Most companies in this space put an accuracy number on their website, but fewer show you where it actually came from. We wanted to be the kind of company that does both, and that decision shapes a lot more of how we operate than people might expect.

It started with a simple question

When we built Olli, the question that kept coming up internally wasn't "how do we get a good accuracy number." It was "how do we build a number we'd trust if we were the ones buying," and those are genuinely different questions that lead to different decisions.

A good marketing number can come from a handful of clean charts run in a demo, but a number you'd actually trust as a buyer has to come from real, messy, day to day charts, run at the volume an agency actually operates at. So that's where we built our own internal QA process to focus, not on proving the tool works in ideal conditions, but on proving it holds up in the conditions agencies actually live in.

What the internal process actually looks like

Every quarter, our QA team runs a full retrospective audit across our customer base, and it isn't a sample chosen to look good. It's a real cross section of charts already run through production, across coding, OASIS, and Plan of Care. Our most recent quarter covered nearly 1,000 chart audits across more than 100 customers, and we publish the full breakdown in our Q2 Analysis by service line rather than leaning on one blended figure.

That process exists because a single number hides too much. Knowing our overall accuracy score is 97.5 tells you almost nothing about where the remaining gap actually sits, but knowing the breakdown by service line, and by the type of error, tells you something you can actually act on, whether you're us trying to improve the product or an agency trying to decide whether to trust it.

Why we went through CHAP Verification on top of that

Publishing our own numbers is one layer of transparency, but having an outside party check our process is another, and we think it matters more. CHAP is the accreditation body agencies in this space already trust, and going through CHAP Verification for our OASIS and Plan of Care Quality Review service meant someone outside our own walls looked at how we actually do the work, not just the number we report at the end of it.

That distinction matters because self reported numbers, even honest ones, are still self reported. A verified process is a different kind of claim, since it says the methodology behind the number has been checked, not just the headline figure itself.

What this actually means for agencies evaluating us

We aren't publishing this expecting every number to look perfect forever. Some quarters will show more room to improve in one service line than another, and when that happens, we'll say so, because an agency evaluating any AI coding or QA vendor should be able to see the real number, see how it was measured, and see whether anyone outside the company checked the work.

We'd rather set that expectation ourselves than wait for someone to ask.

See it for yourself

Numbers on a page are still just numbers on a page, and the real test is what happens on your own charts. We offer a 30-day, low-risk pilot, opt out any time, so you can see the speed and accuracy of our work and the quality of our support firsthand, before committing to anything long term.