Eval directory

Evals for Slang AI

Eval coverage for Slang AI, mapped from its public product surface.

About Slang AI

Slang AI is a voice AI phone answering platform purpose-built for restaurants. It answers inbound calls around the clock, responds to common guest questions, and books, changes, or cancels reservations through integrations with platforms like OpenTable, SevenRooms, Yelp, and Tripleseat. It also routes VIP calls, sends staff alerts, and provides call recordings, summaries, and guest-satisfaction insights.

Industry

restaurant voice AI phone answering

Website

www.slang.ai

Use the eval library for Slang AI

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 Slang 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

Reservation Management

End-to-end voice booking: creating, changing, and canceling reservations, capturing the details a host would capture, and handling the case where the requested slot is unavailable.

“Empower guests to make, change, or cancel reservations, across multiple reservation platforms” www.slang.ai

Mapped capabilities

4 capabilities

New booking capture
Elicit and confirm date, time, party size, and guest name; hold the corrected value when the caller revises mid-sentence.
Modify and cancel flows
Locate an existing reservation from caller-supplied details and apply a change or cancellation without creating a duplicate.
Availability and alternatives
Offer nearby times when the requested slot is full, and surface sister-location openings for restaurant groups.
Special requests
Capture allergy, occasion, and seating notes as structured reservation-level notes rather than dropping them in conversation.

Illustrative example

Input: Hi, can I get a table for eight this Saturday around seven? Oh wait — actually make it nine, my brother's coming too.
Expected behavior: The agent books for nine, not eight, and reads the final party size back to the guest before confirming. If nine does not fit the requested slot, it offers an alternative time rather than silently booking the smaller party.

Guest Question Answering

Prompt-driven answers to the common inbound questions that are not reservations — the surface Slang AI describes as built from analysis of guest call volume.

“Slang AI answers every inbound call, instantly responds to guest inquiries, and automatically manages reservations – 24/7.” www.slang.ai

Mapped capabilities

4 capabilities

Operating details
Hours, address and directions, parking, and closures answered from configured restaurant data.
Menu, allergy, and promo questions
Answer from what the restaurant supplied; decline or defer when the detail was never configured.
Text link follow-up
Send the right artifact — menu, wine list, ordering link, confirmation — as an SMS when the answer is better delivered in text.
Out-of-scope handling
Recognize questions the configured prompts do not cover and move the caller toward a human instead of guessing.

Call Routing & Handling

How a call is directed: who gets it, when, and under what schedule. Covers the operator-configured routing rules rather than the conversation itself.

“Our in-house Conversational Design experts have spent countless hours poring over 17 million+ guest calls” www.slang.ai

Mapped capabilities

4 capabilities

VIP identification and routing
Recognize a known frequent diner and route past the concierge flow to staff or a concierge line.
Business-hours vs after-hours behavior
Apply the correct schedule so after-hours calls are captured rather than forwarded into an unstaffed line.
Forwarding and transfer
Hand off to the configured destination cleanly, including what the guest hears during the transfer.
Escalation on guest insistence
Route to a human when the caller asks for one, without looping the guest back into the automated flow.

Reservation Platform Integrations

Fidelity of the writes and reads against the booking systems Slang AI names as official integrations: OpenTable, SevenRooms, Yelp, Tripleseat, and Fishbowl.

“Slang AI is the only voice platform that directly measures CSAT — consistently delivering 96%+ guest satisfaction” www.slang.ai

Mapped capabilities

4 capabilities

Booking write-through
The reservation confirmed on the call matches what lands in the connected platform, field for field.
Confirmation delivery
Voice confirmation and text confirmation agree with the booked record.
Notes and requests sync
Special requests captured by voice appear on the reservation in the downstream platform.
Integration failure behavior
When the booking platform is unreachable, the guest is told honestly and no phantom confirmation is issued.

Staff Alerts & Priority Topics

Detection and routing of the high-priority call topics Slang AI alerts staff on, distinct from ordinary reservation traffic.

“Staff receive real-time alerts for high-priority topics like private dining, guest complaints, and lost items.” www.slang.ai

Mapped capabilities

4 capabilities

Private dining and events
Recognize large-party or event inquiries as a sales lead and alert rather than force a standard booking.
Guest complaints
Detect dissatisfaction and raise it to staff instead of resolving it inside the automated flow.
Lost items
Capture the item, table, and timeframe and notify staff for follow-up.
Alert precision
Do not fire priority alerts on routine calls that merely mention a keyword.

Illustrative example

Input: I think I left my jacket at your bar last night, table near the window. Can someone check for me?
Expected behavior: The agent recognizes this as a lost-item report, collects the item description and approximate time, tells the guest a staff member will follow up, and triggers a staff alert. It does not attempt to open a reservation flow.

Insights, Recording & CSAT

The post-call data layer: recordings, summaries, topic and cover reporting, and the guest-satisfaction measurement Slang AI captures natively on calls.

“Capture native customer satisfaction scores (CSAT) on every call.” www.slang.ai

Mapped capabilities

4 capabilities

Call summarization
Summary reflects what actually happened on the call, including the outcome and any unresolved ask.
Topic and outcome tagging
Calls are classified into the reporting categories operators act on — reservation, question, complaint, transfer.
CSAT capture
Satisfaction is solicited and recorded without degrading the end of the guest experience.
Cover and answer-rate reporting
Reported covers and answer rate reconcile with the underlying call and reservation records.

Coverage is mapped from Slang 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 Slang AI test?+

The coverage map is generated from Slang AI's own public product surface (restaurant voice AI phone answering): 6 scoring areas — Reservation Management, Guest Question Answering, and Call Routing & Handling, 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 Slang AI evals scored?+

Every case generated for Slang AI — across Reservation Management and Guest Question Answering 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 Slang AI library include?+

The full Slang AI library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, New booking capture and Modify and cancel flows under Reservation Management); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

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

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