Questions

The things people ask on the first call, answered in the open. If yours isn’t here, ask — we would rather answer the hard one now than have it come up after a contract.

How long does it take to start?

The audit needs one thing from you: a folder of remittance files. No install, no integration, no IT ticket. Most customers export 90 days from their clearinghouse in an afternoon, and we return the arithmetic within a week. Working claims takes longer to set up than reading them — that step needs your contract terms verified and your filing credentials, which is usually a week or two of part-time work on your side.

Do you connect to our EHR or billing system?

Not at first, and that is deliberate. Every integration is a security review, a project plan, and a reason for the whole thing to stall before anyone has seen a dollar. Files first: the 835s and 837s your system already produces are enough to find the money and draft the appeals.

Once you want filing automated rather than assisted, we integrate with the clearinghouse you already send through — not the EHR. That is a narrower surface, and it is where claims actually leave your building. Anyone promising to plug into “all major EHRs” on day one is describing a roadmap.

Does the AI write the appeal letters?

It writes the draft. Then a second pass reads that draft against the evidence and checks every factual sentence — the policy quote, the contract rate, the date of service, the dollar figures — back to a source. A draft with a claim nothing supports cannot be approved, and that is enforced in the database rather than by anyone remembering.

The arithmetic is never written by a model. Money, contract rates, and filing deadlines are computed in ordinary code, because that is the part that has to be exactly right every time and it is also the part a language model is worst at.

Who decides what gets filed?

Your staff. Every draft waits for a human to approve it, and the queue is built so that approval takes about two minutes: the finding, the arithmetic, the evidence, and the deadline on one screen. Nothing is sent autonomously, and you can switch any part of it off.

Can we see what it is doing?

Every finding carries its trace — which remittance line it came from, which contract rate it was compared against, which policy version was in force, and the rule that fired. You can hand-check any row against your own files, and during the audit we expect you to.

Filing confirmations are archived with the dispute, because “we never received it” is a real payer response and the receipt is what settles it.

What happens when it gets something wrong?

It will. The interesting question is what the error costs, and the system is arranged so the answer is “a reviewer’s minute”. A wrong finding is visible before anything is filed, because a human approves every send and the trace is on the screen next to the claim.

Where the model is uncertain it says so rather than guessing: a dispute it cannot support with evidence does not become a confident draft, it becomes a dispute nobody has to look at. When a plan type is ambiguous we take the shorter filing window, which is the conservative error rather than the expensive one.

Are we paying for claims we would have worked anyway?

This is the fairest question anyone asks, and the answer is a working rule you write down before the pilot starts — the threshold under which your team was already writing claims off. We work everything and bill below the line. Above it, you keep what you were always going to collect.

What does it cost?

A share of what actually posts, and nothing before that. We read the same remittance feed you do, so a recovery cannot be invisible to either of us. The rate depends on your volume and mix, which is what the audit establishes — quoting a number before seeing your files would be a guess dressed up as a price.

Do you do underpayments, or only denials?

Both, and the underpayments are the part most tools never look at. When a payer takes a bigger contractual adjustment than your contract allows, the claim settles at zero balance with no denial code and no work item. There is nothing in the denial queue to find, which is exactly why it goes unworked for years. See underpayments.

We already have a denials vendor. Does this replace them?

Usually not. Most vendors and most in-house teams work the large denials, because those are the ones that repay the labour. The overlap is smaller than people expect — what we take on is the volume beneath their threshold and the zero-balance underpayments nobody is looking at. The audit tells you the size of that gap before you decide anything.

Is our patient data safe?

Short version: we receive remittance and claim files, not charts; each customer’s data is isolated and encrypted with its own key; every access is logged; no customer data trains any model; and AI processing runs under a signed BAA. The long version, written for whoever has to sign off on us, is on the security page.