weeks and she's still in the documents. She doesn't have time to think about whether the client is underinsured or whether there's a better structure available. She goes in with what she has.

The Move

Sixty days before each policy expiry, an automated pipeline pulls together everything Nordvik already holds on that client: the current coverage terms, recent claims, any notes in the CRM. It also pulls in current market rate benchmarks for that coverage type. An LLM works through the comparison and produces a structured brief: what the client has, where there are gaps against typical coverage for their exposure profile, any upsell opportunities worth raising, and a suggested opening position for the renewal conversation. That brief lands in Ingrid's CRM record before she's made the first call.

The tools to do this exist. The pipeline can be built in Make or n8n. The LLM layer is GPT-4 via the API. The output drops straight into HubSpot or whatever CRM they're running. None of this requires a developer on staff — it requires someone to build it once.

Ingrid still does the client relationship. She still decides what to recommend. She just starts from something useful instead of starting from scratch.

The blind spot

The portals. Everyone assumes that because the data lives in three separate places, it can't really be automated. So nobody tries. But you don't need a perfect integration — you need the key fields, and those can be pulled or even pasted into a structured template to start with. The messy-data problem is real, but it's smaller than it looks once you actually sit down with it.

The pattern

The same setup fits a lot of places where someone has to read and compare documents before a client conversation:

• A recruitment firm preparing for a retained search briefing, pulling together role history, previous candidate feedback, and current salary benchmarks
• A commercial property manager reviewing lease renewals, with rent reviews, maintenance records, and market comparables scattered across different systems
• An accountancy practice preparing for an annual client review, combining management accounts, prior-year notes, and any open queries before the partner gets on the call