AI product studio

We build AI products. Then we build yours.

We built ClaimVision, AI claims intelligence for US property insurers, and Showroom IQ, season intelligence for fashion wholesale, ready for beta. Now we build products like them for other businesses: the workflow your team does by hand, turned into software it uses every day.

  • ClaimVision is generally available.
  • Every ClaimVision finding is cited to its clause and page.
  • Showroom IQ is ready for beta.

Product 01 · ClaimVision · Generally available

Every page read before anyone opens the file.

ClaimVision is an AI claims intelligence platform for US property carriers. It turns every page of a claim into structured data before your team opens the file, and lets them question that data with the evidence attached. It sits on top of the core claims system a carrier already runs, so nothing is migrated.

  1. ClaimVision's photos view on a fictitious example claim: eight storm-damage photos, each tagged with what it shows.

    01 / 03

    Photos and pages, read together.

    The adjuster report, the estimate, the policy and every photo are read as one file. Each photo is tagged with what it shows, so nobody scrolls through a hundred images to find the roof.

  2. ClaimVision's assistant answering a coverage question on a fictitious example claim, marking what needs review and asking for the policy page it doesn't have.

    02 / 03

    Ask a question, get the page.

    Adjusters ask in plain English. The answer cites its sources, and when a document it needs isn't in the file, it says so and asks for it instead of guessing.

  3. ClaimVision's claims overview as a map, with weather radar, hurricane tracks and flood zones over open claims.

    03 / 03

    See across every open claim.

    The same reading powers the portfolio view: open claims on a map, with weather radar, hurricane tracks and flood zones alongside.

Example claim. Fictitious demo data; names are invented. Screens from ClaimVision.

The reading goes to the machine. The judgement stays with your people.

What we can show today

A full day's reconciliation work on a real large-loss claim, done exactly, in under a minute.

On a real large-loss claim, ClaimVision did a full day's reconciliation work, exactly, in under a minute. Every finding is cited to its clause and page, and carriers can test it first on closed files they have already decided. ClaimVision runs on Microsoft Azure. It is built to SOC 2 principles and is working towards SOC 2 Type 2.

Illustrative path from a claim file to a one-page brief A large-loss claim file, a stack of documents, is read into cited findings. Each finding is tagged with a clause and page. Those findings become a one-page brief. A full day of reconciliation work, under a minute. Illustrative, with no real claim data. Large-loss claim file A stack of documents Cited findings Finding Clause and page Finding Clause and page Finding Clause and page One-page brief The same findings, with their sources attached A full day of reconciliation work Under a minute
Illustrative. No real claim data.

Product 02 · Showroom IQ · Ready for beta

Sell the season, not the spreadsheet.

Fashion brands sell wholesale through digital showrooms like JOOR, but the order data lands in exports that someone downloads, merges and reconciles before anyone can say how the season is going. Showroom IQ is our own invention, created during an engagement with a renowned B2B fashion business. It is built and ready for beta, and it connects to JOOR API data.

Sync.
Orders arrive over the JOOR API as the season happens. Nobody downloads an export.
Reconcile.
The ERP stays the golden source. Every figure traces back to where it came from.
Decide.
Forecast rounds lock with an audit trail, and client-ready reports come from the same data.

See Showroom IQ →

The season, read: an illustrative diagram of Showroom IQ Orders from JOOR, actuals from the ERP and targets by region are read together. One line in each source links to a season brief with four findings: orders synced, matched to the ERP, forecast round locked, account report drafted. Illustrative, with no real data. JOOR ORDERS ERP ACTUALS TARGETS BY REGION Europe Americas Asia Pacific Middle East Season brief Illustrative · invented data Orders syncedJOOR API · no export Matched to the ERPevery figure traceable Forecast round lockedaudit trail kept Account report draftedfrom the same data
Illustrative diagram of Showroom IQ. Invented layout, no real data.

For other businesses

We build products like these for other businesses.

ClaimVision and Showroom IQ started the same way: one team, one workflow that ate the week, and documents or exports nobody had time to read properly. If your business has work like that, it may be a product.

It's a fit when…

  • Skilled people spend their day reading documents and re-typing what they find.
  • The numbers live in exports and spreadsheets that someone reconciles by hand.
  • The decision needs judgement, but the reading and reconciling don't.

How we build

  1. 01

    Discover.

    We start with the work itself, not a strategy deck. We sit with the people doing it and map what actually happens: the documents, the exports, the hand-offs, the spreadsheet nobody trusts.

  2. 02

    Prototype.

    Real interfaces on real documents, reviewed with the people who will use them, before a line of production code. If the answer is "don't build this yet", you hear it here.

  3. 03

    Build.

    Narrow and deep: one workflow, fully replaced, on top of the systems you already run, with an audit trail the manual process never had.

  4. 04

    Run.

    We stay with what we build. Products are monitored, supported and improved with the people who use them every day.

Start a conversation

A thin layer of glowing glass, traced with a fine network of light, resting just above a solid slab of stone.
A thin layer on the systems you already run, not a replacement for them.

Why we build

AI won't make us less human.

Used well, it can make us the most human we've ever been. That line started with ClaimVision, and it shapes everything we build.

In most businesses, skilled people spend most of their hours on documents and spreadsheets, and whatever is left on the people they serve. Not because they want to. The reading was the job because nothing else could do it.

Now something else can. Not decide. Read: all of it, carefully, holding the numbers steady, showing its reasoning, and handing the judgement back to a person with everything in front of them.

That's why our products show their work instead of acting as oracles. Tools that replace judgement make people smaller. Tools that inform it make them better at the parts only they can do.

Our sister company, Decision Crew, is an AI consultancy. Same founder, separate firms.

Tom Welgemoed, founder of Decision Agency.
Tom Welgemoed · Founder, Decision Agency

Bring us the workflow. We'll tell you if it's a product.

Thirty minutes. Bring a real example: a file, an export, the spreadsheet everyone argues about. We'll tell you honestly whether there's a product worth building.