Lead Scoring — Your Chatbot Does Not Do It, and the Destination You Picked Decides Whether Anything Does

Quick answer: Lead scoring is a number attached to a prospect so that whoever is calling knows who to call first. It is not a chatbot feature. Across our fifteen platform reviews, exactly one exposes a named lead-scoring primitive in its own builder, and that primitive is a logic block rather than a model. Everywhere else the score is computed downstream, by a CRM or a marketing automation tool, over fields the bot handed it. That makes the buying question a plumbing question rather than a feature question, and our own corpus answers it uncomfortably: of the fourteen lead-capture scenarios our reviews record with a figure, six write to a Google Sheet and only five write to a system that can hold a score at all. The most common thing a chatbot does with a qualified lead, in the platforms we have looked at, is put it in a spreadsheet.

What it is, in one paragraph and one worked example

A scoring model is a list of rules that add or subtract points. Two kinds of rule do the work. Fit rules score what the prospect is: a fifty-person company in your target country is worth more than a student, so company size and job title carry points. Behavior rules score what the prospect did: requested pricing, +20; asked about a feature you do not have, −15; went quiet for thirty days, −10. Sum them, draw a line, and everyone above the line goes into a call queue.

Worked, because the arithmetic is the whole idea. A visitor tells your bot she runs operations at a forty-person logistics firm, in your country, and wants a quote this quarter. Role +15, size +15, country +10, quote request +25 gives 65. A student asking whether you have a free plan scores role +0, size +0, country +10, free-plan intent −20, which is −10. The bot did not decide either of those numbers. It collected six answers. Whatever holds those answers ran the addition.

That is the sentence that reorganizes the purchase. If the score is arithmetic over stored fields, then your chatbot's contribution to lead scoring is which fields it can store where, and the vendor's demo of a friendly qualifying conversation tells you almost nothing about it.

Two adjacent terms, defined here so the rest of the entry can use them: an MQL is a lead the marketing side thinks is worth a sales touch, and an SQL is one the sales side has agreed to work. The distinction is an internal treaty about who owns the follow-up rather than a property of the person, and none of our fifteen reviews uses either abbreviation even once.

What our fifteen reviews record, which is less than you would expect

Fifteen files match sample-reviews/*-review.md. Searching them for the vocabulary of this field produces a set of numbers small enough to print in full:

Read the first and last rows together. The vocabulary that dominates the sales-operations literature is almost absent from the platforms we review, while the name of one particular CRM appears in every single file. That is the finding in miniature: these products do not model your pipeline, they hand it to something that does.

The two lead-scoring matches are not equivalent. Our Landbot review documents a Lead scoring block in the builder's Logic category, visible in a captured screenshot of the Building Blocks sidebar alongside Conditions, Keyword jump, Global keywords, Formulas, A/B test and Goal. That is a real, named primitive, and Landbot is also the only platform in the corpus whose surface mentions BANT at all, as one of five example use-case chips in its assisted agent-generation modal. Our Wati review is the other match, and it is one bullet — "Lead scoring and enrichment, analytics dashboard" — with no mechanism, no tier and no screenshot behind it. One documented primitive and one list item is the honest count.

The three lead-qualification matches describe intent rather than machinery. Chatfuel documents its Fuely AI agent qualifying "against predetermined criteria," sourced to a vendor blog post. Tidio lists lead qualification among the tasks its Smart Actions can automate. Our Voiceflow review names lead qualification in its own audience-fit assessment for enterprise CX teams, which is our editorial judgment about the buyer rather than a vendor claim — a distinction worth keeping, since the other two rest on vendor artifacts. None of the three publishes a scoring model, a points scheme or a threshold.

The number to ask for: where does the field land

Every review is scored against the same lead-capture scenario, a five-field form wired to a destination, which makes the destinations comparable. Fourteen of the fifteen publish a figure. Here they are, grouped by destination rather than ranked by speed, because destination is the argument and the times are not a leaderboard:

Six of the fourteen destinations are a spreadsheet. A Google Sheet is an excellent capture target and it cannot score anything: it has no contact object, no activity timeline, and no place to hang a rule that fires when someone opens an email next Tuesday. Count the rows that write somewhere a scoring model could actually live — HubSpot twice, Salesforce twice, and the HubSpot leg of the Tars pipeline — and you get five of fourteen. Two near-misses are worth naming rather than filing. Botpress Tables is a real structured store with rows, columns and types, but it lives inside the agent workspace rather than in a system of record, so a score computed there is a score your sales team cannot see. The Tidio inbox does hold a contact record, and our review documents no scoring surface attached to it, so it is counted with the in-platform stores rather than with the CRMs.

What the table does not support, which is more interesting than what it does

The obvious hypothesis is that a native connector is fast and a middleware hop is slow, so pay for native. Hold the destination constant — the only fair comparison — and the corpus gives that hypothesis a weak yes, not the strong yes the pricing page implies.

Six platforms write to Google Sheets. Four go native (Chatfuel 7, Manychat 8, Typebot 8, BotPenguin 12; median 8) and two go through Zapier (AiSensy 8-12, midpoint 10; Wati 11; median 10.5). Native wins on median by two and a half minutes, wins six of the eight possible head-to-head pairings, and still loses two of them, because native BotPenguin at 12 is the slowest spreadsheet destination in the set. A first draft of this entry took that one pairing — Wati's Zapier hop at 11 beating BotPenguin's native connector at 12 — and concluded that the numbers refuted the native-is-faster story. They do not. They price it: on a spreadsheet destination the middleware hop is worth roughly two minutes of setup, which is not a purchase decision.

The larger effect is the destination itself. Every spreadsheet destination with a recorded figure sits between 7 and 12 minutes, a five-minute band across six platforms and two different integration architectures. Read at midpoints, and setting aside Tars for the reason below, the CRM destinations run 8 to 25 minutes — a seventeen-minute spread across four. A spreadsheet takes one OAuth and a column map. A CRM takes an OAuth, a decision about which object you are writing to, a property-by-property field map, and a rule for what happens when the contact already exists. That work does not disappear because the connector is native. The native connector moves it into a nicer panel and removes the per-task bill, which is worth money, and does not remove the mapping, which is where the time goes.

Two readings that cut against the paragraph above:

1. Tars's 22 minutes is not like-for-like. It is a three-tool pipeline — Airtable plus a Slack alert plus a HubSpot contact — not a single hop, so it sits in the table as a shape and is excluded from the CRM spread above rather than inflating it.
2. Ranges were read at their midpoints where the review published one (AiSensy 8-12, Landbot 6-10, Intercom 8-12). Reading Landbot at its 6-minute floor would put the fastest CRM destination below the fastest spreadsheet destination, which is the one reading that would overturn the paragraph. We do not take it, because taking the floor of one range and not the others is how a spread gets manufactured.

What to ask a vendor, in one sentence

"When my bot finishes qualifying someone, which object in my CRM gets written, which properties, and does that happen without a paid middleware task?" A vendor who answers with the name of the CRM has answered half of it. The half that decides whether you can score is the property list, because a scoring model can only read fields that exist.

Two of our reviews are worth reading before that call for opposite reasons. Intercom documents a bi-directional native HubSpot sync and a field map to HubSpot contact properties, which is the shape you want. Wati documents its own help center stating that HubSpot connectivity is "possible using Retool" — a third-party tool — and the review says plainly that third-party sites describing Wati as having native HubSpot integration are misreading the vendor's documentation. Same CRM, same claim in the marketing, materially different plumbing.

Related terms

• Lead generation chatbot — the bot that produces the record this entry scores.
• Slot filling — the field-by-field mechanic that collects the answers a score is computed over.
• Entity extraction — how a free-text reply becomes a typed value a rule can read.
• Lead nurturing — what you do with everyone who scored below the line.
• Chatbot conversion rate — the metric that tells you whether the line is drawn in the right place.
• Webhook — the escape hatch when no connector exists, native or otherwise.

FAQ

What is lead scoring?

Attaching a number to a prospect so a sales team can order its day. Points come from fit attributes (company size, role, country) and from behavior (requested a price, went quiet), and the total decides who gets contacted first. It is arithmetic over stored fields, which is why it runs wherever the fields are stored rather than wherever they were collected.

Can a chatbot score leads by itself?

Almost none of the ones we have reviewed can. One of our fifteen reviews documents a named lead-scoring primitive in the builder, a Logic block in Landbot, and even that is a rule element rather than a model. The realistic architecture is that the bot collects, the CRM scores, and your job at purchase time is to check that the handoff between them exists.

What is the difference between an MQL and an SQL?

An MQL is a lead marketing considers worth a sales touch; an SQL is one sales has accepted and is working. The line between them is an internal agreement about who owns follow-up, not a property of the person, which is why two companies with identical scoring models can classify the same lead differently. Neither abbreviation appears anywhere in our fifteen reviews.

Is a scoring model worth building for a small business?

Not until you have more leads than call capacity. Below that threshold the model is overhead: you are going to phone all of them anyway, and the honest version of scoring at that size is a sorted spreadsheet. Build the model when you start choosing which leads to ignore, and see our lead qualification playbook for the sequence.

Does Google Sheets work as a lead destination?

For capture, yes, and it is what six of the fourteen recorded scenarios in our corpus actually use. For scoring, no. A sheet has no contact object and no activity timeline, so behavior rules have nothing to fire against. If your plan involves scoring later, treat the sheet as a staging area with a migration ahead of it rather than as the destination.

What is BANT and do these platforms support it?

Budget, Authority, Need, Timeline — a four-question qualification checklist widely attributed to IBM's enterprise sales practice, in general circulation for decades, and surviving mostly because it is easy to remember. Supporting it requires nothing more than four fields, which is why no platform needs to "support" it. The word appears in exactly one of our fifteen reviews, as an example chip in Landbot's assisted agent-generation modal.

Should I pay extra for a native CRM connector?

Ask what it saves you in setup, then ask what it saves you per month, because the answers are different sizes. On setup, holding the destination constant, our corpus prices the middleware hop at roughly two and a half minutes: the four native spreadsheet destinations have a median of 8 minutes against 10.5 for the two that route through Zapier, and the slowest spreadsheet destination in the set is a native connector. The destination matters far more — spreadsheets cluster in a five-minute band while CRMs spread across seventeen. What a native connector reliably removes is the per-task bill and one failure surface, and those are worth paying for. What it does not remove is the field mapping, which is where the setup time actually goes.

Sources

• Chatbotscape review corpus, searched and read 30 August 2026. Denominator: ls sample-reviews/*-review.md | wc -l returns 15. All seven searches printed in the vocabulary table were executed verbatim from the repository root before publication and return 2, 3, 1, 0, 0, 0 and 15 respectively. The -E flag is printed on the two word-boundary rows for portability rather than necessity: GNU grep honors \b in basic mode too, and grep -rlie '\bbant\b' sample-reviews/*-review.md returns the same 1 on this machine. A draft of this note asserted that dropping -E returns a wrong 0; it was run afterwards and it does not, and the corrected claim is published here rather than the tidier false one.
• Lead-scoring primitives: sample-reviews/landbot-review.md (the Lead scoring entry in the Logic category of the Building Blocks sidebar, and the BANT example chip in the Build-it-with-AI modal, both quoted from the review's own screenshot captions); sample-reviews/wati-review.md (the single "Lead scoring and enrichment" feature bullet).
• Lead-qualification claims: sample-reviews/chatfuel-review.md (Fuely AI qualifying against predetermined criteria, sourced by that review to a vendor blog post); sample-reviews/tidio-review.md (lead qualification among Lyro Smart Actions); sample-reviews/voiceflow-review.md (lead qualification named in that review's own audience-fit assessment for enterprise CX teams — our editorial judgment about the buyer rather than a vendor-published use case, unlike the two rows above it).
• Scenario B destinations and setup times, each quoted from the review that recorded it: chatfuel 7 min / Google Sheets native / 98% field-mapping fidelity; manychat 8 min / Google Sheets / 100% fidelity; typebot 8 min / Google Sheets native / 100% fidelity across 50 submissions; chatbase 9 min / Calendly native; landbot 6-10 min / HubSpot native; intercom 8-12 min / HubSpot native bi-directional; aisensy 8-12 min / Google Sheets via Zapier; tidio 11 min / Flows editor, submissions captured into the Tidio inbox / 100% fidelity across 20 submissions; botpress 11 min / native Botpress Tables / 100% fidelity across 25 submissions; wati 11 min / Google Sheets via Zapier / 100% fidelity; botpenguin 12 min / Google Sheets native; voiceflow 14 min / Salesforce via API integration; tars 22 min / three-tool pipeline / 100% fidelity across 25 submissions; blip ~25 min / Salesforce native, marked PROJECTED in the review. sample-reviews/sendpulse-review.md lists Scenario B without publishing a time, and is the fifteenth file.
• Evidence heterogeneity. This entry originally carried a per-row evidence-class column and dropped it. The five self-contradictions quoted in the callout are, in order: manychat-review.md ("methodology-anchored simulation" against "Hands-on testing — 6-scenario protocol completed 24 May 2026 to 25 May 2026, 9 hours active"); wati-review.md ("pre-registered for this platform" against "6-scenario protocol completed … 9 hours active"); intercom-review.md ("the timed 6-scenario testing" against "roughly seven hours of structured analysis … combining vendor-page walkthroughs"); tars-review.md ("creating a HubSpot contact via Zapier middleware" in the integrations section and "HubSpot contact creation via Zapier" in the scenario summary, against "no Zapier-mediated chain" in a later screenshot reading — a draft of this bullet stitched the first two strings into one quotation that appears nowhere in the file, and grep -c returned 0 on it); blip-review.md ("concrete hands-on testing across Scenarios A-F" and "approximately nine hours of hands-on testing on 2 June 2026" against "Hands-on testing — queued, not completed — 6-scenario protocol projection only"). Unambiguous rows: tidio-review.md labels its figure "Measured (2026-05-28 hands-on)" against a live business trial account, and landbot-review.md describes its evaluation consistently as a structural walkthrough anchored against Manychat, with hands-on validation on a paid Pro account scheduled for 2026-06 and never claimed as done. intercom-review.md also schedules paid-account validation but is on the contradiction list above, and blip-review.md labels its Scenario B row PROJECTED while contradicting itself at file level, so neither is counted as unambiguous. All five contradictions are flagged for the manual editorial track and none was corrected in this run.
• Integration status for HubSpot, the CRM named in all fifteen files: native, vendor-documented and used as the Scenario B destination — with in-product validation still queued — in intercom-review.md and landbot-review.md; native per vendor documentation in blip-review.md, botpress-review.md, chatfuel-review.md, manychat-review.md, tidio-review.md and voiceflow-review.md; tier-gated to the King plan in botpenguin-review.md; explicitly not native in aisensy-review.md (absent from the integrations index, routed via Zapier or Pabbly), chatbase-review.md (via Zapier), typebot-review.md (HTTP request block) and wati-review.md (vendor help center: "possible using Retool"); and named in sendpulse-review.md only as a rival CRM rather than as an integration. That last distinction is why "named in fifteen" is not the same claim as "connects to fifteen."
• A further undisclosed corpus conflict on the anchor row, recorded here rather than resolved: manychat-review.md lists Google Sheets among its native integrations and its own Scenario B gives no path, while landbot-review.md describes the same Manychat anchor as "8 minutes with Google Sheets via webhook." The 8-minute figure is stable across both files; the path is not, which is why the Manychat row is the weakest "native" label in the table.
• Ahrefs Keywords Explorer, US overview and volume-by-country, queried 30 August 2026 — the demand, difficulty, CPC, global-volume, parent-topic and country-split figures in this entry's keyword note, including the checks behind declining 'mql vs sql', 'predictive lead scoring' and 'qualified lead'.
• Chatbotscape evaluation methodology, including the anchored-simulation protocol the setup times come from. /methodology (continuously updated).