What the AI does and where it stops
AI billing and coding review adds a reading step between the finished note and the claim. The software reads the documented services alongside the charges your staff propose, then produces suggestions and flags for a person to act on. It does not submit anything, and it does not change a claim.
Gateway helps configure these workflows around the practice’s services, systems and payer requirements. The review process around the AI is described under coding review support.
Three kinds of output
- Candidate codes. CPT and HCPCS candidates drawn from the services the note documents, each tied to the passage it rests on.
- Gaps. Details a code would need and the note does not contain: a site or side, a size or depth, a time, a quantity. Diagnoses that do not relate to the service billed are raised here too.
- Conflicts. Proposed code pairs that meet a National Correct Coding Initiative (NCCI) edit, quantities above a published unit limit, and payer-specific issues the practice has confirmed.
Each item is a question for a reviewer, not an answer.
What the reviewer sees
Every item carries its source and its reason: the sentence in the note that prompted it, the edit or payer rule it was compared against, and the date of service the rule was applied to.
The reviewer accepts it, rejects it or sends a documentation question back to the clinician. The workflow records the choice and who made it.
NCCI edit checks before submission
NCCI procedure-to-procedure edits are pairs of codes that should not normally be reported together for the same patient on the same date. Each pair carries an indicator showing whether a modifier can ever be used to report both. For some pairs it cannot.
The workflow checks proposed pairs against the edits in effect on the date of service and shows the reviewer which kind applies. Units are compared with published limits in the same pass. The edit tables are revised on a regular schedule, so part of Gateway’s work is keeping the checks current.
Modifier review flags
When an evaluation and management service and a procedure share a date, the claim is flagged for review. Modifier 25 requires a significant, separately identifiable evaluation and management service, and the note has to show one. An edit on a claim is never a reason by itself to add a modifier.
The flag places the relevant parts of the note beside the rule. It does not add or remove a modifier. The same applies to modifiers for distinct procedures, sides or sites: the workflow asks the question and your coder answers it.
Where AI gets it wrong
We do not quote accuracy rates. A model can misread a templated note, treat copied-forward text as a service performed today or miss a rule a payer changed last month. The workflow is built with that in mind.
- Nothing reaches a payer without staff review.
- Reviewers’ accept and reject decisions are tallied by flag type.
- A sample of claims that raised no flags is checked by a person.
- Checks that mostly produce noise are changed or switched off.
One visit, start to finish
Take an office visit that includes a procedure. The workflow brings together the visit note, the proposed codes, the payer’s requirements and the authorization status. It flags a possible conflict between two codes, shows the edit and routes the question to the reviewer the practice has named. The reviewer reads the note and decides.
Visits like this are routine in wound care practices, where an evaluation, a procedure and a product billed by unit may share one date.
What stays with the practice
- Code selection and every final coding decision
- Whether a service was medically necessary and adequately documented
- The clinical record, which only the clinician authors or amends
- Approval of each claim, and responsibility for it
Reviewed claims continue through the practice’s medical billing process. Recurring gaps make a good agenda for changes to documentation workflows, and our guide to documentation habits that support clean claims covers the basics.
How an engagement starts
Use the practice support inquiry form or call 314-325-0126. Do not include patient information.
We start with one defined process, such as one clinician’s office visits with a same-day procedure. We confirm which systems hold the note and the charges, what data access is appropriate and who will review. After a few weeks we go through the results with your team and expand only where the flags proved useful.
Common questions
How accurate are the code suggestions?
We do not state an accuracy figure. Results vary with the specialty, the way notes are written and the rules in play, which is why every suggestion goes to a reviewer. Once the workflow is running, the practice sees its own tallies of accepted and rejected suggestions and can judge from those.
How is this different from the claim scrubber we already have?
A scrubber generally checks a finished claim for formatting problems and known edit conflicts. It does not read the note. An AI-assisted review compares the documentation with the proposed codes, so it can raise questions a scrubber cannot, such as a billed service the note does not describe. The two can run together.
Will it push our coding toward higher-paying codes?
It is not designed to. Candidates are tied to what the note documents, and flags run in both directions. A code the documentation does not support is raised as readily as a documented service left off the charge. The reviewer decides in every case.
Will it work with our record and billing systems?
That depends on the systems and the access they allow. Some offer supported interfaces, others only scheduled exports, and the workflow is designed around what is available. We confirm this in the first conversations and tell you plainly if a system limits what can be done.