Bene Hotline, our inbound neuro-agent, works as an AI customer service agent on your phone line. It checks who is calling, reads the order, booking or account in your systems, makes the changes you allow and logs the call in your CRM. A request that needs a person goes to one, with the case already written up. Every call is scored: the agent’s own calls with the analytics included in its package, your team’s calls with Locator, our separate speech analytics service.
Customer service calls come from people who already do business with you: a parcel on its way, an appointment, a card, a car in the workshop. The agent identifies them, reads the record and acts on it within the permissions you set. The calls quoted on this page are demo recordings from our call library, with personal data and company names masked.
Where the parcel is, what time the appointment starts, the balance and recent payments, what a policy covers, when a branch opens. The agent finds the customer by order or tracking number, or by the number they are calling from, and answers from your system and your knowledge base. On our English bank demo, once the lost card is blocked, the caller asks for the balance and hears it, with the latest payment, before hanging up. Where-is-my-order calls follow the same pattern: one lookup, one answer. Where a fact is missing from the record, the agent says so and routes the call.
It rebooks a missed delivery, moves an appointment, blocks a lost card, books a service with the extra check the caller asked for, or logs a complaint for review. On our garage demo, a caller with an amber engine light and a service due hears: “We can run diagnostics alongside the service.” The call ends with a full service booked, engine light diagnostics, a tyre tread check and collection from home. The agent writes each change into your CRM or order system, or books the slot in your calendar, with the transcript attached. What it may change, and in which systems, is agreed in writing before launch.
Which requests go to a person is agreed before launch, one request type at a time. On our bank demo, the caller wants the replacement card sent to a work address, and the agent tells them that changing an address “needs an advisor” and offers to put them through. Whoever picks up sees who called, what was asked and what the agent has already done, so the customer does not have to start again. If nobody is free, the agent books a callback. Our article on when an AI voice agent should hand the call to a human walks through that exit path.
People phone when the matter is urgent or the website has not helped. When the call goes badly, many of them never say so; they simply leave.
According to ContactBabel, the share of customers who prefer the phone for urgent, complex issues rose from 28% to 47% between 2018 and 2024, and its 2024 figures show 8.4% of inbound callers hanging up before anyone answers. Out of hours the same requests arrive with nobody to take them. After an outage, a storm or a delayed shipment, they arrive all at once. The agent answers concurrent calls up to the capacity agreed for your line, at night and at the peak alike. A caller who hangs up gives you no ticket and no name: an unanswered call leaves no record in the reports you read, so the loss never shows.
Zendesk CX Trends 2026 found that 56% of dissatisfied customers never complain and just switch, so the call is often the only signal you get. Manual quality control hears only a sample: a QA team listening by hand covers 3-5% of calls, our own figure from running live call operations. Here every call is scored. Analytics on the agent’s own calls is included in its package, and Locator reviews 100% of the calls your team takes, against a checklist agreed with you, on 30+ parameters. A fault that repeats across calls shows up in the scores, and you can open the call where it happened.
The test is whether the answer can be read from a system or a policy, and whether the change is one your staff make the same way every time.
Card services: a lost or stolen card blocked, a replacement ordered, the balance read out. Delivery: the recipient asking where the parcel is and what happened, rebooking a missed drop, checking cash on delivery. Clinics and care homes: bookings, changes, what a visit involves. Garages: a service booking, a warning light, a courtesy car or collection from home. Retail: order status and returns within your policy. The first four each have demo recordings you can play with the transcript alongside. Our page on delivery operations covers the recipient’s calls; the one for dealers and service centres, the workshop line; and the clinic page, bookings and changes.
A small business closes the line at night and at weekends; a large floor drowns at the peak. The agent takes the evening and the overflow on the same rules as the day. If you need one line covered out of hours, our AI answering service page describes that setup, and a floor that splits its queue by call type is the AI call center case. This page covers calls from customers who already have an order, a booking or an account with you. Lines run in English, French and Spanish, the three languages of our call library. The founders ran live call operations for about ten years, with up to 1,500 operators at peak, across France, Mexico and beyond.
With each client, Benerra agrees which requests go to a person before the agent takes a live call. Those requests are written into the brief the agent is built from, together with the line the agent says when one comes up.
Medical questions, exceptions to policy, compensation, a distressed caller, a complaint that needs someone with the authority to decide. On our fertility clinic demo that ends in an evening appointment, a first-time caller asks about their chances, and the agent answers: “Only the doctor can look at your full history and give you an honest answer.” In the same turn it offers two consultation slots. In a 2024 Five9 consumer survey, 75% of consumers said they still prefer talking to a real human, so the route to a person has to be short. Our article on what an AI voice agent should never be asked to handle lists the boundaries we write into a brief.
If a record is not connected, the agent cannot see it, and it says so. On our Spanish-language courier demo, the caller asks where a parcel is; the agent explains that the tracking team holds that detail, takes the tracking number for them and tells the caller they will confirm on WhatsApp. Prices, policies and exceptions have to be kept current, and that job usually sits on your side. What makes an agent improvise, and how the brief closes that off, is covered in why your AI agent invents things that aren’t in your brief on our blog.
Once the repeatable requests stop reaching people, the calls your team still takes are the longer, harder ones: the exception, the upset customer, the decision. In a 2025 Gartner survey, only 20% of service organisations had actually cut headcount. Plan for new targets, scorecards written for difficult calls, and training on the cases the agent hands over, which arrive with the history already written up. Our article on people plus AI agents sets out how to plan that division of work.
Pick a few call reasons whose answer sits in a system, prove them on real calls, then add the next ones.
Tag a week of calls by the first thing the caller asks: order status, rebooking, a card, a booking change, a complaint. Next to each reason, note the system that holds the answer (order tool, booking calendar, card platform) and whether settling it means looking up the record or changing it. Reasons that only need a lookup can go live first, those that change a record once the permissions are agreed in writing, and those that need someone to decide become the handover rules. The steps after that are laid out in where to start with AI in your calls, in four stages.
Bene Hotline, our inbound reception and routing agent, is €0.20 a minute excluding VAT, the same nominal figure in USD, with volume discounts running to −25% across the tiers up to 1M minutes a month. Latin America is priced in USD at roughly half those rates. Billing is on talk time, call analytics is included, and each tier is a monthly package of minutes. Minimum purchase: 3,000 minutes a month, at each product's Base rate. Before go-live we confirm in writing which systems are connected (telephony, CRM, order or booking tool), what the agent may write into them, where it stops and who it hands to. Go-live timing is quoted per project rather than promised as a number of days. The cost calculator runs the numbers on your own volume.
