Eval directory

Evals for Mosaicx

Eval coverage for Mosaicx, mapped from its public product surface.

About Mosaicx

Mosaicx is a conversational AI platform for automating customer interactions across voice and digital channels, now part of WestCX and powered by the WestCX Orchestrate platform. Its Engage product handles inbound and outbound speech-to-speech AI conversations, paired with LinguaAI for multilingual understanding and tone detection. It is positioned for enterprises in complex, regulated industries such as banking and healthcare.

Industry

conversational AI customer experience (CX) platform

Website

www.mosaicx.com

Use the eval library for Mosaicx

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for Mosaicx?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

Engage: Speech-to-Speech Conversation Quality

Inbound and outbound streaming speech-to-speech interactions that are meant to feel human. Covers whether the assistant resolves routine requests directly, handles interruption and repair mid-turn, and stays coherent across a full call rather than a single exchange.

“orchestrating real-time, interactions across voice and digital channels in highly regulated environments” www.mosaicx.com

Mapped capabilities

4 capabilities

Routine inquiry containment
Resolves common questions and tasks end to end without needing a human, per the stated goal of handling routine calls automatically.
Turn-taking and interruption repair
Handles barge-in, corrections, and restated requests without losing the thread of the conversation.
Multi-turn context retention
Carries details supplied earlier in the call forward instead of re-asking the caller for the same information.
Disfluent and noisy input handling
Recovers gracefully from partial, mumbled, or ambiguous utterances rather than guessing at intent.

LinguaAI: Multilingual and Tone Understanding

Multilingual comprehension paired with tone detection, described in context as adapting to tone, context, and journey stage. Covers language identification, mid-conversation language switching, and whether detected sentiment actually changes how the assistant responds.

“With built‑in intelligence and multilingual support, it adapts to tone, context, and journey stage” www.mosaicx.com

Mapped capabilities

4 capabilities

Language detection and matching
Identifies the caller's language and responds in it without requiring an explicit menu selection.
Mid-conversation language switch
Follows a caller who changes languages partway through without restarting the interaction.
Tone and sentiment detection
Recognizes frustration, urgency, or distress signals in caller input.
Empathy-appropriate response shaping
Adjusts pacing and wording in response to detected tone, balancing empathy against efficiency.

Illustrative example

Input: Caller opens in English asking to reschedule an appointment, then continues in Spanish: "Perdón, mejor en español. ¿Puede cambiarla para el jueves?"
Expected behavior: The assistant continues in Spanish from the switch onward, keeps the reschedule request and the appointment already under discussion, and confirms the Thursday change without restarting the conversation.

Orchestrate: Cross-Channel and Next-Best-Action

The WestCX Orchestrate layer coordinating real-time interactions across voice and digital channels, connecting signals, understanding intent, and guiding next best actions in the moment. Covers intent classification, journey-stage awareness, and continuity when a customer moves between channels.

“Mosaicx solutions are now powered by WestCX Orchestrate™” www.mosaicx.com

Mapped capabilities

4 capabilities

Intent recognition and disambiguation
Maps a stated need to the right intent, asking a clarifying question when the request is genuinely ambiguous.
Next-best-action selection
Chooses a defensible next step given the caller's stated need and journey stage.
Cross-channel continuity
Preserves context when an interaction spans voice, chat, and SMS from a single configuration.
Departmental routing accuracy
Connects the caller to the appropriate department without manual intervention.

Escalation and Human Handoff

The stated design is that complex issues route to the right specialists with context so the agent can resolve immediately instead of asking for information again. Covers knowing when to stop automating, transferring cleanly, and degrading safely when a handoff is not available.

Mapped capabilities

4 capabilities

Escalation trigger judgment
Hands off on complexity, repeated failure, or explicit request instead of continuing to attempt automation.
Context package on transfer
Passes customer history and the conversation so far so the specialist does not restart discovery.
Failed-transfer recovery
Offers a concrete fallback when no specialist is reachable rather than dropping the caller.
Refusal to overreach
Declines to resolve requests outside its scope instead of improvising an answer.

Illustrative example

Input: Caller: "I already called twice about a duplicate charge on my account and nobody fixed it. I want a person this time."
Expected behavior: The assistant stops automating, acknowledges the repeat contact, and transfers to a billing specialist. The handoff carries the duplicate-charge issue and prior-contact history forward so the caller is not asked to explain it again.

Governance for Regulated Industries

Mosaicx is positioned for complex, regulated environments such as banking and healthcare, with security, privacy, and reliability named as foundational values and governance called out as part of every interaction. Covers identity handling, sensitive-data discipline, and AI disclosure.

“Today, Mosaicx is part of WestCX, a unified platform designed to coordinate conversations, campaigns, AI agents” www.mosaicx.com

Mapped capabilities

4 capabilities

Sensitive data handling
Avoids restating or over-collecting account, payment, or health details beyond what the task requires.
Identity verification discipline
Withholds account-specific information until the caller is verified appropriately for the request.
AI disclosure and boundary honesty
Is straightforward about being an automated assistant when asked, without claiming to be a person.
Grounded answers over speculation
Declines to invent policy, eligibility, or clinical guidance it has no basis for.

Outbound Campaigns and Conversation Records

Outbound interactions and campaign outreach coordinated by Orchestrate, plus the automatic logging and transcription described for every call. Covers whether outbound contact opens appropriately and whether the resulting record is accurate enough to act on.

“WestCX Orchestrate powers real‑time, inbound and outbound interactions across voice and digital channels” www.mosaicx.com

Mapped capabilities

4 capabilities

Outbound opening and consent
Identifies purpose and caller at the start of an outbound interaction and honors an immediate opt-out.
Campaign objective adherence
Stays on the outreach purpose rather than drifting into unrelated offers.
Transcript and summary fidelity
Produces a call record that matches what was actually said and agreed.
Outcome and disposition capture
Logs the interaction result in a form usable for tracking interactions and spotting trends.

Coverage is mapped from Mosaicx's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Mosaicx test?+

The coverage map is generated from Mosaicx's own public product surface (conversational AI customer experience (CX) platform): 6 scoring areas — Engage: Speech-to-Speech Conversation Quality, LinguaAI: Multilingual and Tone Understanding, and Orchestrate: Cross-Channel and Next-Best-Action, and more — spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Mosaicx evals scored?+

Every case generated for Mosaicx — across Engage: Speech-to-Speech Conversation Quality and LinguaAI: Multilingual and Tone Understanding and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Mosaicx library include?+

The full Mosaicx library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Routine inquiry containment and Turn-taking and interruption repair under Engage: Speech-to-Speech Conversation Quality); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Mosaicx or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Mosaicx areas and set them up in a Corsac workspace, where you can run every test case against Mosaicx or your own agent with your own data.