Eval directory

Evals for AKASA

Eval coverage for AKASA, mapped from its public product surface.

About AKASA

AKASA provides a generative AI platform for the healthcare revenue cycle, trained on clinical and financial data. Its solutions include a Prebill Optimization Suite, Coding Optimizer, CDI Optimizer, and an AI Advisor research assistant for revenue cycle teams. It targets health systems seeking to reduce denials, close documentation gaps, improve coding accuracy, and improve margins.

Industry

generative AI for healthcare revenue cycle management (RCM)

Website

akasa.com

Use the eval library for AKASA

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for AKASA?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

Coding Optimizer

GenAI review of coded encounters to surface missed coding opportunities, flag quality and compliance risk, and support revenue integrity before the claim goes out.

“Unify coding and CDI to close documentation gaps, ensure accuracy, and improve quality with GenAI-powered optimization.” akasa.com

Mapped capabilities

4 capabilities

Missed code opportunity surfacing
Identifies codes supported by the chart but absent from the coded account, with the documentation passage that supports each suggestion.
Quality and compliance risk flags
Flags coding that is unsupported, over-specified, or otherwise a compliance exposure rather than only upside opportunities.
Present-on-admission and severity capture
Handles POA indicators and severity-affecting diagnoses, the capture area Cleveland Clinic reported improving.
Coder-facing rationale
Explains why a suggestion was made in terms a working coder can accept, reject, or escalate.

Illustrative example

Input: Coded inpatient account. Admission H&P documents a stage 2 sacral pressure ulcer present at arrival; the wound care note repeats it. No pressure ulcer diagnosis appears on the coded account.
Expected behavior: Surfaces the omitted pressure ulcer diagnosis as a missed coding opportunity, marks it present on admission, and cites the admission H&P as the supporting documentation. It presents this for coder review rather than applying the code itself.

CDI Optimizer

GenAI assistant that uncovers documentation gaps in the clinical record and helps CDI teams drive toward accurate, complete patient documentation.

“Uncover documentation gaps and enable accurate patient documentation with a GenAI assistant” akasa.com

Mapped capabilities

4 capabilities

Documentation gap detection
Detects clinically implied conditions that lack the specificity or linkage documentation requires.
Query-worthiness judgment
Distinguishes gaps that warrant a physician query from those already adequately documented.
Non-leading query support
Drafts or supports queries that present clinical evidence without steering the provider to a particular diagnosis.
Complete clinical story assembly
Reconciles evidence across notes, labs, and orders into a coherent account of the encounter.

Prebill Optimization Suite

The unified coding-and-CDI layer that closes documentation gaps and checks accuracy before bill drop, positioned as one workflow rather than two disconnected reviews.

Mapped capabilities

4 capabilities

Unified coding + CDI reconciliation
Resolves conflicts when the coding view and the CDI view of the same encounter disagree.
Prebill timing and hold decisions
Determines whether an account should be held for review or released, given the bill-drop deadline.
Encounter prioritization
Ranks the worklist so limited staff time lands on the accounts with the most impact or risk.
Handoff between coder and CDI
Preserves context, prior decisions, and open questions as an account moves between roles.

AI Advisor

Research assistant for revenue cycle teams that finds answers, source documents, and surrounding context faster than manual lookup.

“AKASA is the leading provider of generative AI solutions for the healthcare revenue cycle” akasa.com

Mapped capabilities

4 capabilities

Grounded answers with citations
Returns answers tied to the specific guideline, policy, or payer document they came from.
Document and passage retrieval
Locates the relevant section of long reference material rather than the document as a whole.
Abstention on unsupported questions
Declines or qualifies when the corpus does not contain an answer instead of composing a plausible one.
Revenue cycle domain vocabulary
Interprets RCM shorthand — DNFB, DRG, MCC, denial category — without requiring the user to expand it.

Illustrative example

Input: Revenue cycle analyst asks AI Advisor: "What is this payer's timely filing window for corrected claims?" for a payer whose contract documents are not in the connected corpus.
Expected behavior: States that the connected documents do not contain a timely filing window for that payer, and points to what is available or where to look next. It does not supply a specific number of days as if sourced.

Clinical and Financial Grounding

Whether platform output stays anchored to the clinical and financial data it claims to be trained on, and behaves defensibly under audit.

“It can unlock clinical data (like chart records) that were previously opaque to computers.” akasa.com

Mapped capabilities

4 capabilities

Evidence traceability
Every suggestion points back to a locatable passage or data field in the source record.
Fabrication resistance
Does not assert diagnoses, dates, or documentation that the record does not contain.
Ambiguous or conflicting record handling
Surfaces the conflict when notes disagree rather than silently choosing one reading.
PHI handling in output
Keeps patient identifiers out of surfaces where they are not needed.

Mid-Cycle Team Workflow

How the platform fits the staffing and accountability reality of mid-cycle teams — human review, override, and the reporting leaders use to judge impact.

“Find answers, documents, and context faster with a research assistant built for revenue cycle teams” akasa.com

Mapped capabilities

4 capabilities

Human accept, reject, and override
Staff decisions are recorded and the system respects a rejection rather than resurfacing it unchanged.
Reviewer-facing audit trail
A compliance reviewer can reconstruct what was suggested, by what evidence, and who acted on it.
Impact reporting for leaders
Reports capture volume reviewed, accepted suggestions, and effect on coding quality measures.
Onboarding for a shifting workforce
Usable by newer coders as experienced staff retire, the workforce pressure AKASA cites.

Coverage is mapped from AKASA's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for AKASA test?+

The coverage map is generated from AKASA's own public product surface (generative AI for healthcare revenue cycle management (RCM)): 6 scoring areas — Coding Optimizer, CDI Optimizer, and Prebill Optimization Suite, and more — spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the AKASA evals scored?+

Every case generated for AKASA — across Coding Optimizer and CDI Optimizer and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the AKASA library include?+

The full AKASA library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Missed code opportunity surfacing and Quality and compliance risk flags under Coding Optimizer); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against AKASA or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped AKASA areas and set them up in a Corsac workspace, where you can run every test case against AKASA or your own agent with your own data.