AI Automation

Sales Intake

Practical AI

Lead Qualification

Product Design

Our "Book a Call" Button Was Asking the Wrong Question

4/17/2026

6 min read

Every small consultancy has the same button. Ours said "Book a Call." It dropped people straight into our calendar. They'd pick a time, and 30 minutes later we'd be on a video call asking, "So, tell me about your business."

The button did what it was designed to do. The problem is we were asking it the wrong question.

---

The Setup

For a while, "Book a Call" was our funnel. One click from the nav bar. Friction stayed low, submissions came in steadily, on paper it was fine.

In practice, every call started from zero. Neither side knew anything about the other. The prospect hadn't thought much past "AI sounds interesting." We hadn't thought much past "looks like a lead." The first third of every conversation was spent figuring out whether we should even be talking.

And the misses were predictable. Two or three calls a week we'd know within a minute: wrong stage, wrong problem, wrong expectation. We'd spend the remaining 29 minutes being polite.

---

The Form We Tried

The obvious fix was to put more stuff before the calendar. We built a form — ten questions, a mix of dropdowns and free text. Submit it, then pick a time.

Submissions fell off a cliff.

The reason is obvious in hindsight. A form before the calendar reads like a job application. You haven't decided we're worth your time, and we're handing you paperwork. The asymmetry is wrong: we're asking you to qualify yourself to us before we've given you anything that makes us worth qualifying for.

We rolled it back within two weeks.

---

The Reframe

The thing we kept missing: we were arguing about when to put the form — before the button, or after. The real answer was that the form shouldn't exist as a separate step at all. Qualification doesn't belong before the booking or after it. It belongs inside it.

The "Book a Call" button itself needed to change. Not what surrounded it — the button.

---

What We Built Instead

The button now opens a chat with an AI co-pilot. It asks about your business, one question at a time, and adapts based on your answers. Typical session is about 5 minutes.

It doesn't feel like a form. It feels like the beginning of the conversation you came for.

The co-pilot's job isn't to qualify in the scoring sense — there's no "do they have budget" check. Its job is to extract the context that would otherwise eat the first ten minutes of the actual call. What's the business? What's eating time? What have you tried? What would "this is working" look like in 90 days?

At some point, when the co-pilot has enough to work with, the flow switches. A slot picker appears right there in the same view — no redirect, no separate calendar page, no context switch. The booking is part of the same conversation.

The transcript goes straight to us. When the call lands on our calendar, the context is already attached.

---

Why Chat Beats a Form

Two things surprised us once it was running.

Adaptivity is most of the value. The co-pilot doesn't ask the same five questions to everyone. It follows up. If you say you're drowning in quote requests, the next question is about quote requests. If you mention a specific tool, the next question is about that tool. A form can't do that. A form asks question four whether or not question three made it obsolete.

Momentum matters more than information. A form breaks momentum — you see ten fields and decide whether it's worth the cost. A chat keeps momentum — each answer costs you one sentence and produces the next question. You're already talking by the time you'd have abandoned the form.

The net result isn't that we collect more information than a form would have. It's that we collect roughly the same amount, from more people, and it's more relevant because each question responded to the previous answer.

---

Why a Human Still Reads the Answers

One small decision we made early: there's no LLM-generated brief in the middle. The transcript goes straight to us. We read it before the call.

We considered adding a summarization step — "here's a three-bullet brief, go into the call with this." It was tempting. But we're a small team, and the 60 seconds it takes to read a transcript is cheap insurance against the model editorializing something the client meant differently. For now, the model's job ends when the chat does.

If we were a 50-person team, we'd probably automate the read. At our size, the transcript is the brief.

---

What Actually Changed

No hard numbers yet — we haven't run it long enough for honest stats. Qualitatively: calls start with context. Bad-fit conversations surface earlier, sometimes inside the chat itself. The energy of a 30-minute call when both sides walk in informed is nothing like the energy of a 30-minute call that starts with "So, tell me about your business."

That's the shift we were chasing. Everything else is downstream of it.

---

If You're Copying This

A few things we'd tell someone building their own version:

• Don't bolt a chatbot onto the booking page. Replace the button, don't add a popup next to it. One decision point, one flow.
• Keep it short. Five questions, five minutes. If your chat asks fifteen things, you've just rebuilt the form you replaced.
• Let the model follow up. The adaptivity is the unlock. Scripted question trees lose the advantage.
• Put the slots inside the chat. If the user has to leave the conversation to book, you've reintroduced the problem.
• Read the transcripts. At least at first. The model will surprise you in both directions.

The "Book a Call" button was a 15-year-old UI pattern we had stopped questioning. Replacing it with something that actually talks back turned out to be one of those small changes that quietly fixes a much bigger problem.

---

If you want to try it yourself, our book-a-call button now opens a 5-minute chat. You can start one here.