About
A specialist company built around one federal workflow.
American Veterans Care builds authorization intelligence for VA Community Care dental workflows. Not healthcare AI in general, not prior authorization in general: this workflow, read to the line level, with the product designed around what its rules can actually settle and what they never will.
We are early, and this page says exactly how early. What we are not is unfamiliar with the problem.
AVC prepares the decision. Humans make it.
Domain
We read the rules before we designed the product.
VA publishes what it will and will not pay for in dental Community Care, and it is not a short document. Most of it is mechanical (frequency limits, lookback windows, same-day exclusions, per-lifetime caps), and a determined engineer can evaluate it. A smaller part is not mechanical at all, and is written in language no rule engine will ever settle: adequate bone support, willingness to maintain oral hygiene.
That seam is the product. AVC evaluates the part that is decidable, and routes the part that is not to a person, explicitly, with the reason attached, rather than by producing a confident-looking answer to a question that has no mechanical answer. The eight classifications are fixed and the vocabulary is closed for the same reason.
These rules are published in the dental benefits manual attached to a pending, unawarded VA solicitation. They are what VA has published, not confirmed current practice, and not a relationship with VA. We hold them as the best available statement of the rules, which is why the interviews below ask what practices actually experience.
What we found in the published rules
- 206pages of dental benefit rules
- Read in full, not summarized from a secondary source.
- 810per-procedure-code policy blocks
- Each carrying its own limits, windows and exclusions. A large majority are deterministically checkable.
- 7documentation artifact types
- What the requirements actually reduce to once read end to end: narrower than the workflow suggests.
- 19clinical topic areas no rule can settle
- Written in irreducibly judgment-laden language. These are why the product refuses to decide.
Mission
Give the decision-maker a case, not a pile of documents.
A Veteran’s dental authorization moves through several organizations before anyone can act on it. Along the way the facts that matter (what was authorized, what is being requested, what evidence supports it) end up distributed across documents in different formats produced by different systems.
Someone then reassembles that by hand, on every case. The delay that follows is rarely a clinical disagreement. It is more often an administrative one: a missing chart, an inconsistency nobody caught, a scope question that could have been raised two weeks earlier.
AVC exists to remove the reassembly step: to present the authorization question as one structured, source-linked case so the person deciding can spend their attention on the judgment instead of the reconstruction.
Philosophy
Human-centered is a constraint, not a slogan.
There is a version of this product that tries to decide. It would be easier to demo and impossible to defend, because a system that issues determinations in a benefits context has to be right in a way that no current technology can guarantee, and the cost of being wrong lands on a Veteran waiting for care.
So AVC is deliberately bounded. It organizes evidence, applies documented rules, states what is missing, and says when a question needs a person. When it cannot answer, it says so rather than producing something answer-shaped.
The clinician and the reviewer keep the judgment
Clinical judgment belongs to a licensed clinician. Benefits judgment belongs to the authorized decision-maker. AVC’s contribution is upstream of both: making sure that when the judgment happens, the person making it is looking at a complete, traceable, honestly-labelled case.
The security and governance rules that enforce this are written out in detail on the Security & Governance page.
Current stage
We prove the problem before we scale the solution.
The document work told us what the rules say. It cannot tell us what a returned case costs an office on a Tuesday afternoon. So the current phase is field evidence: structured interviews across every role that touches this workflow, run to a fixed script so the answers can be compared rather than collected.
The decision thresholds are fixed before the first interview, never after. That ordering is the whole difference between research and reassurance, and it is the same discipline that governs what this website is allowed to say.
We publish the stage because the buyers we are built for can tell the difference between a company that is early and a company performing traction. One of those is a temporary condition. The other is a permanent tell.
Where we are, precisely
Complete
- Published rules analysis
- 206 pages
- Primary-source verification
- 40 claims
- Engine design & rule policy
- Documented
Not yet
- Customers
- None
- Pilots running
- None
- Production deployment
- None
- Measured outcome data
- None
Research process
Four roles, one script, comparable answers.
This workflow crosses four kinds of organization, and each one sees a different part of it. We interview all four to the same structure, because a problem described identically from four vantage points is a fact, and one described from a single vantage point is an anecdote. We ask about the work as it happens today, never about whether someone likes the idea of the product.
Dental practices and groups
The people preparing VA Community Care treatment packages: what actually takes the time, what comes back, and how often.
Billing and operations leaders
Where the rework sits, what a returned case costs an office, and whether anyone is currently measuring it.
VA Community Care personnel
How dental authorization work reaches them and what makes a submitted package easy or hard to act on.
Administrator-side contacts
Reviewers and utilization-management teams handling authorization volume, and what case readiness means on their side of the file.
We go to the primary source. Every time.
Most of what is written about federal healthcare workflows is a summary of a summary, and the errors compound quietly. We re-derived all 40 of our material claims from the original government records. Thirty held exactly. Five needed qualifying, three were corrected, one was refuted outright, and one could not be verified at all. Those ten are the reason the standard exists.
One of those corrections was a figure we had calculated ourselves from two correctly-quoted VA numbers. Both inputs were sourced. The product of them was not a fact, and multiplying two rounded quantities had manufactured a precision that was never there. We now publish no population figure we computed ourselves.
- Statute and regulation are cited from the CFR or USC directly, not from a secondary summary of them.
- VA policy is taken from official VA, VHA, and federal sources.
- Procurement facts come from official procurement records.
- A citation has to support the exact claim attached to it, not a claim adjacent to it.
- A material claim we cannot verify is either marked as a hypothesis or removed.
Product principles
Eight rules the product is held to.
These are not aspirations written after the fact. They are design constraints: each one rules out a version of the product that would have been easier to build or easier to sell.
- 01
Human authority
AVC prepares decisions. Authorized clinicians and decision-makers make them. The product is built so it cannot quietly take that authority on.
- 02
Evidence before prediction
The first job is to find and organize what the documents actually say. A prediction that is not anchored to evidence is not useful to a reviewer.
- 03
Explainability
Every material flag carries its source, the rule that produced it, the version of that rule, and the reason. A reviewer can interrogate any conclusion.
- 04
Deterministic rules first
If a reliable rule answers the question, we use the rule. A model is not asked to guess at something already determined.
- 05
Existing records remain authoritative
AVC does not alter original clinical documents, VA records, or source authorization records. It reads them and structures around them.
- 06
Workflow compatibility
The product has to fit the way this work is already done. A tool that requires an organization to reorganize itself will not be adopted.
- 07
Data minimization
Collect only what the workflow requires. Every sensitive field has a purpose, a source, an access policy, and a retention rule. AVC does not collect SSN.
- 08
Claims must be earned
Nothing is published as fact until it has been measured or formally achieved. This principle constrains this website more than any other.
What we will not claim
Until it is measured or formally achieved, it does not go on this website.
AVC will not publicly claim performance improvements, time or cost savings, certifications, compliance status, or clinical outcomes that have not been measured or formally achieved. No percentages, no dollar figures, no customer logos, no testimonials, because there is no data and there are no customers.
We also will not imply a relationship with, endorsement by, or approval from the U.S. Department of Veterans Affairs. There is none.
When we do have measurements, they will arrive with the method attached: what was measured, over what population, compared to what baseline. A number without that is a marketing artifact, and this is not an audience that rewards those.
The team
Who is building this.
Founder and team details will be published here. This space stays empty until they can be stated plainly, the same standard applied to everything else on this site.
You know this workflow from the inside. We want that view.
We have read the rules. You run the cases. The conversation worth having is between those two things, including, especially, where your experience contradicts what the manual implies. Twenty-five minutes, structured questions, no pitch deck.