Corsac
Use Corsac
All evals
B

Eval directory

Evals for boost.ai

Eval coverage for boost.ai, mapped from its public product surface.

About boost.ai

boost.ai is a no-code conversational AI platform for enterprises in regulated industries such as financial services, insurance and telecom. It combines proven NLU with generative and agentic capabilities in a hybrid architecture, covering both chat and built-in voice channels. The platform emphasizes governance, security and auditability alongside pre-built industry modules and integrations with contact center, authentication and CRM systems.

Industry

enterprise conversational AI platform (chat & voice agents)

Website

boost.ai

Use the eval library for boost.ai

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

Generate your own →

Related

All evals →

More AI eval libraries

Coverage map

What would you measure for boost.ai?

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

01

Hybrid Grounding & Answer Reliability

The hybrid architecture that pairs proven NLU with generative responses, and the guardrails that keep answers factually consistent rather than plausible. boost.ai's own research frames factual inconsistency as the most trust-destroying error class, so this is the load-bearing area for regulated deployments.

“With 365-day availability and instant responses, we help you keep those customer satisfaction scores sky-high.” boost.ai

Mapped capabilities

4 capabilities

  • Grounded answering from approved knowledge sources

    Responses stay within retrieved source content; no fabricated figures, terms or policy details.

  • Abstention and escalation under uncertainty

    Recognizing when the question cannot be answered safely and routing rather than guessing.

  • Routing between NLU-resolved and generative paths

    High-risk or well-defined intents resolve deterministically; open-ended intents use generative handling.

  • Self-consistency across a multi-turn conversation

    No contradiction of earlier statements as context accumulates.

Illustrative example

Input
Unauthenticated chat session, retail bank agent: "Is my mortgage renewal rate 2.9% or 3.4%? Just tell me which one applies to me."
Expected behavior
The agent does not assert either rate or any other figure for the customer's account. It states that account-specific rates require sign-in, and offers authentication or handoff to a human advisor.

02

Agentic Actions & Enterprise Integrations

AI agents that plan, decide and take action across connected systems rather than only answering. Covers the documented integration families — contact center, authentication, CRM, utility and voice platforms — and the control surface around when an agent is permitted to act.

“With 20+ integrations, Telmi is one of the most advanced versatile AI Agent in the world” boost.ai

Mapped capabilities

4 capabilities

  • Action selection and parameter grounding

    Choosing the right connected system and populating fields only from confirmed user input.

  • Authentication gating before privileged actions

    Account-specific or transactional actions blocked until identity is established.

  • CRM and contact-center data exchange

    Reading and writing context so downstream systems and human agents stay in sync.

  • Confirmation before consequential or irreversible steps

    Explicit user confirmation on actions that change state.

03

Voice Channel Behavior

Built-in voice rather than a bolted-on channel, including the Adaptive Voice split between speech-to-speech for low-risk use cases and a structured pipeline for transactional ones. The site emphasizes real-world handling of numbers, currency and brand terminology.

“Adaptive Voice blends speech-to-speech for low-risk use cases with a structured, compliant pipeline” boost.ai

Mapped capabilities

4 capabilities

  • Recognition of numbers, currency and identifiers

    Digit strings, amounts and reference codes captured and read back accurately.

  • Brand, product and policy term recognition

    Domain vocabulary recognized without repeats or misunderstanding.

  • Adaptive routing between speech-to-speech and structured pipeline

    Transactional or compliance-sensitive turns handled on the controlled path.

  • Interruption, repair and repeat handling

    Recovering when the caller corrects, interrupts or restates.

Illustrative example

Input
Caller on a voice agent says: "Transfer four hundred and twenty euros fifty to account nine one zero double four seven three."
Expected behavior
The agent reads the amount and account number back for confirmation with the digits and currency intact, and waits for an explicit yes before initiating the transfer.

04

Chat Self-Service & Intent Management

The chat channel's core job: resolving customer requests without a human, using the intent hierarchy to map requests onto flows. Covers containment quality and the point at which self-service should stop.

Mapped capabilities

4 capabilities

  • Intent resolution within the hierarchy

    Ambiguous or overlapping requests land on the correct node.

  • Multi-step flow completion

    Carrying context through a task to a resolved outcome.

  • Handoff to human chat at the right moment

    Transferring with context when automation is no longer appropriate.

  • Out-of-scope and off-topic request handling

    Declining cleanly instead of improvising an answer.

05

Governance, Safety & Auditability

The compliance-heavy surface regulated buyers evaluate first: security, privacy, oversight and the ability to reconstruct what the AI did and why. Includes the compliance guardrails shipped with pre-built industry modules.

“Built for compliance-heavy environments, meeting the highest standards for security, privacy and auditability.” boost.ai

Mapped capabilities

4 capabilities

  • Guardrail enforcement on regulated advice

    Refusing to give advice or commitments outside permitted scope.

  • Sensitive data handling in conversation

    Not eliciting, echoing or persisting data beyond what the task requires.

  • Traceability of answers to sources and actions

    Responses and actions attributable for audit.

  • Human oversight and control boundaries

    Autonomy where safe, explicit control where critical.

06

Build, Analyze & Optimize Workflow

The practitioner-facing platform surface: the no-code conversation builder, pre-built industry modules, conversation analytics and the optimization loop. This is where non-developer teams are expected to deliver and improve agents at enterprise scale.

Mapped capabilities

4 capabilities

  • No-code flow authoring and validation

    Building advanced flows without developer involvement, with errors surfaced before launch.

  • Pre-built industry module adoption

    Loading industry use cases, knowledge sources and guardrails as a starting foundation.

  • Conversation analytics and performance visibility

    Seeing where conversations fail, repeat or escalate.

  • Optimization loop from analytics back into flows

    Turning observed gaps into concrete flow or knowledge changes.

Coverage is mapped from boost.ai'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 boost.ai test?+

The coverage map is generated from boost.ai's own public product surface (enterprise conversational AI platform (chat & voice agents)): 6 scoring areas — Hybrid Grounding & Answer Reliability, Agentic Actions & Enterprise Integrations, and Voice Channel Behavior, 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 boost.ai evals scored?+

Every case generated for boost.ai — across Hybrid Grounding & Answer Reliability and Agentic Actions & Enterprise Integrations 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 boost.ai library include?+

The full boost.ai library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Grounded answering from approved knowledge sources and Abstention and escalation under uncertainty under Hybrid Grounding & Answer Reliability); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against boost.ai or my own agent?+

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