Changelog

Updated Conversation Scoring

We’ve adjusted the scoring rules in consultation with customers and experts. The updated rules place greater emphasis on achieved conversation goals and fully met criteria. This makes successes more visible and encourages learners to keep practicing.

Scores for Fully Met Criteria

Fully met criteria now receive 90 to 100 points. In a comparison sample, the average score for the same conversations increased from 64.8 to 70.6 points using the updated scoring method.

The Conversation Goal Determines Success

A conversation is considered successful when the learner achieves the conversation goal. The quality of their approach continues to be assessed through phases and criteria.

This definition of success also applies to program tasks with the Only count on successful conversation option enabled.

Define Clear Goals and Criteria

• Describe the conversation goal as a specific outcome, such as: “A demo appointment has been agreed with a date and time.”
• Define conversation steps and techniques as phases and criteria. This allows the outcome and the quality of the conversation to be assessed separately.

Under Conversation Types, check that your goals clearly describe the intended outcome.

Note on Existing Reports

Existing analyses retain their scores. Higher scores following the update may result from the scoring adjustment and do not automatically indicate learning progress. Keep this change in mind when comparing results under Analytics.

Clearer conversations, even in noise

Practice conversations now feel closer to a real call. AI personas speak with a new, more natural voice, and they hear you clearly even when the room around you is loud.

A more natural voice

Personas speak with new speech synthesis. Emphasis, pauses and emotion come through more naturally, so the conversation feels less like talking to a machine. Every persona uses the new voice automatically, and you do not need to change anything.

Understands you in loud rooms

A new voice isolation step filters what your microphone picks up before the persona listens. Colleagues talking nearby, an open-plan office or a busy café no longer reach the persona. The persona stops far less often in mid-sentence because someone else spoke, and the transcript only contains what you said.

How to get the most out of it

• Use the microphone closest to you, such as a headset. Voice isolation works best when your voice is the nearest one.
• Did a persona keep interrupting you before? Try the same conversation again. It should now let you finish.
• Check the transcript in your feedback. Stray words from background voices are gone, so the scoring reflects what you actually said.

Open AI-Personas and start a conversation to hear the difference.

AI guide in practice calls

Learners now get help while they practise, not only afterwards. During a practice call, the new guide follows the conversation type and shows which phase comes next.

What learners see

• Conversation phase indicator: the recommended phase, with skipped phases marked.
• Offer answer help: learners can ask for a sentence they could say next.
• The guide only appears in practice calls. Assessments and recruiting stay as they are.
• After every conversation, learners can rate it with 1 to 5 stars and add a comment.

Turn it on

The guide is in beta and off by default. Admins switch on each part separately under Coach in practice.

Open Settings → AI-Personas to switch it on, then start a practice call.

More accurate transcripts, four new languages

Two things got better for every learner. Transcripts are more accurate, especially in Swiss German, and the interface is now available in six languages.

More accurate transcripts

• After every practice conversation, a new transcription model transcribes the full recording again.
• It hears the whole conversation, so dialect and fast speech come out far more reliably – especially Swiss German.
• Scores, objections and filler-word counts all use this transcript.
• The audio is deleted afterwards unless Keep recordings is on in Compliance.

Four new languages

Sleak is now available in Spanish, French, Italian and Chinese, alongside English and German. Personas are translated too, so learners can practise in their own language.

Open Settings → Profile and choose your language under Language and Time.

Persona Builder for every workspace

The Persona Builder is now available in every workspace. Describe who your team should practise with, and the builder drafts the persona – role, scenario and voice – for you to refine in the chat.

Build from your material

• Attach PDFs and images with Add attachments, paste them or drag them onto the chat – up to 5 files, 10 MB each.
• Switch on Web search so the builder can look up a company or an industry.
• Files are only used in that chat and are not stored.

Drafts and publishing

• Your changes stay in a draft until you click Publish. Once you publish, all users switch to the new version.
• Unpublished changes shows what changed. You can discard one change or all of them.
• The Version history lets you restore an earlier version into the draft.
• AI-Personas Management shows Published personas and Drafts in separate tabs.

Your existing personas moved over with their voices unchanged. Open AI-Personas and create a new persona to try it.

New conversation type builder

Building a conversation type is now a conversation. Describe what your team should practise, and the builder writes the phases, criteria and scorecard next to the chat, where you can refine them.

Start from your own material

Attach the documents your team already works with, such as a call guideline, a sales deck or a list of common objections. The builder reads them and turns them into phases and criteria. Ask it to search the web, for example for a sales method or for facts about your market.

Share and publish

Every change stays in a draft until you click Publish. Until then, learners keep practising with the published version. Use Share to give the conversation type to the whole workspace, to teams or to single people. Publishing also translates the scorecard into English and German.

Version history

Every publish creates a version. Open Version history to see all published versions and Restore an older one into the draft. Past conversations keep the score from the version that was live at the time.

How to get the most out of it

• Start from a template in Start from a template, or from your own material, instead of an empty chat.
• Export slides as a PDF before you attach them. The chat accepts PDFs and images up to 10 MB each.
• Write the situation in detail: who calls whom, what should happen at the end, and which objections come up. "A sales rep calls an operations lead at a restaurant group who has never heard of us. The goal is a demo appointment." gives a far better draft than "cold call".
• Refine in the chat: "weight the needs analysis higher" or "add a criterion for asking about the budget".

Open Conversation Types in the Manager Panel to try it.

Audio recordings off by default

Many teams do not want audio of their practice conversations stored. Each workspace now decides whether recordings are kept – transcripts, scores and feedback work either way.

What changes

• New workspaces keep only the transcript. The recording is deleted right after the transcript is made.
• Existing workspaces keep their recordings as before, until an admin switches this off.
• Recruiting and assessment conversations are always recorded, because they are formal evaluations.

Open Settings → Compliance and use Keep recordings under Training Sessions.

Upload Your Own Calls

Until now, Sleak analyzed the calls you ran inside the platform. Starting today, you can also upload your own call recordings and put your real sales conversations through the same AI analysis as your training sessions.

How it works

Upload a recording directly from the calls view. Sleak transcribes it, runs the full conversation analysis, and files the results alongside your training calls – scores, coaching insights, and skill-gap data included.

• Real conversations, real feedback – coaching insights based on what actually happens with your customers, not only simulations
• One place for everything – uploaded calls appear in the calls table with all existing filters, scores, and analysis views
• Team-ready – the same permissions and team scoping apply as for every other call

Open Conversations and click Import to try it with one of your own recordings.

Multi-Project Recruiting & Unified Table Filters

Two major changes ship today: a full recruiting project hub that replaces the old single-project redirect, and a unified filter system rolled out across every major table in the platform.

Multi-Project Recruiting Hub

Your recruiting dashboard is now a full project listing. From /recruiting, you see all projects at a glance — status, assigned persona, applicant count, and due date included. Create new projects in seconds, edit existing ones, or remove what you no longer need, all without leaving the hub.

• Project hub — dedicated listing page replaces the old auto-redirect to the first project
• Self-service CRUD — create, edit, and delete projects via dialogs, gated by existing permissions
• Project switcher — jump between projects directly from the applicant view header, no round-trip to the hub required
• Optional video step — when no intro video URL is configured, that step is skipped automatically in the applicant flow

Unified Table Filters

Calls, personas, manage personas, applicants, teams, and users all now share the same filter building blocks: a single Filter button that opens a structured dropdown, active filters shown as dismissible chips below the search bar, and a clear-all control. Previously each page had its own ad-hoc filter implementation.

This means you get the same keyboard-friendly, chip-based experience everywhere — and future filter additions to any table will be faster to ship.

Head to Recruiting to try the new hub, or open any table page to see the updated filters in action.

Scoped Persona Sharing, Call Filters & Analysis RBAC

Filtering your workspace just got significantly smarter. This release ships three tightly connected upgrades: scoped persona sharing filters, score-range and persona filters on the calls table, and role-based access control for the Analysis page — all backed by new database infrastructure for full-text member search and an infinite-scroll member combobox rolled out across the product.

Persona "Shared With" filter — scoped to your teams

The old visibility filter exposed internal DB values ("workspace", "restricted") that were confusing and showed teams you weren't part of. It's replaced with a Shared With filter that mirrors the public persona grid experience.

Non-admin members see only the teams they belong to. Admins and editors see all org teams. This means the filter is always relevant — no noise from teams that have nothing to do with you.

Score-range and persona filters on Calls

You can now filter your calls table by score range (0–100) and by persona. Both filters integrate into the standard filter dropdown and update the table in real time. The score-range input uses a new reusable FilterNumberRangeSub component — available to any future table that needs numeric min/max filtering.

Analysis page — permission-gated with team scope

The Analysis page now requires the analysis:read permission. Members without it are redirected to a permission error instead of seeing an empty or broken state.

Team admins can access the page but only see data for members within their managed team — enforced entirely server-side. The client cannot manipulate the scope.

Infinite-scroll member search, everywhere

A new AddOrgMemberCombobox replaces the previous member search across programs, teams settings, and user settings. It uses infinite scroll (IntersectionObserver sentinel) backed by a new useInfiniteOrgMembers hook with debounced full-text search powered by a fullName generated column added to the profiles table.

Head to Manage → Personas, Calls, and Analysis to see these changes in action.