Skip to content
← SignalSpring 2026 No. 23
Positioning4 min

A chat flow is just a form with mood lighting

The brief

A scripted chat flow looks like a conversation. It behaves like a menu. The visitors you most want to hear from are the ones whose questions never fit a menu.

A potential client is scrolling Instagram at 11pm and stops on your post. Comment GUIDE for the free workbook, you wrote. They want the workbook. They comment.

The DM comes back: Hi! What's your email so I can send it over?

They send the email. They also add: Quick question while I have you — does this also cover [the specific thing they actually came in for]?

The next DM ignores the question: Great! What's your biggest priority right now? Reply A, B, C, or D.

They try again: Sorry — I just wanted to ask about [the specific thing].

To make sure I send you the right resource, please reply A, B, C, or D.

By the third loop, they've left the DM open and moved on. They won't come back. They won't dig further. They won't send a cold follow-up asking what they actually wanted to know. The window closed.

The names you'll never know

This is the version of you that meets them now. A flow that can only answer what someone could have answered for themselves on the site. A handler that breaks the moment the visitor brings their own language to the door.

The harder part is that you paid for this on the way in, and you'll pay for it again on the way out.

To set up that system, someone — maybe you — built a decision tree. Mapped the intents. Wrote the fallback. Tested the paths. Trained it on the five inputs that cover ninety percent of inbound, and you knew at the time that the other ten percent would hit a dead end and leave. You shipped it anyway because every alternative was harder.

So now you have a system that does two things at once. It handles the boring inbound. And it filters out the interesting inbound. Anyone whose question is exactly the question you imagined gets through. Anyone whose question is anything more specific — and the more specific questions are usually the better ones — bounces.

You will read the analytics later and see strong session counts and reasonable completion rates, and decide it's working.

It's working for the visitors you already knew how to serve. It's failing — quietly, without ever logging a complaint — for the visitors you'd actually want to hear from.

And it's failing where you can't see it. The analytics tell you that someone bounced. They don't tell you who that person was, what they were actually asking, or whether the question was sharper than your tree knew how to answer. You will never know the names of the visitors your system filtered out on your behalf.

Same room, louder doorman

The static link page was the wrong room for them. The scripted chat flow is the same room with a louder doorman.

Both share the same flaw. They assume the visitor came already shaped to fit your menu. The visitors who matter most are the ones who didn't.

You can spend another quarter tuning intents. You can pay for the upgraded tier with multi-turn fallback. You can add a sixth path and a seventh path to the tree. The shape of the problem won't change. A chat flow is just a form with mood lighting. The structure is still wrong; the visitor still has to translate themselves into your categories before you'll respond.

Meeting them where they are

What would change isn't the conversation. The conversation was the right idea. The visitor is already in motion, already in conversation with themselves about why they're here — meeting that with another conversation is the right shape.

What would change is what's on the other side of it. That's what we've been building. It's called Parlei.

Parlei meets each visitor dynamically — based on why they actually showed up, not on which keyword triggered the flow. If they came for the workbook, they get the workbook. If they came with a sharper question along the way, the sharper question is the one that gets answered. The visitor doesn't have to translate themselves into A, B, C, or D before they're heard.

And then — quietly, after the visitor has gone — Parlei sends you a brief. Who came by. What they were asking. What got said. Not opt-in counts. Not completion rates. The shape of the visitor and the shape of the conversation, summarized for someone who wasn't there.

The visitor you want

The person who lands at 11pm with a specific question is the visitor you want. They've already shown you they care enough to ask sharply.

The least you can do is have something on the other side of that question that doesn't make them pick from four buttons.

If the inbound problem has felt like a tone-of-voice problem, or an hours-of-operation problem, or a contact-form problem — and a chat flow looked like the next thing to try — it might be worth pausing before you wire up another tree.

The thing on the other side of the link is allowed to be more like a person than that.

Continue reading