How it works

Four layers. Only two of them are ours to charge for.

Densery sits on top of your existing core, policy admin or claims system rather than replacing it. Nothing is ripped out, nothing is migrated, and the first document type goes live in under three weeks.

LAYER 01 · COMMODITY, PASSED THROUGH

Perception — reading the document as it actually arrives

Mixed layouts, multiple scripts, handwriting, photographs of paper, forty-page PDFs and engineering drawings. We buy best-in-class extraction, including open-weight models that run inside your walls, and we route between them. This layer is close to free across the industry and we do not pretend otherwise.

LAYER 02 · OURS

Ontology — your institution's concepts, typed

Counterparty, beneficial owner, exposure, claim, policy, lien, obligation, jurisdiction — defined once, with the relationships and constraints between them. This is where a specialist's judgement gets codified: which combination of fields makes a file complete, which mismatch is material, which exception a regulator will ask about. It is the reason two institutions with identical documents get different answers.

LAYER 03 · OURS

Completion — the work finished into the system of record

The agent performs the task rather than proposing it: the file is assembled, the check is made, the decision is recorded, the record is written back through your API. Approval gates sit wherever your policy requires a human signature, and the agent stops there and waits. Confidence below threshold routes to a named person with the source page attached.

LAYER 04 · OURS, AND THE POINT

Evidence and correction

Every action is logged with provenance — which document, which page, which model, which version, which rule, which human. Every correction a reviewer makes is captured as training signal for your ontology. The system gets measurably better on your documents, and the record of how it got there is exportable.

Deployment

Three ways to run it. One of them is why people call us.

ON-PREMISE

Your hardware, your building

Including a GPU appliance we supply and install. No customer data crosses your network boundary at any point. This is the configuration for institutions under a legal residency obligation, and it is the deployment our banking customer runs.

PRIVATE CLOUD

Your tenant, your keys

Deployed inside your own VPC with customer-managed encryption keys, for institutions whose policy permits cloud but not shared infrastructure.

HOSTED

Managed by us

The fastest path to a first production process. Appropriate where there is no residency constraint — though if that is your situation, be honest with yourself about whether you need us at all.

On-premise capability is no longer scarce. Open-weight document models run on a single consumer GPU. What is scarce is on-premise combined with a governance stack — and it only matters where the constraint is legal rather than a preference.

Integration

It writes back, or it is not finished.

A system that produces a spreadsheet of extracted values has not completed the work; it has moved it. Densery writes into the system that holds the truth — core banking, loan origination, policy administration, claims, case management, document management — through your existing APIs, under your existing service accounts and permissions.

Where a legacy system has no API, we operate it the way your staff do, under a named service identity with every action logged. That is less elegant and it works.

Model routing is ours to manage, not yours to worry about: routine steps go to inexpensive models, hard reasoning to frontier ones, and anything under a residency constraint to a model running inside your walls. You are never locked to one lab's pricing or roadmap.

WHAT WE DELIBERATELY DO NOT DO
  • Rebuild OCRExtraction is commoditised. Building our own would be a worse product at a higher price.
  • Require a migrationYour core system stays exactly where it is.
  • Train on your data by defaultNothing is used to improve anything outside your instance without a written data-rights agreement and a commercial term attached to it.
  • Sell you a platform to build onIf the deliverable is a capability your team has to finish, we have not sold you anything.

The next step

Ninety minutes, your documents, three numbers.

A scoping session is not a demo. Bring twenty real files, redacted if you need to. We take three numbers off you — annual volume, fully loaded cost per file today, and what happens when the output is wrong — and hand back a one-page value case in your own KPIs.

If the arithmetic says we are not a fit, we will tell you in the room rather than six weeks later.

QUALIFY YOURSELF OUT

We are a fit if all three are true

  • More than 250,000 pages a year, or 25,000 claims or files
  • A legal obligation — regulator, board risk committee or parent-company policy — to keep the data in-house
  • An AI or agent pilot that did not reach production

If your data can go anywhere and your documents are already clean and digital, you do not need us. Use a hyperscaler document API and spend the money on something harder.