Compliance

Make sure every call follows the script.

Compliance is the page that tracks whether your team is saying the things they are required to say on every call. Define the phrases your team must say, then see who is on script, who is drifting, and where to coach next. Matching combines keywords and semantic intents, with breakdown per script, per agent, and per call.

Page anatomy

Region	What you see
Filter bar	Date and hour controls, see Overview.
KPI strip	Four summary cards specific to script adherence.
Per-script table	The project’s script library with how each script is performing.
Per-agent table	Average adherence per agent, lowest first.
Calls tabs card	Two-tab card: Needs attention and Top performers.
Edit library dialog	Opens from the per-script card header; admins use it to manage scripts.
Per-call drill-in sheet	Slides in when you click a row in the calls tabs.

KPI strip

Four cards reactive to the active filter window.

KPI	Meaning
Overall pass rate	Share of applicable script checks that passed across all scored calls.
Average adherence	Mean per-call score across applicable scripts.
Calls scored	Count of calls scored in the window.
Lowest pass	The script with the worst pass rate in the active range.

Per-script table

The project’s script library. Each row shows:

Column	Notes
Script	Script name (e.g. Greeting, Agent identity disclosure).
Severity	Low, Medium, or High.
In scope	Number of calls where the script’s trigger applies.
Pass rate	Share of in-scope calls that passed the script.
Avg score	Mean phrase-level score across in-scope calls.

The card header carries an Edit library button that opens the script library dialog.

Per-agent table

Shows average adherence per agent across the active filter window, sorted lowest first so coaching candidates are at the top. Same table mechanics as the per-script table (sortable columns, pagination at 10 rows by default).

The Calls tabs card

A two-tab card titled with Needs attention as the default tab.

Tab	What it shows
Needs attention	Calls where at least one required phrase was missed. Worst score first, up to 50 rows. The subtitle reads Calls flagged for missed required phrases ( total).
Top performers	Calls that passed every required phrase. Best score first, up to 50 rows. The subtitle reads Calls that passed every required phrase ( total).

Both tabs reuse the same table.

Column	Sortable
Date	Yes
Agent	Yes
Duration	Yes
Score	Yes

The pagination defaults to 10 rows per page, with options for 10, 25, 50.

When the tab has no rows, the empty state reads:

Tab	Empty message
Needs attention	All scored calls passed every required phrase.
Top performers	No calls passed every script in this range yet.

The Edit library dialog

Opens from the Edit library button in the per-script card header. Admins can:

Action	Behaviour
Add script ▸ From template	Drops in one of the four starter scripts: Greeting, Agent identity disclosure, Data privacy prompt, Identity verification.
Add script ▸ Blank script	Creates a new empty script you can fill in.
Set Severity	Per script: Low, Medium, High. Drives sort priority and the Lowest pass KPI.
Set Trigger	Per script. Either Every call or Only when intent is in . The intent picker is multi-select and reuses the project’s detected intents.
Edit phrases	Each script has one or more phrases. Each phrase carries a label (EN and AR), keyword chips (EN and AR), an intent multi-select, and a Required to pass toggle.
Save changes	Saves the library to the project. The next score run picks up the new library.
Cancel	Closes the dialog without saving.

Phrase matching rules

Each phrase matches via two independent signals. A phrase must carry at least one signal or the dialog rejects the save.

Signal	How it matches
Keywords (EN)	Word-bounded match: a chip hello matches “hello there” but not “othello”.
Keywords (AR)	Case-insensitive match across the chip list.
Intents	The phrase matches if the call’s detected intent is in the picked list. Reuses the Intent Analysis catalog. No extra cost.

Limits: maximum 50 keywords per phrase, maximum 80 characters per keyword.

Intent matching is great for phrases with many variants. Tie a Greeting script to the greeting intent and it catches every greeting variant the AI already labelled, in either language, with no keyword maintenance.

The per-call drill-in sheet

Click any row in the Calls tabs card to open a side sheet.

Section	What it shows
Session metadata card	Date, agent, duration, sentiment, intents, words, file.
Inline audio player	Built-in browser audio with play, pause, seek, and volume.
Transcription	The full transcript, RTL or LTR per language.
Per-script breakdown	One row per script: Passed / Failed / Not in scope.
Per-phrase rows under each script	Each row shows the phrase label, Matched or Not found in transcript, the matching signal (Keyword EN, Keyword AR, Intent), and a short transcript excerpt around the match when applicable.

CSV export

The page-level Download CSV button emits one row per (call × script × phrase) for the active filter window.

Property	Notes
Hard cap	200,000 rows. Past the cap, you’ll see a message asking you to narrow the date filter.
Encoding	UTF-8 with BOM (Arabic renders correctly in Excel).

Common questions

The starter scripts use English phrases. Do they catch Arabic conversations?

Each starter has both EN and AR keyword chips, plus the greeting / verification etc. intents. They match Arabic out of the box.

Why do some calls show as Not in scope for a script?

Each script has a trigger. If the trigger does not apply to a call (for example, the script’s intent list does not include the call’s detected intent), the call is Not in scope for that script.

A call shows Failed for a phrase even though the agent said the right thing.

Two things to check. First, the call’s intent may not be in the phrase’s intent list. Second, the speech-to-text engine may have transcribed the phrase differently. Open the side sheet — the transcript shows you exactly what was recorded.

Can I make a phrase optional?

Yes. Toggle off Required to pass on the phrase. The phrase still appears in the side sheet’s drill-in, but it does not affect the call’s pass result.

The Edit library dialog will not save.

Most likely a phrase has no keyword and no intent. The dialog requires at least one matching signal per phrase. Add a keyword chip and try again.

What does Severity actually do?

It does not change the matching logic. It drives the sort order of the per-script table and the Lowest pass KPI, so a high-severity script with a 70% pass rate appears above a low-severity script with the same pass rate.

My Calls tabs cap at 50 rows but I have hundreds of failing calls.

The tab caps at 50 to keep the page snappy. Use the Download CSV export to get every (call × script × phrase) row for deeper analysis.

Will the LLM rubric score replace this whole flow?

No. The upcoming LLM rubric is an additional 0 to 100 score with a rationale, surfaced alongside the deterministic phrase scoring. Both signals will live in the same drill-in sheet.

Where to go next

Agent + Team Scorecards

Per-agent metrics across the contact center.

Intent Analysis

Source the intent list used in script triggers.

Alerts

Threshold rules and rolling-baseline anomalies.

Goals

Project-wide KPI targets.