What Conversation Intelligence Tools Capture, and What They Don't

A few weeks ago, we wrote about a pattern that many conversation intelligence and roleplay platforms run into. A rep passes every checkpoint the scoring model tracks and still doesn't close at the rate the dashboard implies. We named the gap. This piece is about what's actually producing it.

What These Platforms Are Built to Read

Most conversation intelligence and roleplay tools work from the same raw material: a transcript and the call's timing metadata. From that, they calculate talk-to-listen ratios, run keyword and objection-handling checklists against the script, and populate forecast fields like deal size, close date, and who else was on the call.

That's a text-and-timing layer, and it's genuinely useful. Knowing a rep let the prospect talk enough, handled the required objections, and surfaced a competitor mention is a real signal for pipeline hygiene and coaching structure. None of that is in question here.

Why a Transcript Can't Carry the Rest

The limitation isn't a bug in any one platform. It's built into what a transcript is. A transcript captures which words were said and in what order. It doesn't capture how they landed.

“I can get you that by Friday” reads identically on a transcript, whether the rep said it with confidence or while backpedaling. Same words, same keyword match, same checkpoint cleared. A scoring model built on transcript and timing data has no way to tell those two calls apart, because the thing that's different between them was never in the data it was built to read.

That's a structural gap, not a tuning problem. Adding more keywords to the checklist doesn't fix it. The information the checklist would need doesn't exist in a transcript to begin with.

What a Sincerity™ Score Adds to the Same Call

Sincerity™ reads the same conversations these platforms already have access to and adds the layer that a transcript can't capture. It works with audio, video, or both, depending on what's available.

From audio alone, it reads four dimensions: prosody (pitch, tempo, rhythm, pause behavior), voice quality (micro-variations in tone and tension), emotional consistency (whether tone holds steady across the call), and a linguistic layer that reads sentiment out of the transcript itself, the one piece these platforms already have. From video, it reads facial expression, eye behavior, head motion, whether a smile reaches the eyes or stays at the mouth, and posture, which is useful when audio quality is poor or unavailable. When both are present, they combine into a single score weighted by each side's confidence in its own read.

Worth saying plainly: none of this checks whether what a rep said is factually true. It describes whether the way they said it lines up with what they appeared to feel when saying it. That's a different question than “did they hit their talking points,” and it's the one a transcript was never built to answer.

Add-on, Not a Replacement

This isn't a pitch to rip out what these platforms already do well. Talk ratios, objection checklists, and forecast fields stay useful regardless of whether delivery gets measured too. Sincerity is built to sit alongside that data as one more field on the same record, not to replace the scoring model a platform has already built around transcript and timing.

If your platform already scores a call on what was said, the open question is whether it's also scoring how it landed. That's the field most of them don't have yet.

We wrote about where this gap first shows up in “Why Pass/Fail Roleplay Scoring Doesn't Predict Who Closes.”