Lekha.AI — how we rebuilt an enterprise revenue cycle, end to end.

Front-desk staff still key insurance details by hand from scanned cards and paper intake forms, spending 5–10 minutes per patient. Coverage errors only surface weeks later when a claim is rejected, because no one checks that the plan on file is active on the date of service.

Superbill creation is a seven-step manual sequence that takes 25–30 minutes, with paper-based practices re-keying every code, modifier and service level by hand. Handwritten codes have no reliable capture path, so providers code conservatively from memory and leave reimbursement on the table.

Claims go out with no pre-submission validation, so NCCI edits, modifier rules and ICD↔CPT pairing are checked manually or not at all. Payer-specific rules live only in billers’ heads, and every rejected claim triggers a full rework cycle of 25–60+ minutes.

ERA files are downloaded by hand from every clearing house portal, and denial codes are interpreted one claim at a time. With no pattern visibility, the same denial reason recurs unnoticed across payers — and each affected claim consumes 30–60 minutes to manage.

Appeals for complex denials are written from scratch every time, with payer-specific language and evidence requirements that vary widely. Supporting documentation is assembled by hand for each one, so high-value denials are often written off rather than appealed.

Patient statements are generated and chased manually, with no prioritisation — a small balance gets the same effort as a large one. Outreach runs on a single channel with low response rates, and aging balances live in spreadsheets instead of the workflow.

There’s no consolidated AR visibility across practices and clearing houses, so follow-up is driven by whoever remembers rather than what pays back most. Secondary claims are submitted late or missed, and payer denial rates stay invisible until someone builds a report.

MIPS measures are assembled by hand at reporting deadlines, with quality and cost data pulled from several systems manually. Gaps are discovered too late in the performance year to correct, and reporting effort competes directly with day-to-day billing work.

The architecture is locked today, so billing AI cannot be sold independently, and there’s no audited, safe write path into the database. Billers have no single console for exceptions, and the documented schema may not match the live production database.