Lead capture: every extra question costs you people and buys you quality

Every question a lead bot asks does two things at once. It tells you more about the person, and it loses some of them. That trade is the whole design problem, and most implementations resolve it badly by treating more information as straightforwardly better.

It is not straightforwardly better. A bot asking eight questions produces well-described leads and fewer of them, and whether that is an improvement depends on numbers specific to your business: what a lead is worth, what it costs you to talk to an unqualified one, and how many you can handle.

What we'll cover

• The trade, stated properly
• Where your optimum sits
• The three questions worth asking
• Order matters more than count
• Building it with CodeWords
• Speed beats qualification
• The handover, where most leads are lost
• Scoring that somebody actually uses
• What never to ask a bot to decide
• Making it survive
• Limits worth knowing about
• What to measure first
• Frequently asked questions

The trade, stated properly

Two numbers move in opposite directions as you add questions.

Completion rate falls. Each question is a moment somebody can stop, and the drop is steepest on anything that feels intrusive or that they cannot answer quickly.

Lead quality rises. You know more, so your team wastes less time and can prioritise.

The net effect depends on what an unqualified lead costs you. If a salesperson spends forty minutes discovering that somebody was never a fit, questions are cheap and worth asking. If qualifying takes two minutes on a call your team was making anyway, the questions are costing you leads for very little.

Most businesses have never worked this out and default to asking more, because a form with more fields feels more professional. It is the more expensive option in a lot of cases.

Where your optimum sits

Three numbers settle it, and you can estimate all three in an afternoon.

What a closed deal is worth, and roughly what proportion of enquiries close.

What it costs to handle an unqualified enquiry, in the time of whoever handles it.

How many enquiries you can actually handle. This one is decisive and gets ignored. A team at capacity should be filtering hard. A team with spare capacity should be capturing everything and sorting later, because a lead lost at question four is worth nothing at all.

The shape that falls out is usually: capacity-constrained businesses with expensive sales time should ask more; businesses with spare capacity and cheap qualification should ask almost nothing beyond how to reach somebody.

The three questions worth asking

Where you are asking anything, these earn their place ahead of the rest.

What they actually want. In their words, not from a list. A single open question produces more useful information than four multiple choices, and it is the one people answer willingly because it is what they came to say.

When. Timing separates a live enquiry from research more reliably than budget questions do, and people answer it honestly.

How to reach them, and what they prefer.

Budget is the question everyone wants to ask and the one that loses most people. Where the work has a genuine range, saying so and asking whether that is the right territory works better than asking what they have to spend.

Order matters more than count

Ask the cheap question first and the expensive one last, where cost means how likely it is to end the conversation.

What they want is nearly free, because it is what they arrived to tell you. Timing is cheap. Contact details are moderately expensive and should come after somebody has invested a little. Budget is expensive. Anything about their organisation's internal structure is expensive and rarely useful early.

The common failure is asking for a name and email first, before the visitor knows whether you can help. That reads as being captured rather than helped, and a share of people leave at exactly the point where they had the most intent.

Get the shape of what they need first. Then the contact details are a reasonable thing to ask for, because there is now something to come back to them about.

Building it with CodeWords

CodeWords connects to more than 3,000 integrations, and the connection is made once and reused.

1. Open CodeWords and start a new automation.
2. Describe what should happen in plain language to Cody, the automation builder: what you need to know, in what order, and what a qualified lead triggers.
3. Authorize the connections to your CRM and whichever channel the enquiries arrive on.
4. Describe the exceptions: somebody who will not answer a question, somebody who asks something unexpected, an existing customer, an enquiry outside what you do.
5. Record every conversation including the abandoned ones, since where people stop is the only data that tells you whether you are asking too much.

You describe the outcome; Cody builds it, connects it, and deploys it. The free plan covers light use, with Pro at $39 per month and Business at $100 per month as usage grows; details are on the pricing page.

Speed beats qualification

If you improve one thing, make it how fast a person responds rather than how well the bot qualifies.

Somebody enquiring is usually enquiring in several places, and the order of replies matters. A lead answered while they are still thinking about the problem is in a different conversation from one answered the next morning, when they have already spoken to somebody else.

Which means the bot's most valuable job is often not qualification at all. It is acknowledging immediately, capturing enough to be useful, and getting a person in front of the enquiry quickly. A lightly qualified lead reaching a human in ten minutes beats a thoroughly qualified one reaching them tomorrow.

Measure your time to first human response. Most businesses do not know it, and most are slower than they believe.

The handover, where most leads are lost

The bot does its job and then the lead evaporates, which is a process problem rather than a bot problem.

Send the whole conversation, not a summary of fields. What somebody actually wrote carries information no form captures.

Name somebody. A lead arriving in a shared inbox belongs to nobody. Assign it, with a time by which it must be picked up.

Tell the customer what happens next, specifically. "Someone will be in touch" is vague enough to feel like nothing. "Sarah will call you this afternoon" sets an expectation that also makes your team accountable.

Make the handover visible so an unactioned lead is obvious. A lead sitting untouched for two days should be somebody's problem.

Scoring that somebody actually uses

Most lead scores are ignored, because they are computed from things that never predicted anything.

Two rules. Base it on what actually correlated with closing in your own history, rather than on a template. And keep it to a handful of factors, because nobody trusts a number they cannot explain.

Check it against reality quarterly. If your highest-scored leads are not converting better than the rest, the score is decorative and worth deleting rather than tuning. A score nobody trusts is worse than none, since it gets used as a reason to ignore leads.

What never to ask a bot to decide

Whether to reject an enquiry. Filtering into a lower priority is fine. Telling somebody you cannot help them is a person's call, and bots get it wrong in ways that end up on social media.

Anything about price beyond published figures. A number a bot mentions becomes a number the customer expects.

Commitments about timing or availability it cannot verify.

Anything where somebody has expressed frustration. That goes to a person, immediately.

Making it survive

Log abandoned conversations, since where people stop is the whole dataset.

Measure time to first human reply, and treat it as the primary number.

Pass the full transcript to whoever picks the lead up.

Check the score against outcomes quarterly, and delete it if it does not predict.

Re-read twenty real conversations monthly, particularly the ones that ended early.

Limits worth knowing about

More questions is not more leads, and the trade is specific to your economics.

A bot cannot judge fit, only collect what it was told to collect.

Qualification cannot fix slow follow-up, which is usually the larger loss.

Scores decay as your market and mix change.

Some enquiries are unqualifiable in advance, and for those the right design is capture and route quickly.

What to measure first

Where people stop. Instrument the abandoned conversations before changing anything, because the question that loses you most people is rarely the one anybody predicted, and without that data every change to the flow is a guess.

Two habits make the difference. Ask what they want before asking who they are, since contact details requested before any value is offered read as capture rather than help. And optimise time to first human reply ahead of qualification quality, because the lead you answer first is frequently the one you win regardless of how well you described it.

Frequently asked questions

How many questions should a lead bot ask?

It depends on what an unqualified enquiry costs you and whether you have spare capacity. Capacity-constrained businesses with expensive sales time should ask more; businesses with room should capture nearly everything and sort afterwards.

What are the most useful qualifying questions?

What they actually want, in their own words; when they need it; and how to reach them. Timing separates live enquiries from research more reliably than budget questions.

Why is asking for budget a problem?

It loses more people than any other question and is frequently answered inaccurately. Stating your own range and asking whether that is the right territory works better.

What order should questions come in?

Cheapest first, where cost means how likely a question is to end the conversation. What they want, then timing, then contact details. Asking for a name and email first reads as being captured.

What matters more than qualification?

Time to first human reply. A lightly qualified lead reaching a person in ten minutes generally beats a thoroughly qualified one reaching them tomorrow, because enquirers are talking to several suppliers.

Where do captured leads get lost?

At the handover. Send the full transcript rather than fields, assign a named person with a time to pick it up, tell the customer specifically what happens next, and make unactioned leads visible.

Is lead scoring worth doing?

Only if it is based on what actually correlated with closing in your own history, kept to a few explicable factors, and checked against outcomes quarterly. A score nobody trusts becomes a reason to ignore leads.

What should a bot never decide?

Whether to reject an enquiry, anything about price beyond published figures, commitments it cannot verify, and anything where somebody has expressed frustration.

What should I instrument?

Abandoned conversations, by the question at which they stopped. It is the only data that tells you whether you are asking too much, and the costly question is rarely the predicted one.

Should the bot or a person qualify?

Use the bot for what is cheap to ask and unambiguous, and a person for judgement. The bot's most valuable contribution is often speed rather than depth.

Related reading

• Asking without interrogating
• Connecting conversations to your CRM
• Designing the conversation