The Strictness Dial: You Decide How Tough Your AI Auditor Is

Today we're handing you a dial for the thing every admin has wanted to tune since day one: how tough your AI auditor is.

Let's be honest for a second. Don't you hate it when the AI is a little too thorough? Your agent greets the customer warmly, covers everything that matters, the customer clearly gets it, and the audit still comes back with a ding because the explanation came out in the agent's own words instead of the exact phrasing on the checklist. Nobody was misled. Nothing was skipped. And yet there it is: another violation, another coaching conversation nobody needed, another agent wondering why the robot grades like it's finals week.

Now flip it around. Some questions deserve a grader with no chill whatsoever. When a CMS disclosure is on the line, "close enough" is not a phrase you ever want in the building.

So which auditor do you want: the tough one or the understanding one? Wouldn't it be nice if you could just choose? Question by question?

Now you can. Say hello to the strictness dial.

Five settings, one slider

Every question in your compliance checklist now has an AI Grading Strictness slider, sitting right next to its violation tier. Slide left for mercy, slide right for menace:

• Most Lenient. If the substance reached the customer in any reasonable form, it counts. Even the customer explaining it back in their own words is evidence.
• Lenient. Benefit of the doubt on paraphrasing, informal wording, and things covered out of order or across the call.
• Balanced. The default, and exactly how Calladhere grades today. Touch nothing and nothing changes.
• Strict. Every substantive element must be explicitly stated. A hint is not a statement.
• Strictest. Explicit, complete, and unambiguous, or it's a no. Zero inference. The grader your compliance officer dreams about.

There's a bulk action too, so you can set a whole section in one click and then fine-tune the exceptions.

Not a vibe, a rubric

Here's the part we're proud of. The slider does not whisper "be 30 percent nicer" to the model and hope for the best. Research on AI graders is blunt about this: models can't calibrate to a number. So each stop on the dial swaps in a written evidence standard that spells out exactly what earns the benefit of the doubt, where the line between a pass and a partial sits, and how much inference is allowed before the answer is no.

That standard governs everything about the question, including any Advanced Prompt criteria you've written. Set a question to Lenient and the AI judges your custom criteria with the same generosity. And because our independent second reviewer receives the exact same standard, your two audit passes never end up arguing about a setting you chose on purpose.

The lines that never move

Before anyone asks: no, the dial cannot talk the AI out of compliance. At every setting, from Most Lenient on down, legally required disclosures, regulatory requirements, and explicit customer consent must actually be spoken on the call. Strictness changes how the grader judges the wording, ordering, and completeness of what your agent did say. It never excuses what was never said.

Every audit also records the strictness that was in force for each question, so when someone asks "why did this pass last week?", you have the answer on file instead of a shrug.

The bottom line

Strictness is now a per-question decision, not a personality trait of your auditor. Keep your CMS disclosures at Strictest, give the greeting question a little grace, and let Balanced carry everything in between. Your checklist finally grades the way you would.

The strictness dial is available now to agency admins in the compliance checklist editor, for both Medicare and ACA organizations. There's nothing to enable and nothing extra to pay: adjust a slider, save, and every new audit grades your way.

Want help deciding where each question should sit? Reach out to your account manager or email us at hello@calladhere.com.