Chatbot lead capture: every field has a price

How many fields, at what moment, into whose inbox. Three separate decisions, usually collapsed into one form, and the third one quietly governs the other two.

Two hundred addresses in a spreadsheet and nine replies. That is the usual result of a form that opens with the chat window and wants a name, an email and sometimes a phone number before the visitor has asked anything, and it is a result no amount of rewriting the field labels improves. An address is a price, and the visitor sets it. They pay it when something has already changed hands and refuse it when nothing has, which means lead capture is really three decisions that most shops collapse into a single form: how many fields you ask for, at what moment you ask, and whose inbox the record lands in. The third one governs the other two. A perfectly timed two-field ask that lands in a dashboard nobody opens has collected nothing at all.

Start by counting last month's bounces

Before you change a setting, get the tell. Take the addresses you captured last month, wherever they landed — your inbox, a sheet, your CRM — and send them something real. A plain email from a person on your team, answering the thing they asked about. Not a tracking ping, not a newsletter blast.

Then count three things. How many hard-bounced. How many are role addresses like info@ or a domain nobody reads. How many were typed to get past the field: asdf@asdf.com, the same address six times, a first name spelled with one letter. A workable line is one in ten. If more than one address in ten fails one of those tests, you do not have a lead list. You have a toll gate that visitors paid in fake currency.

That count is your baseline, and it is the only number worth defending. A month that captures half as many addresses and bounces at a quarter of the rate is a better month, even though the total went down. Measure lead capture by volume and you will keep tuning toward junk, because junk is easy to collect and real interest is not.

Form-first fails because nothing has been exchanged yet

A visitor who has been on your site for forty seconds owes you nothing. No answer, no price, no size confirmation, no shipping date has changed hands. Asking for an address at that point is asking someone to pay before they know what they are buying. Two things happen. They close the tab, or they type whatever gets them past the field. In your reporting those look like the same event until you check which addresses were real, which is exactly why a toll gate looks like it is working.

So answer first. Once the assistant has said the 42 is in stock, that shipping to their country is $9 and takes four days, or that the price at 200 units is not the price on the page, an address gets cheap. They want the follow-up now, and the follow-up is what the address buys.

This is also why the chat has to be able to answer before capture is worth switching on at all. The assistant has your catalog and your written policies to work from and nothing else, so the ceiling on answering first is precisely how much of your shop exists in writing. A visitor who gets "I don't have that information" and is then asked for their email address has been charged for nothing, and will price the next ask accordingly.

Fields cost completions, and the cost is not linear

Name and email get filled in. Name, email, phone, company and budget do not. The fifth field is not five times worse than the first. Somewhere around the third, a visitor stops filling in a short form and starts deciding whether to bother at all. Watch that in your own numbers before you trust anyone else's, including this page: run two fields for a month, count the bounces, then run four and count again.

Phone is the sharpest example. A phone number inside a text conversation reads as a request to be called, and someone who chose to type at you rather than call you has already said what they want. Company and budget have a milder version of the same problem. They read as qualification, and a browsing visitor does not think of themselves as a lead being qualified.

Collect the minimum that lets you follow up. For most shops that is a first name and an email, and nothing else. The company, the quantity, the delivery date, the budget: everything you wanted on the form comes out of the reply, because you will now be able to have that conversation. A field is a question asked at the worst possible moment. The same question in a reply is asked while someone is already talking to you, and it gets a real answer.

Required, optional and off are three different jobs

Required means details before any answer. It converts a browsing visitor into a closed tab, so it belongs on a narrow set of cases: a quote request, a custom order, a wholesale inquiry where the next conversation costs an hour of someone's time either way. On a product page it is a loss, and a loss you will never see, because the people it cost you do not show up in any list you can open.

Optional means the assistant answers, then offers the exchange. You capture fewer addresses and a much larger share of them are real, because everyone who gave you one chose to. This is the right default for almost every shop. If you change one setting after reading this, change required to optional, then re-run the bounce count in a month and compare the two.

Off is a legitimate setting and it is under-used. If your business is people buying a thing today, and there is no follow-up you would actually send, lead capture is a tax you are charging on your own checkout. Leave it off, let the assistant answer questions and put the right variant in the cart, and switch it on the day you have something worth emailing about.

Four moments worth an ask

Pick the moments deliberately instead of letting a form appear on a timer. Four earn the ask in most shops.

In each of these, say the exchange out loud. "So we can email you when it is back in your size" converts. "So we can send you a price for 200 units" converts. "To continue" does not, because it names no benefit and reads as a toll. The sentence in front of the field decides more than the setting behind it.

Notice what is not on the list. A visitor asking about shipping, a visitor asking whether you take returns, a visitor who has been reading a product page for ninety seconds. Those are questions with answers. Answer them.

Do not promise something the assistant cannot do

This is the constraint that gets broken most often, and it is worth being blunt about our own product. Starly does not send automated back-in-stock alerts. Nothing watches your inventory and fires an email when the count goes above zero. If the assistant says "we'll let you know when it's back", that promise is kept by a person on your team reading the lead and writing the email, or it is not kept at all.

So write the promise you can keep. "I'll pass this to our team and someone will email you when it's back" is true, and it is not the weaker version. A person replying is worth more than a template. The option you cannot survive is the third one, where you promise an alert that nobody sends. The visitor waits, the item comes back, they buy it somewhere else, and you have spent a warm lead teaching someone not to trust your chat.

The same test applies to every promise around the ask. If the sentence in front of the field implies a text message, a call within the hour, or a delivery update sent to them later, check that a person is going to do it. The wording is a contract, and the assistant will repeat whatever you wrote for as long as the card says it.

Where the record lands, and whose name is on it

A captured address is worth nothing until a named person acts on it. Starly emails each lead to a team member you choose, and can POST it to a webhook so it lands in whatever CRM, sheet or internal tool your pipeline already runs on. Neither route is cleverer than the other. What matters is that both end somewhere a person is already looking, because a list inside a dashboard is where leads die. Opening the dashboard is a task somebody has to remember, and on a Thursday nobody does.

Pick one owner and one deadline before you turn capture on. One name against the inbox, and a response time counted in hours rather than days. Two people sharing it is the same as nobody having it, and a rotation nobody wrote down is worse than either. A lead answered on day three is a lead you paid for and threw away: the person has bought elsewhere, or forgotten what they asked, or filed your email under things that arrived too late to be useful.

The lead record itself is thin by design. It carries the fields you asked for and the reason the assistant asked for them. Before you reply, open the conversation transcript in Starly and read it. The question, the product they were looking at, and the answer that did not satisfy them are all there, and a reply written from the transcript sounds like a person who was in the room rather than a template with a first name in it.

If nobody can commit to a same-day reply, leave capture off. Addresses you will not act on cost you the conversation and return nothing.

Say why, in one line

Nobody wants a privacy paragraph at the moment of typing an email address. They want two facts: what it is for, and when something will happen. "So we can email you the quote, usually within a day" does both in ten words. A block of text about how you handle data invites people to stop and read carefully, which is the opposite of what you want mid-ask. Keep the legal detail on your privacy page and link to it.

There is a related line that gets crossed without anybody deciding to cross it. An address given so you can send a quote is not an address given so you can send a newsletter, and the person who typed it knows exactly which of those they agreed to, even though your export will not. If you intend to market to these people later, ask for that separately, in its own sentence, and expect a decent share to decline. The cost of skipping the question is nothing at all until it is not: an unsubscribe is cheap, a complaint about a list somebody never joined is not, and either way you have taught a warm contact that answering your chat window has consequences they did not agree to. Keep the quote list and the mailing list as two lists, even when it is the same twelve people.

One specific reason beats a general reassurance. Compare "we respect your privacy and will never share your details" with "so our wholesale contact can send you the trade price list". The second names who, what and why. The first says only that somebody has been thinking about lawyers. Whichever you write, the assistant will say it in the customer's own language, so keep it plain enough to survive translation: one clause, no idiom, no joke.

When capture is the wrong move entirely

If someone on your team is available and the question has an answer, capture is the wrong move. You would be collecting a follow-up for a conversation that could simply finish. The same goes for low-ticket, high-volume shops. If the average order is $20 and repeat business comes from a newsletter people already signed up for, an in-chat form is friction on a sale you were going to make.

Lead capture also cannot fix bad answers. If the assistant does not know your shipping rules, or half the catalog never imported, every address you capture is a receipt for a failure, and the pile of them is a to-do list rather than a pipeline. Fix the answers first. The capture rate on a chat that is useful is a different number from the capture rate on one that is not, and only the first tells you anything about your visitors.

And do not use a required form as research. If you want to know who your visitors are, that deserves a method: analytics, a post-purchase survey, ten phone calls to customers you already have. Not a gate in front of your own product answers.

The two numbers to watch afterwards

Track the share of captured addresses that got a reply from a real person within one working day, and the share that bounced or went unanswered. Everything else — capture rate, leads per week, conversations opened — moves for reasons that have nothing to do with whether this is working.

Re-run the count monthly for the first three months. If both numbers hold and the volume is modest, lead capture is doing its job. If the volume is large and either number is bad, what you have is a form, and the fix sits upstream of every setting on this page: answer first, ask for less, and only ask when you have something to send.

Common questions

Should the chatbot ask for an email before it answers anything?

Only for high-value inquiries like quotes, custom orders or wholesale, where the next step costs a person's time either way. Asking before any answer converts a browsing visitor into a closed tab, and the addresses you do collect that way include a high share of fake and abandoned ones. For everything else, let the assistant answer first and then offer the exchange: fewer addresses, far more of them real.

How many fields should a chat lead form have?

Two is usually right: first name and email. Name and email get filled in; name, email, phone, company and budget do not, and the drop-off is not linear. Around the third field, people stop filling in the form and start deciding whether to bother. Collect the minimum that lets you follow up, then get the company, quantity or budget from the reply, because you will now be able to have that conversation.

Can Starly send automatic back-in-stock notifications to people who ask?

No. Starly captures the fields you configured and emails the lead to a team member you choose and/or POSTs it to your webhook. Nothing watches your inventory and fires an alert when stock returns. So the assistant should promise a person emailing them, not an automated alert, and someone on your team has to actually send it. Promising a notification the tool cannot send is worse than not asking for the address at all.

Where do the captured leads actually go?

Starly emails each lead to a team member you nominate, and can also POST it to a webhook so it lands in whatever CRM, sheet or internal tool your pipeline already lives in. Pick one owner and one response deadline before turning capture on. A list sitting in a dashboard nobody opens is where leads die, and a lead answered on day three has usually already bought somewhere else.

Is lead capture worth turning on for a small shop?

Not always, and off is a legitimate setting. If your customers buy the same day and there is no follow-up you would genuinely send, a form in the chat is friction on a sale you were going to make anyway. It is also the wrong choice if nobody on the team can commit to replying within a working day. Turn it on when you have a real reason to email someone back, such as a quote, a restock or a trade price, and a named person to do it.