After the demo

An AI document layer on loan writing.

In the session we showed Flint what is possible when documents are read by AI the moment they land, not keyed by hand days later. This is a working demonstration of the idea, not software built for Flint yet. It is here so Ben and Redom can see the shape of it again, and decide whether it is worth building for real.

Borrower document portal Live AI extraction Broker view in HubSpot

What you saw

Two moments, one pipeline.

The demo walked the path a real loan takes, from the borrower uploading a file to the broker seeing what it means. Two parts did the work.

A Flint-branded borrower portal that reads documents as they land.

The borrower drops a file in. The portal recognises what the document is on upload, then reads it with AI in seconds. The live moment in the room: it read a company financial statement and pulled the numbers a credit analyst actually needs, on screen, with a confidence score against each one.

Read from the statement · on screen, in seconds

Net profit$486,200Confidence 98%
Depreciation$72,400Confidence 96%
Interest$58,900Confidence 97%
Servicing add-back$617,500Confidence 95%

Figures shown are from the demo document, not a Flint client. The point is the behaviour: the add-back is assembled from the read figures, not typed by a person.

A broker view inside HubSpot that already knows what is outstanding.

On the broker side, the same upload surfaced inside HubSpot. It tracked what is still missing against the file, and showed the servicing numbers without anyone re-keying them. The broker reads the position instead of building it.

The Phase 1 outcome

Take the admin layer off loan writing.

Phase 1 lifts Loan Processor and Data Entry out of the loan-writing process, so analysts spend their time on credit, not on keying.

Faster Faster for credit analysts. The figures arrive read and structured, so the analyst starts from a position instead of a pile of PDFs.
Fewer Fewer manual touches. Each hand-off where someone re-types a number is a place an error or a delay can enter. The demo removes them.
Tighter Tighter file control. Outstanding items are tracked against the file from the first upload, not chased at the end.

The seam we target first

From Documents received to Application approved.

Where this goes

An agent working the systems Flint already uses.

The demo is the first slice. From here it extends the existing Middle to Broker Engine data pipe, and grows into an agent that works across Flint's stack rather than sitting beside it.

Extend

The Middle to Broker Engine pipe

Build on the data path that already exists between Middle and Broker Engine, rather than introducing a new one.

Operate

An agent on your own systems

Not a separate tool to log into. An agent working inside the platforms your brokers already use every day.

Control

Human kill-switch, onshore data

A person stays in the loop with the ability to stop it, and the data stays onshore. Built for a regulated process.

Works across HubSpot Middle AFG / Broker Engine

To build it for real

Two things from Flint, and we can scope it.

The demo proves the behaviour. To turn it into something built for Flint's actual process, we need a clear picture of how loans move today and access to the two systems they move through.

A short Loom of the current process

One walkthrough of how a loan moves from documents received to approval today, in your own words. It tells us exactly which touches to remove.

Intros to AFG and Middle

A warm introduction to the right people at AFG and Middle, so we can confirm the integration points and build against the real pipe, not a guess.

If the demo felt close to how you already work, that was the point.

Everything you saw is a demonstration of what is possible, not delivered software. With the Loom and the two intros, we can scope Phase 1 properly and show you exactly what it takes to run it on real Flint files. Whenever you are ready, we are.