Guide 02
Most websites lose the same way. Someone arrives with a real question, finds a contact form asking for everything at once, and leaves. This is about the conversation that goes in that gap: what to ask, in what order, when to stop asking, and where the answer should end up so somebody actually follows up.
A contact form is a demand made of a stranger. It asks for name, email, phone and a free-text description of your problem, all at once, before it has given anything back, from someone who has committed to nothing. Then it says "we will get back to you", which everyone reads as "sometime, maybe".
A conversation inverts every part of that. It asks one thing at a time. It answers questions along the way, so the visitor gets value before they give any. It can react to what was said, which a form cannot. And it can tell someone what happens next with a real timeframe rather than a shrug.
The gain is in the sequencing, not the novelty. A form that asked one field per screen and answered a question between each would perform similarly. The reason nobody builds that is that it would be a chatbot.
Worth being straight about the limits: this does not manufacture demand. A page with no traffic converts nothing regardless of what is on it, and a bot on a page where nobody has a question is an annoyance with a typing indicator. The gain shows up where people arrive with intent and the site was previously making them work for it.
We have put numbers to this elsewhere: what changes when a conversation replaces a contact form covers where the additional leads actually come from, which is mostly people who would have left rather than people who would have converted anyway.
Almost every small business lead conversation reduces to the same shape. The wording changes by industry; the sequence rarely should.
That is the whole thing. The instinct to add a fifth and sixth question is strong and should be resisted, because each one costs completions and the marginal answer rarely changes what you do. The test for any additional question is narrow: would a different answer change who calls them, how fast, or what they say? If not, the question is for your comfort, not the lead's benefit.
Lead scoring gets imported from enterprise sales into businesses that have no use for it. A twelve-point model assigning weights to seven attributes is a sophisticated way of producing a number nobody acts on differently.
For most businesses the qualification question is binary and immediate: is it worth someone's time to call this person back today? Answering that needs two facts, what they want and when, which the conversation above already collected.
Where sorting genuinely helps is routing rather than ranking. A lead that says "burst pipe, right now" and one that says "quote for next spring" should go to different places at different speeds. That is not a score, it is a rule, and rules are easier to write, easier to explain to whoever picks up, and easier to correct when they turn out wrong.
One caution about disqualifying. It is tempting to have the bot politely end conversations with people who look like a poor fit. Be careful: the signals available in four questions are thin, and a business that automatically waves off anyone whose budget sounds small will occasionally wave off somebody who was asking on behalf of somebody else. Route low-priority leads to a slower queue rather than to nothing.
For the mechanics of turning answers into a routing decision, qualifying a visitor without interviewing them goes through the rules themselves rather than the principle behind them.
Placement decides more of the outcome than the script does, and it is the part most people never revisit after launch.
Exit intent deserves a mention because it is genuinely effective and easy to overdo. Catching somebody as they leave is a legitimate last ask. Catching them with a modal that covers the page they were reading is why people install blockers.
A captured lead sitting in an inbox nobody watches is worth exactly nothing, and this is the most common failure in the whole category. The bot works, the leads arrive, and three weeks later somebody discovers forty conversations that were never actioned.
Three things have to be true before you turn any of this on.
There is an uncomfortable implication worth stating. If your business cannot follow up within a few hours, capturing more leads will make things worse rather than better: you will have a larger pile of people who now have a reason to be annoyed with you specifically. Fix the follow-up first, then turn up the volume.
Where a calendar is involved, the follow-up problem partly solves itself, which is the argument in letting people book inside the conversation: a booked slot needs no callback and cannot sit unactioned in an inbox.
Conversations started is the vanity number. It measures how aggressively the widget opens, which you can raise at will and which tells you nothing. Four that mean something:
Read a sample of transcripts weekly alongside those. The numbers tell you where people leave; the transcripts tell you why, and the why is often embarrassingly simple, like a question phrased in a way that reads as nosy.
All of this assumes the thing exists and has been built with some care. If it has not been yet, the guide to building a website chatbot covers the sequence, from deciding the one job it does through to what breaks in month two.
On nearly every site the most common opening message is some version of "how much does this cost". How a chatbot handles that one exchange decides more leads than the rest of the script combined, and most businesses handle it badly in one of two directions.
The first mistake is deflection. "Pricing depends on your needs, can I get your email?" is heard, correctly, as a refusal. The visitor asked a direct question, got a wall, and now has to pay in contact details to get past it. A good number of them will simply go and ask a competitor who answers.
The second mistake is over-answering. Reciting a full rate card to somebody whose situation you know nothing about invites them to price-shop a number that may not even apply to them, and it ends the conversation at the exact moment it was about to become useful.
The version that works gives a real range, then earns the right to narrow it. "Most jobs like this land between six and nine hundred. It depends mainly on access and how much has to come out. What are you looking at?" That answer does three things at once: it treats the question as legitimate, it filters out anyone for whom the range is impossible, and it hands back a question the visitor now wants to answer, because a better number is in it for them.
This is also the cleanest reason to give a bot access to your actual pricing information rather than a script of approved phrases. A range you would say out loud on the phone is a range the bot should be able to say. If there is genuinely no number you are willing to state, that is worth examining separately, because your competitors are stating theirs.
One related habit: when the honest answer is "it depends", say what it depends on. "It depends" alone is noise. "It depends on square footage and whether the existing floor comes up" is information, and it usually prompts the visitor to volunteer both.
There is more to say about this one exchange than fits here, including what to do when you genuinely cannot publish a number: answering the price question without losing the lead takes it further.
The four questions look different by industry mostly in the third slot. Three versions, to make the pattern concrete.
In all three the skeleton is identical and only the third question is industry-specific. If you find yourself needing two questions in that slot, pick the one whose answer would change the callback, and get the other from the person doing the calling.
Each of those three has been worked through properly on its own: real estate, ecommerce, and software and subscription businesses, where the third question and the escalation rules differ most.
Every lead capture conversation has a moment where automation should stop. Recognizing it is a design decision, not a technical one, and getting it wrong is expensive in both directions.
Hand off too early and you have built an expensive contact form: the visitor types two sentences and gets told somebody will be in touch, which is what the form already did. Hand off too late and you have a person repeating themselves to a machine that keeps missing the point, which is worse than having no chat at all because it is actively insulting.
Three reliable triggers for stopping. The visitor asks twice for a human, which is a request that should never be argued with. The conversation covers something with real consequences attached, like a medical symptom, a legal deadline, or a complaint about work already done. Or the bot has answered the same underlying question twice and the visitor is still asking, which means it does not have the answer and repetition will not produce it.
When the handoff happens outside business hours, say so plainly and give a time. "Nobody is here until 8am, but this is in front of them first thing and I have your number" keeps the lead. Silence loses it, and a cheerful promise that somebody will respond shortly, at eleven at night, loses it twice, because you have now also been caught out.
Three or four before contact details, and only ones whose answers change what you do next. If you cannot say what you would do differently based on an answer, do not ask it.
Whichever you will use within the hour. A phone number you never call is worse than an email you answer, because they gave you the more personal detail and got less for it.
Usually, because of sequencing rather than technology. A form asks everything at once and gives nothing back until submit. A conversation asks one thing at a time and answers questions along the way.
After some sign of intent, not on arrival. Fifteen to thirty seconds, scroll depth, or exit intent. Be more forward on a pricing page than on a blog post.
That you can tell whether it is worth calling them back today. For most businesses two facts do it: what they want and when. Everything else is a question asked to feel thorough.
Four questions
The industry pages below are the same four questions with the third one already chosen for you.