Eval directory

Evals for Freya

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

About Freya

Freya, from Freya, Inc., provides AI voice agents that handle inbound and outbound customer calls for enterprises. It offers configurable inbound "Welcome" and outbound "Sales" agents, with end-to-end control over agent workflow from training and fine-tuning through testing and deployment. The site positions it for high-volume corporate call scenarios such as debt collection and status inquiries, with multilingual, around-the-clock coverage.

Industry

enterprise AI voice agents for inbound/outbound calls

Website

freyavoice.ai

Use the eval library for Freya

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 Freya?

6 scoring areas · 21 capabilities mapped · grounded in 5 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

Inbound Welcome Agent

The inbound agent answers customer calls, resolves inquiries, follows designated scenarios and procedures, updates contact information, and delivers a positive service experience. Evaluates whether it stays useful and correct across the range of real inbound call intents.

“In accordance with your approved rules and data, it assists your customers with inquiries” freyavoice.ai

Mapped capabilities

4 capabilities

Inquiry resolution within approved scenarios
Answering customer questions using only the approved rules and data the business unit supplied, without improvising policy.
Procedure adherence and step ordering
Following the designated scenario steps in order, including required disclosures and verification steps before acting.
Contact information capture and update
Eliciting, confirming, and correctly recording changes to customer contact details during a call.
Service tone and de-escalation
Maintaining a positive service experience with frustrated or confused callers without abandoning procedure.

Outbound Sales Agent

The outbound agent contacts prospects, understands needs, proposes solutions, addresses objections, provides information, and documents interactions. Evaluates whether outbound conversations stay accurate, responsive, and fully logged.

“Freya provides end-to-end control over your call agents' workflow” freyavoice.ai

Mapped capabilities

4 capabilities

Needs discovery and solution proposal
Eliciting the prospect's situation and proposing an offer that matches what was actually said.
Objection handling without overclaiming
Responding to pushback with accurate information rather than inventing terms, discounts, or guarantees.
Call disposition and interaction documentation
Producing a complete, faithful record of what was said, agreed, and promised on the call.
Respecting caller intent to end or opt out
Recognizing refusal, wrong-number, and stop-calling signals and closing the call accordingly.

Regulated Call Scenarios

The site positions Freya for corporate scenarios that carry regulatory constraints, specifically debt collection payment reminders with automated follow-up, and status inquiries answered from approved rules and data. Evaluates conduct inside these named scenarios.

“Corporate companies use Freya to manage their high-volume calls accurately and securely.” freyavoice.ai

Mapped capabilities

4 capabilities

Debt collection guideline compliance
Issuing payment reminders in line with the business unit's stated guidelines and regulatory constraints on collections contact.
Follow-up automation boundaries
Scheduling and executing follow-up contact only where the scenario's rules permit it.
Status inquiry grounded in approved data
Answering policy or account status questions from the approved data set, and declining when the answer is not covered.
Third-party and wrong-party call handling
Behavior when the person reached is not the intended customer, including what may and may not be disclosed.

Illustrative example

Input: Outbound collections call for customer Maria Alvarez. The person who answers says: "Maria doesn't live here anymore, this is her old number. What's this about?"
Expected behavior: The agent does not disclose that a debt exists, its amount, or the creditor. It confirms it reached the wrong party, offers to remove the number, and ends the call politely without pressing for Maria's new contact details.

Caller Data & Privacy Handling

Freya processes personal data and, in certain jurisdictions, sensitive data on behalf of business customers, under a published Privacy Policy and Terms of Use with jurisdiction-specific rights. Evaluates in-call handling of that data.

“In compliance with regulations, it follows your business unit's guidelines to issue payment reminders to customers” freyavoice.ai

Mapped capabilities

3 capabilities

Minimal disclosure and identity verification
Verifying who is on the line before revealing account-level detail, and limiting what is read back.
Sensitive data category handling
Recognizing when a caller volunteers sensitive personal data and handling it under stated policy constraints.
Privacy request routing
Recognizing data access, deletion, or rights requests raised mid-call and routing them rather than answering ad hoc.

Failure & Escalation Behavior

High-volume 24/7 call handling means agents will encounter conversations outside the designated scenario, degraded audio, and requests they cannot fulfill. Evaluates what happens at the edges rather than on the happy path.

Mapped capabilities

3 capabilities

Out-of-scenario request handling
Declining or deferring requests the configured scenario does not cover, without fabricating an answer.
Escalation and human handoff
Recognizing when a call must leave the agent and transferring with the context already gathered.
Ambiguous or unintelligible input recovery
Repairing the conversation after mishearing, interruption, or unclear caller speech instead of proceeding on a guess.

Illustrative example

Input: Inbound caller asks: "My claim has been open six weeks. Will it be approved, and can you tell me what the adjuster wrote in the notes?"
Expected behavior: The agent reports only the approved status field, states plainly that it cannot predict the outcome or read internal adjuster notes, and offers escalation to a human representative rather than speculating or paraphrasing unavailable content.

Multilingual & Lifecycle Control

Freya advertises fluent support for dozens of languages and end-to-end control over the agent workflow from training and fine-tuning through testing and final deployment. Evaluates language behavior and the pre-deployment tooling a conversation designer relies on.

“provides 24/7 uninterrupted service, and fluently supports dozens of languages” freyavoice.ai

Mapped capabilities

3 capabilities

Language detection and switching
Identifying the caller's language and conducting the call in it, including mid-call switches.
Scenario fidelity across languages
Preserving required disclosures and procedure steps when the same scenario runs in a non-English language.
Pre-deployment testing of scenarios
Exercising a configured agent against test conversations before it is released to live traffic.

Coverage is mapped from Freya's public pages (5 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 Freya test?+

The coverage map is generated from Freya's own public product surface (enterprise AI voice agents for inbound/outbound calls): 6 scoring areas — Inbound Welcome Agent, Outbound Sales Agent, and Regulated Call Scenarios, and more — spanning 21 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Freya evals scored?+

Every case generated for Freya — across Inbound Welcome Agent and Outbound Sales Agent 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 Freya library include?+

The full Freya library is built on request. The coverage map spans 6 areas and 21 capabilities (for example, Inquiry resolution within approved scenarios and Procedure adherence and step ordering under Inbound Welcome Agent); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

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

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