Two leads fill out your chat on the same afternoon. One asks about your enterprise plan, mentions a team of sixty, and wants to start next quarter. The other asks if you have a free trial and never comes back. If both land in the same inbox with the same priority, your best salesperson spends the day guessing. Lead scoring is how you stop guessing.
The idea is simple: turn what a lead does and says into a number, so the strong ones rise to the top. Conversational AI makes this better than the old form-based version, because a chat reveals things a form never asks.
What lead scoring is really for
Scoring isn't about labeling people. It's about deciding where your limited human attention goes first. A small team can't call every lead within five minutes, so the score answers a practical question: who gets the fast, personal follow-up, and who gets a slower automated nurture?
Get that routing right and your closers spend their hours on deals that can actually close. Get it wrong and they burn the day on tire-kickers while a hot lead cools in the queue.
What a conversation reveals that a form doesn't
A form gives you what someone was willing to type into blank boxes. A conversation gives you intent. When a lead asks "can I migrate my existing data?" they're telling you they already have a system and are seriously considering a switch. That's worth more than any checkbox.
Chatbots capture these moments naturally. The questions a person asks, the plan they poke at, whether they push to talk to a human, how specific their needs are. All of it is signal, and most of it never fits on a form. A conversational tool lets you score on behavior in the moment, not just declared facts.
Signals worth scoring
Not every signal is equal. Some strongly predict a good customer; some are noise. Here's a starting set for a small business, with rough weights you'd tune over time:
| Signal | Why it matters | Sample weight |
|---|---|---|
| Stated budget in your range | Direct sign they can buy | +30 |
| Team or company size fits | Predicts fit and deal size | +20 |
| Near-term timeline | Ready now, not "someday" | +25 |
| Asked to talk to sales | Explicit buying intent | +20 |
| Specific product question | Real evaluation, not browsing | +10 |
| Only asked about free plan | Often low intent to pay | -10 |
| No contact info given | Can't follow up | -15 |
Weights are a starting point, not gospel. Yours will look different, and that's fine.
Building a simple score
Resist the temptation to build something elaborate. A model you can explain in one sentence beats a black box you can't trust. Here's a workable path:
- Pick three or four signals that best predict a real customer for you. Budget, timeline, and fit are safe first picks.
- Assign each a point value, positive or negative. Keep the numbers round and easy to reason about.
- Set a threshold. Above it, the lead is hot and routes to a person now. Below it, it goes to nurture.
- Have the chatbot ask for those signals naturally during the chat, then tally the score as answers come in.
- Route on the result. Hot leads hit your inbox or CRM immediately; the rest get an email capture and a follow-up sequence.
In SpideyChat you'd build the qualifying questions into the chat flow and pass the scored result, plus the transcript, to wherever your team works. The point is not a perfect algorithm. It's a consistent, honest first pass so nobody gets ignored and nobody gets over-served.
A worked example
Take Bloom & Ledger, a fictional bookkeeping service for small firms. They set a hot threshold of 50 points.
A visitor tells the bot they run a 15-person agency, want to switch from a spreadsheet within the month, and asks about monthly pricing. That's fit (+20), timeline (+25), and a real product question (+10), for 55 points. The bot flags them hot and offers a call that afternoon. A founder takes it, and the deal moves.
Another visitor only asks whether there's a free version and gives no email. That's -10 and -15, a clearly cold score. The bot still answers helpfully and invites them to leave an email for a guide, but nobody drops what they're doing to chase it. The attention went where it paid off.
A score is only useful if it changes what you do
A number nobody acts on is just decoration. The point of scoring is to trigger different treatment, so decide up front what each band means in practice.
A workable three-tier setup:
- Hot: route to a person immediately, ideally with a live handoff or a same-day call offer. These are the leads worth interrupting someone's afternoon for.
- Warm: capture the email, drop them into a short follow-up sequence, and let a rep reach out within a day or two.
- Cold: answer helpfully, invite them to subscribe, and don't spend human time chasing. Some will warm up later on their own.
Write these actions down next to the thresholds, and make sure whatever you use for chat actually does them automatically. A hot lead that sits unrouted for six hours may as well have scored cold; the speed of the response is part of the value. The score isn't the deliverable. The faster, smarter follow-up it triggers is.
Keep it honest and adjustable
A scoring model is a hypothesis, not a fact. The only way to know if yours works is to check it against reality.
Once a month, pull your closed deals and your dead leads and look at the scores they had. If deals keep closing from leads you scored low, your weights are wrong, and that's good news. Adjust. Maybe timeline matters more than you thought, or budget less. A model that never changes is a model nobody's checking.
A few guardrails keep this useful rather than misleading:
- Don't over-fit to a handful of deals. Wait for a real sample before big changes.
- Keep the human override. A rep who senses a great fit should be able to promote a lead regardless of score.
- Watch for gaming. If leads learn to say the magic words, your signals lose meaning. Weight behavior, not just claims.
Scoring done this way turns your chat from a lead firehose into a ranked queue your team can actually act on. Start with three signals, ship it, and let a month of real outcomes tell you what to fix next.