An AI phone assistant that does more than take your number

It happened to me just recently. Early one morning I called a company with a simple question: can you help me with this? Nobody was in the office yet, so I ended up with an external answering service. That in itself is customer-friendly: at least someone picks up, instead of a phone that keeps ringing or a voicemail. A friendly operator took my name and number and promised that someone would call me back. That was all she could do: she knew nothing about the company, let alone about my question.

So that company pays per call for a notepad. There's a better way.

At Off the Radar, the AI conference in Brussels, voice was one of the big topics. The expectation was clear: in the coming years, talking to AI will become as normal as chatting with it.

Two demos, two sectors

We built two demos for potential clients, in two sectors that have nothing in common. In neither of them will you hear "press 1 for appointments, press 2 for billing". You simply ask your question, the way you would at a front desk.

In both cases the goal is the same: your team gets more time for work that really adds value, instead of answering the same questions all day. The assistant is a safety net, not a replacement.

What already works well?

A first version is up and running quickly. What the assistant needs to know comes from information your company already has: your opening hours, your services, the questions customers ask most, and the agreements on what it can and cannot say. After a few days of testing and fine-tuning, we had an assistant you can confidently let a customer hear.

The voices are also much better than the robotic voices of the past. There are Flemish voices, and some of them sound genuinely natural.

Behind the voice sits a language model, and there's a lot to choose from. That makes a big difference. With a cheaper model, things sometimes went wrong on something as simple as "it's Sunday, so the workshop is closed".

Where it still falls short

It's no silver bullet. Reading out numbers and license plates correctly, and getting a friendly tone, took some fine-tuning. And the demos work with sample data. In practice, the assistant needs to check a real calendar or book an appointment in the planning system. Those integrations take the most work, and that's exactly what we do every day.

It has to really know your company

The most important lesson wasn't about technology. A good assistant knows your company like an experienced colleague at the front desk. It knows not only what you do and what customers often ask, but also what you stand for and how you talk to customers. Formal or informal? Short and businesslike, or a little warmer? What would you never promise on the phone? You write that down once, and it sticks to it in every call.

On top of that, a few rules apply to every assistant:

• It doesn't make things up. If the answer isn't in the information you gave it, it says it doesn't know and puts you in touch with a colleague.
• Check first, then offer. It checks whether a department is open before offering to transfer you. Nothing is more annoying than "I'll put you through" followed by "sorry, they're closed".
• It repeats what matters. It reads back a license plate or a callback number and asks you to confirm before moving on.
• It's honest. Ask whether you're talking to a computer, and it says yes and keeps helping you.

And now for ourselves

At Sevendays, too, the phone sometimes rings when nobody is available: in the evening, on the weekend, or because we're all on another call. That's where our own assistant comes in. And who calls varies quite a bit:

• A client with a new idea. For example, a process they'd like to automate. The assistant asks follow-up questions the way we would: how does it work today, how much time does it take, who works with it? That conversation is processed the same way as an email, as I described in five AI colleagues working with me day and night. So the next morning there's a well-defined request waiting, instead of just a name and a number.
• A company that doesn't know us yet. It explains how we'd approach it and asks follow-up questions where needed: what software do you already use, how many orders do you process per day? Then it suggests a meeting.
• The salesperson trying to sell us something. They're politely asked to send an email.

Just as important is what it doesn't do. It doesn't share a list of our clients, doesn't quote prices for custom work, doesn't reset passwords and doesn't give out the team's mobile numbers. Not even when someone asks very cleverly.

Try it yourself

Our own digital assistant knows what we do: helping companies work more efficiently and become more customer-friendly, with custom applications, AI and integrations between their systems. Press the button and ask it a question about your own business. A call lasts up to two minutes. For example, ask it:

• "We still plan our crews in Excel. Can you build an application for that?"
• "Every day we retype order forms from emails. Can AI take that over?"
• "Our webshop and our accounting software don't talk to each other. Can you connect them?"
• "Our customers often call to ask where their order is. How do we solve that?"

Is this a fit for your business?

Many companies forward their phone to an answering service after office hours. Imagine that service could actually answer: whether they can help you with your question, whether a part is in stock, when your car will be ready. A call outside office hours then stops being a missed opportunity and becomes a customer who got help.

Start small. Pick one time of day when you miss calls today, for example after 6 p.m., and the five questions people ask most at that time. Write down what the assistant can answer and when it should bring in a colleague. That's enough for a first version you can test yourself.

Curious what that would look like for you? Feel free to ask your question below.