Reading 4.6 million pages a month so that people don't have to.

The problem

The client provides coding, review and revenue cycle services to US healthcare organisations. The work depends on patient records that often run to hundreds of pages per case, held in many different clinical and billing systems. Volume was growing fast, and every case needed a qualified reviewer to find the relevant evidence, reach a coding decision and then enter the result back into the client's own record system.

The financial side had the same shape. The data needed to report on billing and collections sat in dozens of source systems, each with its own interface, and some reachable only through a remote desktop.

What we built

We work as the client's technology partner, alongside its clinical and operational teams. Together we have built two production platforms, and a Chisl team runs and develops both.

Reading the record

The platform reads every document in the record and extracts the information a reviewer needs. Each suggestion links straight to the passage in the source document that supports it, so reviewers can check it without searching the whole record.

The right model for each request

AI requests go through one gateway that can send work to several model providers and to open models we host on dedicated GPUs. Routing rules decide which model handles which task, shift traffic by time of day, and fall back automatically when a provider fails. New models are compared on real cases before they go live, and every request is traced without logging patient content. Running cost is managed as part of the service: routing changes have brought model spend down while volume held steady.

Closing the loop

Helping with the review is only part of the job if staff then have to type the answer into another system. Robotic process automation retrieves records from the client's clinical systems and writes reviewed results back into them, including systems that can only be operated through a remote desktop session.

Bringing the financial data together

The second platform collects billing and collections data from dozens of source systems through whatever route each one offers: APIs, browser and remote desktop automation, or secure file transfer. Workers scale from zero to hundreds as work arrives, and a new client is onboarded by changing configuration rather than writing a new integration. The data lands in a cloud warehouse, is checked by automated tests, and is presented in dashboards where each client sees only its own data.

Built for patient data

Private networking, managed identities, secrets held in a vault, and security scans that block a release when they fail. The platforms run on Azure, and the engineering controls sit alongside the client's own compliance obligations.

Where it got to

Both platforms are in production and used every day by the client's teams. Document volume grew many times over during 2026 on the same platform, while model spend came down. The next step is moving the remaining financial reporting onto the newer data platform, running old and new side by side until the numbers reconcile client by client.