Sales and support usually share one chat console, and the console was mostly built by people thinking about support. It will tell an agent that a visitor sounds frustrated, or that the agent's own tone has gone cold. Neither of those answers the question a salesperson actually has, which is whether this conversation is close to a purchase. A cheerful chat about a feature can be miles from a sale, and an irritated visitor can be a week from signing.
The buying-intent meter exists to close that gap. It is a small gauge on the right-hand rail of the agent console that reads the visitor's words and settles on a label: Hot, Warm, Lukewarm, Cool, or Disqualified.
What it is, and what it is not
Start with the disclaimers, because they are what make the meter trustworthy. It is a rule-based scorer, not a model. It reads a curated list of phrases, applies fixed arithmetic, and shows the result. It sends nothing to any CRM, writes no records, and never replies to anyone on its own. Nothing about the label reaches the visitor, and nothing about it is used to route or prioritise the chat.
It also only ever reads the visitor's messages. What the agent writes does not move the number, which matters more than it sounds: a scorer that counted the agent's enthusiastic pitch would be measuring the agent, not the lead.
Seven categories, and why each counts once
Visitor messages are matched against seven groups of buying language. Pricing, when someone asks what it costs or about discounts. Purchase, when they talk about buying, ordering or paying. Next-step, when they ask how to get started or proceed. Trial or demo, covering free trials, sample accounts and booking a walkthrough. Stakeholder, the business-to-business tell, where a visitor mentions their team, their manager, procurement, or needing sign-off. Timeline, where a date or a deadline appears. And compare, when they mention alternatives or ask what makes you different.
A category can score at most once per message. A visitor who writes what does this cost, and how much for five seats registers one pricing hit, not two, so a wordy question cannot outweigh a decisive one. The phrase lists are also deliberately narrow. A false Hot label changes how an agent behaves, so the design prefers to miss a borderline signal rather than manufacture one.
How the score actually moves
Hits accumulate over the whole conversation, and each category is capped at three. That cap is the most important rule in the system. A visitor who mentions price seven times has told you one thing about themselves, not seven, and without the cap a single anxious pricing conversation would out-score a genuinely broad buying discussion.
The label then follows the total. Six or more reads as Hot, three to five as Warm, one or two as Lukewarm, and zero as Cool. In practice that means breadth wins over repetition: three or four different kinds of buying signal in one chat is the shape of a real opportunity, and the labels plateau at Hot rather than chasing a maximum.
Pullback counts double
A separate list catches explicit withdrawal. Just looking, not right now, maybe later, I will think about it, not for us. Each distinct phrase of that kind subtracts twice what a buying signal adds, and enough of them push the score negative, which is the Disqualified label.
Two things about that list are worth knowing. It is about pullback, not objections: someone saying your price is too high has not withdrawn, they have raised an objection, and that belongs to a different panel. And it is asymmetric on purpose. A visitor telling you plainly that they are not buying today is a stronger, clearer statement than any single positive phrase, so it is weighted accordingly.
The one-line hint under the gauge
Under the gauge sits at most one short line of advice, and it is conditional rather than constant. A Hot score prompts you to propose the next step, but only when the conversation has no open commitment and no decision recorded, because if you have already promised to send a quote the advice would be noise. A Warm score suggests keeping the momentum, and only when the visitor's most recent message contained no pullback.
If the latest message pulled back, the line changes to suggest exploring why. And if the visitor just asked about price and you have not yet replied, it says so, because an unanswered pricing question is the single most reliable way to lose an otherwise willing buyer. When none of those conditions hold, the line stays empty. It does not fill silence with encouragement.
It resets on transfer, deliberately
When a chat is transferred or accepted by another agent, the score starts again from zero for the receiving agent. This looks like data loss and is actually the correct behaviour: the score is a reading of one agent's conversation, and a Hot badge inherited from someone else's judgement invites the new agent to skip the listening that produced it. The signals are still in the transcript, and if they are real they will re-appear within a few messages.
The score also feeds the wrap-up panel. A one-line buying-signals summary with the per-category counts is included when you copy or save a chat recap, and a Hot visitor closing a chat with no commitment and no decision recorded raises a gentle amber flag before you finish, on the theory that a hot conversation with no next step is worth ten seconds of second thought. A disqualified visitor never raises it, since an explicit no is a perfectly good resolution.
What the label is safe to be used for
Use it as a prompt during the conversation. It is good at reminding a support-shaped agent that this particular chat has turned into a sales conversation, and good at flagging the visitor who quietly signalled a deadline while you were troubleshooting.
Do not use it as a lead score in a report. It is in-memory only, it never persists, it resets on transfer, and it is derived from phrase matching against English text. Every one of those makes it unfit for a funnel metric. If you want lead quality measured, measure the outcomes you record deliberately, and let the meter do the job it is good at, which is nudging the person in the conversation while the conversation is still happening.
What to measure
Measure the handoff, not the gauge. Count how many chats you record as sales opportunities, how many reach a named next step, and how many of those convert. If your agents say the meter is usually right, the number worth watching is whether Hot conversations end with a commitment recorded, since that gap is exactly what the wrap-up flag is trying to close.
It is also worth reading a handful of Disqualified transcripts each month. If genuine buyers are landing there, someone is being talked out of a purchase before they were finished deciding, and that is a coaching finding no automated score will hand you.
Put it into practice
- Explain the arithmetic to the team. A transparent rule gets trusted; a mysterious badge gets ignored.
- Say out loud that it never reaches the visitor and never triggers an automated reply.
- Treat the pricing prompt as urgent. An unanswered price question is the costliest silence in chat.
- Expect the reset on transfer and brief the receiving agent with words, not badges.
- Never report on it. It is in-memory, per-agent and resets; it is not a metric.
- Close Hot chats with a next step, which is precisely what the wrap-up flag is asking for.
- Review Disqualified transcripts monthly to catch buyers being lost early.
A meter like this is useful for the same reason a good colleague sitting nearby is useful: it notices the thing you were too busy to notice, and then leaves the decision to you. Keep it in that role and it will earn its space on the rail. Promote it to a metric and it will start describing your phrase lists rather than your pipeline. The qualifying questions that actually decide a lead are covered in chat for lead qualification, and they remain the agent's job.