Release Notes.

Written and illustrated reports, one-time or recurring

A report is a written, illustrated document generated by the AI from your data: it includes a summary, structured sections, key figures, tables, concrete examples, and recommendations, along with a description of the methodology used (time period, filters, scope, sampling). It can be created in four steps from the Reports page, or by entering a single sentence into Copilot from a dashboard.

• A one-time response to a question that comes up in the moment, or a recurring question—daily, weekly, or monthly—that arises during your ritual.
• Every run is recorded on the Reports page, and the document is shared via a link.
• A report requested from Copilot uses the scope displayed on the dashboard.

The Copilot opens from every page

The Copilot is present on nearly every page of the platform, and it knows where it is. From a dashboard, it reads the displayed widgets and the active filters; from an evaluation, it works on the open conversation; from a grid, it helps define the criteria. On pages that do not display any figures, it remains available but never suggests a figure.

Pre-written answers, ready to be displayed

Every chart and every quote appears at the point in the text where it is discussed.

• Customer quotes are taken directly from the original conversation, are dated, and are clickable.
• While it's working, Copilot explains what it's doing, step by step. Once the answer is written, those steps collapse into a single expandable line.
• When Copilot writes a report, a button appears below its answer to open it, and it remains there when you reopen the conversation.
• The Copilot identifies and names filters on a section score or an additional score, and names the score behind an indicator when you ask what a widget measures.

Report a problem or suggest an improvement via chat

Something isn't working—a button that doesn't respond, a page that won't load, an export that never comes through—just let Copilot know in the chat. It asks all the necessary questions at once, shows you a summary, and forwards the issue to our teams, automatically including the page, the active filters, and the technical context. Feature requests follow the same process. You'll receive a confirmation in your notifications.

A chart summarizing the entire analysis

A chart combines the different fields of a custom analysis—one bar per field—so you can see at a glance which ones stand out and compare them before drilling down into the values of each.

• On a time axis, one line per field shows how often each field is observed over the period.
• Clicking a bar displays the field's details, followed by its values and their subcategories, along with a breadcrumb trail to navigate back up.
• A field added to the analysis appears in existing charts.

Charts That Are Easier to Read

The dashboard charts have been redesigned to provide information at a glance.

• The total appears above the stacked bars, and the value of each slice is shown in the pie chart legend.
• Labels remain horizontal, a widget title can span two lines, and the plot fills the entire height of its card.
• Bars can be arranged horizontally.

Free-text answers and customer feedback are added to the filters

• A "filled in" or "empty" filter applies to an open-ended question and can be pinned just like the others. The verbatim count can be narrowed down to a single specific question.
• A speaker selector—whether it's a customer, agent, or bot—lets you keep only the speech you care about in the "Segments" tab, as shown in the charts that count speech per message.

Good to know: The verbatim count now ignores empty answers, so the numbers on existing dashboards may decrease.

Versions and a draft for analyses and grids

Changes to a custom analysis or an evaluation grid remain in a private draft until you publish them, and each publication has its own version.

• An evaluation remains linked to the version it was assigned to.
• A selector lets you view a previous version and restore it to a new draft.
• The name and scope are set separately, without creating a version.
• When Copilot generates criteria, it also adds them to the draft: only you can publish. It allows you to edit a grid that has already been used.

Test and analyze a few conversations before publishing

You pin a small sample of existing conversations to an analysis, run the version you're writing on them, and compare the results with those produced by the live version. The test runs on the draft, publishes nothing, and leaves the production results unchanged. A conversation is added to the tests from its details panel.

Pause an analysis without losing its results

An analysis can be paused: it stops applying to new conversations, and everything it has generated remains visible. It resumes with a single click, with no need to recompute anything. The analyses you’ve already run cover everyday events, a campaign, an incident, or a spike in activity—without needing a separate analysis for each one: pausing is intended for temporary analyses.

Grades remain in the right hands

The evaluation visibility setting applies to guests. An evaluated person always sees their own evaluations, including the AI explanation and the discussion. Where the setting requires it, only the scores are hidden, and they are hidden everywhere: in the details, in the PDF export, and in Copilot.

Import conversations in bulk

An import screen processes an entire batch of conversations—up to fifty items at a time.

• Audio recordings, or written transcripts of conversations in the form of a spreadsheet or JSON file for chat and email, that can be read immediately without having to wait for a transcription.
• An entire folder can be dropped all at once.
• The agent, the date, and the attributes are set for the entire batch, row by row, or from a mapping file whose template you can download.
• Before the import, every issue is reported along with the row it appears in: duplicate, unsupported format, future date, unknown agent.

Every chart speaks for itself

Some charts are inherently complex, and others have filters applied at design time that not every user is necessarily aware of. Here are two ways to understand exactly what a widget displays:

• Ask the Copilot ("ask a question about this widget"): it explains what is counted, how, and over what scope.
• Click the widget's icon, which shows the filters that have been applied.

The source of any figure can be verified with a single click, at any time.

Conversation segments become available

The new Segments tab displays the conversation segments that actually match the search term or the applied filters. Go from the total count to the verbatim content in a single step, without leaving the dashboard.

• Each segment displays its source, date, and agent.
• Search terms are highlighted.
• The original conversation opens with a single click.

Building your views gets faster

Fewer repetitive clicks, and everyone picks up where they left off in their exploration.

• Duplicate a widget with a single click to create your views.
• Let the platform name the widgets based on their configuration.
• Pin your favorite filters: they'll be right at your fingertips from one session to the next, personalized for each user.

Alerts That Hit the Mark

Alerts and widgets now span multiple dimensions of the same analysis—such as a topic and a sentiment. Setting up a notification for when "delivery" appears with a negative sentiment takes just a few clicks: you’ll be alerted to the specific combination that matters to you, not the topic in general.

The interface is available in Arabic

In addition to French and English, the entire interface is available in Arabic, with a right-to-left layout designed for natural reading. Each user selects their language, and the platform remembers it.

More pleasant, more reliable, everywhere

The platform continues to improve, and your feedback is incorporated directly into it.

• Harmonized dates, more organized lists, and clearer navigation.
• Administrators invite a user with the appropriate workspaces and permissions from the start.
• A series of fixes based on your feedback: anonymized audio quality restored, filters that remain consistent even when the analysis changes, and improved stability for dashboards with many charts.
• Behind the scenes: continued efforts to strengthen security and data isolation, and consistent performance even with very large conversation volumes.

The interface now supports French and English

Internationalization now applies to the entire application: alerts, data sources, custom analyses, agent space, evaluation grids, knowledge bases, Live Assist, and audit logs are displayed in the language of your choice.

• Dates, durations, and numbers automatically follow the format of the selected language.
• Changing languages takes effect immediately, without reloading the page.

Answers with clickable sources

When Copilot includes verbatim quotes in its response, a “cited sources” block appears below it, with one card for each excerpt: quote preview, conversation, attribution, and date.

• Clicking a card opens the cited conversation in a window, with the relevant excerpt highlighted.
• Quotes are linked directly to real data: Copilot cannot invent a source.
• Sources remain clickable when you reopen an existing Copilot conversation.

A more accurate, more readable "Usage" page

The Usage page has been reorganized into two tabs (“AlloIntelligence” and “Transcription & Anonymization”), with a detailed count for each card and a new “Knowledge Bases” card.

• The count now accurately reflects the analysis work performed: each grid item and each custom analysis counts as one analysis.
• Rerunning an evaluation counts only the items that are actually reanalyzed: unchanged results are not counted a second time.

Usage Tracking for Live Assist

A new "Live Assist" tab on the Usage page provides insight into real-time assistance activity.

• Assisted hours, number of calls, and number of active assistants during the selected period.
• Detailed breakdown by assistant and by agent.

More Reliable AI Evaluations

• The quality of automatic evaluations has improved significantly, with a particularly marked improvement in conversations in Arabic.
• When a criterion does not apply to a conversation (no actual exchange, condition not triggered), the item is now marked “not applicable” instead of being scored at the lowest level.

Live Assist sessions via agent email

Telephony integrations can now create a Live Assist session by identifying the agent using their email address: this is useful for systems that do not know the agent's internal identifier.

Live Assist · Opt-out from recording and detection of refusal

You can now disable recording (audio and transcription) for an assistant, and detect during a call that a caller refuses to be recorded in order to stop the assistant. This is truly a matter of consent and trust.

A smoother experience for Copilot and Live Assist

You remain in control of the scrolling while Copilot is writing: you can read the beginning of an answer at your own pace.

Copilot · New Features

The now unified Copilot offers three new features: it applies all the dashboard’s filters; it returns clickable links to the conversations cited as examples; and it saves and reuses prompts along with a history of exchanges. This last feature is a welcome replacement for a previously popular function: generating dashboard summaries and saving summary templates. It provides more detailed, reusable analyses with direct access to the cited conversations.

Historical Quality Management / Customer Voice · Continuity Without Regression

Features carried over from the current tool to the new platform: VoC dashboard export, generation of PDF reports, and copying a chart's data (to retrieve all the chart information as it appeared on the previous platform).

Dashboard · Comparison, Interactivity, and Speed

The ability to compare two periods—already available for KPIs, line charts, and bar charts—has now been extended to pie charts and lists. Charts are now clickable to filter the dashboard, which is a significant improvement in usability. Additionally, request caching significantly reduces load times.

Real-time conversation translation

A user can translate a conversation into their language (FR / EN) to read, evaluate, and work with it, even if the conversation took place in another language.

Copilot · active filters, ad hoc analyses, and relevance

The Copilot now displays all active filters and can remove a specific one or reset them all at once while preserving the time period; its summary reflects the scope that is actually applied and flags any inconsistencies rather than returning an empty result. CSAT, NPS, and histogram analyses are calculated and displayed correctly; filtering by category once again returns the correct results (and prompts for clarification when a name is ambiguous); and the Copilot only displays actions relevant to the current page.

Analysis & Evaluation · Clear Comparisons and Preserved AI Categories

In a widget in comparison mode, a value missing from the previous period now displays a true 0 for count/sum metrics (averages, scores, and ratios retain the dash). And editing an analysis now only resets the AI categories for the field that was actually modified: the others retain their history, without unnecessary regeneration. This ensures readable comparisons and preserves category history.

Conversations & Agents · Identify the agent at a glance

Whenever a transcript is displayed (in an evaluation, conversation list, or alert), the agent's tooltip shows their name and a link to their agent space; the name also appears in the conversation list on the dashboard.

Live Assist · More Comprehensive Refusal Detection

Refusal detection is broader: refusal is recognized based on meaning (use of AI, recording, or analysis), rather than just set phrases. The safeguard remains in place: a question, a simple mention of AI, or a moment of hesitation does not trigger the stop.

Dashboard · New Quality Analyses

Three charts provide comprehensive quality tracking, with new applications:

• Quality score trend: See whether quality improves or declines over a period of time.
• Score distribution: the breakdown of conversations by score range—that is, the proportion of high and low scores—beyond the simple average.
• Top trends: surface weak signals—topics that are rising or falling sharply and newly emerging categories—to spot them early.

Several usability issues with the dashboard have also been resolved.

Unified Copilot: QM + dashboard all in one place

Two previously separate Copilots have now been merged: the coaching Copilot (on the Quality Monitoring side, which analyzes an agent’s performance) and the dashboard Copilot (which analyzes dashboards and performs qualitative analyses on conversations). Now unified, the Copilot can be used on a wider range of pages: you can use it on a dashboard as well as on evaluations for coaching, and cross-reference Voice of Customer data (customer-oriented) with Quality Monitoring data (agent-performance-oriented)—which makes perfect sense on a unified platform.

Agent space

A page dedicated to an agent, which their team leader or supervisor can use to focus on the agents on their team. At this stage, it lists the Live Assistants the agent has access to (and allows you to initiate a session with them directly from the page), as well as all the calls made by the agent, whether they have been evaluated or not.

Data Accuracy and Isolation

Greater isolation of analyses between workspaces, deduplication of the conversation list, more reliable filters, and synchronization during large imports. Accurate, up-to-date, isolated numbers.

Live Assist · Smarter Assistants

Increased character limit for prompts (up to 5,000) for more advanced assistants, and verification of access permissions for assisted calls.