Before you trust AI with a veteran's file: seven questions for any vendor
AI is arriving in veterans' claims work whether anyone invited it or not. Some of it will genuinely help the people doing this job. Some of it will put veterans at risk while charging them for the privilege. The difference is not in the marketing — it's in the answers to seven plain questions. Print this. Ask it of every vendor who comes through your door, including us.
Free, accredited help is the veteran's legal right. Any tool that charges veterans — directly, by subscription, or as a cut of backpay — is competing with the free system you provide, not strengthening it. Red flag: consumer pricing aimed at veterans. The honest answer is that the tool is sold to offices and organizations, and the veteran is never billed for anything.
An accredited representative's signature is not a formality — it is the law and the veteran's protection. Ask whether the tool can submit anything on its own. Red flag: any "auto-file" or "one-click submit" feature. The right answer: every output is a draft for your review and signature, and no submission pathway exists inside the product at all.
"AI-powered" is not a number. "80% median rating" is a marketing line, not a measurement. Ask how the tool is tested, against what, and to see the results — especially the ones it got wrong. Red flag: a vendor who can't show you a benchmark, or who only shows wins. A tool tested honestly against published Board of Veterans' Appeals decisions, with its misses published and explained, is a tool you can actually trust — because you can check it.
The dangerous failure isn't being wrong. It's being confidently wrong — a clean, plausible answer with nothing behind it. Ask what the tool does when the record doesn't support a conclusion. Red flag: a system that always gives a confident answer. The right behavior: it tells you it isn't sure, names the specific fact that would decide the question, and leaves the call to you.
You are about to hand a vendor the most sensitive records a person owns. Ask directly: encryption at rest and in transit? A signed Business Associate Agreement? An audit log of who touched which file? Red flag: vague "enterprise-grade security" with no HIPAA commitment and no BAA. If a vendor won't put veteran-data protection in a contract, the protection isn't real.
Many AI products improve themselves by learning from what users feed them. With veterans' medical and service records, that should be a hard no. Red flag: a privacy policy that reserves the right to use your data for "product improvement." The right answer is a contractual guarantee that veteran data is never sold, shared, or used to train models — full stop.
You already have a system of record. A tool that demands you abandon it, retrain your staff, and migrate your data is buying itself a failed rollout. Red flag: "rip and replace." The better answer: it does the analysis your current software was never built to do, and its output drops into the workflow you already run.
The standard, not the sales pitch
We wrote these questions to be used on us. We built Guidon to answer every one of them the right way, and we published our answers as a five-point pledge and an open accuracy page precisely so you wouldn't have to take our word for it. But the questions matter more than our answers. If they help you evaluate every vendor in this space — and hold all of us to a higher bar — then they've done their job, whichever tool you choose.
The veterans you serve deserve nothing less than the most demanding buyer in the room. Be that buyer.
Questions or a vendor you'd add to this list? chancey@guidonhq.com.