AI Voice Agent for Restaurants: The Dinamo Steak House Case
6,709 cold calls, 1,188 conversations, and 40 warm leads ready for human handover. This is how an AI voice agent for restaurants works a cold database.
Every restaurant accumulates phone numbers in its database that no one has time to call. In this case study, we show what happens when that list is handled by an AI voice agent for restaurants: 6,709 attempts, 1,188 conversations, and 40 warm leads delivered to the manager, over two months for a "first visit" segment. The goal: reactivate a cold database without burdening staff, filtering truly interested contacts, and scheduling useful callback windows.  ## TL;DR - 6,709 outbound calls, 1,188 real conversations, and 40 warm leads filtered. - 3.4% positive rate on conversations; average duration 35 seconds. - AI qualifies and schedules callbacks; closing remains human. - Cold database cleaned: fewer numbers, more useful appointments for the team. ## The Client and Context Dinamo Steak House, located between Oleggio and Novara, specializes in lava stone-grilled meat, with Pata Negra as a signature dish and themed evenings (Cervellone, karaoke, Sarabandinamo). Over the years, it has collected thousands of contacts from promotions and events. The priority: reactivate a pool of contacts that had never been systematically worked, without taking time away from the dining room staff and marketing. ## The Challenge: Working a Cold Database at Scale - Volume: thousands of numbers, high probability of no answer. - Filtering: need to identify genuine interest before human contact. - Timing: many contacts can't talk "now" but request a callback. Reacting a database of cold contacts is repetitive and low-yield per attempt. It's ideal for automation and automated cold calling with lead qualification. ## Implementation: What the Agent "Martina" Does (and Doesn't Do) Martina, powered by DeepAgent, is configured for the "0 visits" segment. Clear scope: she doesn't sell and doesn't confirm reservations. She makes initial contact, presents the offer, verifies interest, and only passes "yes to further discussion" leads to the Communications Manager. ### Operational Flow 1. **Recognizable Opening:** Introduces herself on behalf of the restaurant and immediately explains the reason for the call. 2. **Value Proposition:** Presents the €49 Carnivore Coupon, linking it to the experience (lava stone meat, Pata Negra). 3. **Interest Verification:** Understands if there's curiosity, information requests, or disinterest. 4. **Light Qualification:** Identifies actionable signals (questions about conditions, willingness for a callback). 5. **Objection Handling:** Clarifies what's necessary, prepares for handover for commercial details. 6. **Agreed Recontact:** Gathers specific time slots (e.g., "tomorrow at 6 PM"). 7. **Human Handoff:** The manager only calls back those who have agreed to a conversation. Result: automated cold calling and lead qualification become a single efficient phone call.  ## Results: April 15 – June 17 | KPI | Result | |---|---:| | Call Attempts | 6,709 | | Completed Conversations | 1,188 | | Interested Leads (Positive Outcomes) | 40 | | Positive Conversation Rate | 3.4% | | Average Conversation Duration | 35 seconds | | Total Conversation Time | ~693 minutes · 11.6 hours | The 3.4% indicates those who expressed interest and agreed to a concrete next step (callback with the manager). This doesn't equate to coupons sold or covers: these happen later and require CRM or reservation tracking to attribute ROI. ## What People Say on the Phone From the 1,188 conversations, six recurring themes emerged, useful for future campaigns: - **How the coupon works:** inclusions, restrictions, booking, validity. - **When to call back:** precise windows turn missed attempts into appointments. - **Menu and experience:** meat, alternatives, use cases. - **Events and evenings:** Cervellone, karaoke, Sarabandinamo as independent draws. - **Identity and location:** who are you, where are you, why did I get the call. - **Reschedule ≠ rejection:** work, travel, family; a callback saves good contacts.  ## Two Priorities for the Next Campaign ### 1) Anticipate Coupon Conditions Questions about validity and restrictions are frequent: including them in the opening reduces repetitive objections, shortens call times, and focuses efforts on warm leads. ### 2) Track Post-Handoff Outcomes Connecting leads to results (activated coupon, reservation, no-show, average check) is essential for calculating ROI and correctly valuing a lead qualified by an AI voice agent. ## Impact on the Team: Less Noise, More Useful Conversations The most significant effect is not just the 3.4%, but the "cleaning" of the contact pool: rejections, no-answers, and unavailability are handled by the AI, and the human team receives a short list of requested conversations, each with an agreed-upon callback window. ## How to Replicate This in Your Restaurant - **Segment:** Start with "0 visits" or contacts inactive in the last 12 months. - **Define Agent's Scope:** No closing, yes qualification and callback scheduling. - **Clear Offer:** A trial proposition (e.g., coupon) with simple conditions. - **Script and Compliance:** Clear opening on identity, contact source, and privacy. - **Tracking:** Integrate CRM or reservation system to measure end-to-end ROI. ## Conclusion An AI voice agent works "cold" leads at scale, filters genuine interest, and prepares for commercial handover. The result: less wasted time and more qualified opportunities for your restaurant. Want to see how it works on your database? Request a demo and try a DeepAgent agent in real-time. ## Frequently Asked Questions ### How long does it take to deploy an AI voice agent for my restaurant? Typically 1-2 weeks: segment selection, script definition, flow configuration (qualification, handoff, callback windows), and testing. Automated cold calls only begin after successful testing on a small sample. With CRM integration, an additional 2-5 days are added for field mapping and reporting. ### How do I measure ROI if the agent doesn't close sales or reservations? Track the entire funnel: qualified leads, activated coupons, reservations, show rates, and average check. Integrate your CRM or reservation management system to link the 40 warm leads to actual outcomes. This way, you can assess the value of reactivating a cold contact database and accurately attribute revenue to the AI. ### Can the agent handle complex objections or detailed commercial requests? It handles frequent objections (price, validity, booking) and gathers signals of interest. For specific conditions or negotiations, it performs a handoff to the manager. This separation keeps conversations brief and maximizes lead qualification, leaving negotiation and closing to the human team. ### What minimum volumes make sense for automation and outbound? It makes sense from a few thousand contacts upwards, especially if they haven't been contacted recently. For high volumes, the AI voice agent maintains consistent coverage, schedules callbacks, and scales without burdening the team. Even with smaller lists, it's useful for testing offers and messages before handing over to the team. ### Does the AI respect privacy and consent in cold calls? Yes. Transparent opening about identity and purpose, referencing the contact source, and the option for immediate opt-out are part of the script. Data is managed according to regulations, with conversation logs and respect for contact preferences. This increases trust and reduces friction in automated cold calls.