Scoring
The score answers one question: how good is this lead for this business? Leadveyor splits the work in two. A language model answers your criteria about the message, with quotes. Code turns those answers into a number, a tier and a breakdown — the same way every time.
- Wedding? → 9+ months · “getting married next April”
- Guests? → 100–200 · “100–150 guests”
- Location? → near the kitchen · “in Collingwood”
- Funeral? → no
- Wedding, 10 months away+30
- 100–200 guests+15
- Near the kitchen+20
- Score65 → critical
In the example, the model answers four criteria: wedding — 9+ months away (quote "getting married next April"); guests — 100 to 200 (quote "100–150 guests"); location — near the kitchen (quote "in Collingwood"); funeral — no. Code adds 30 + 15 + 20 = 65 points, which falls in the critical tier (60 and above) on a scale where low starts at 0, normal at 20 and high at 40.
Why the model doesn't give the score
A model asked "rate this lead from 1 to 100" gives an opinion: it shifts with wording, can't explain itself precisely, and can be argued with by the message. Leadveyor never asks that. The model only answers concrete questions about the message — is it a wedding? how many guests? — and must quote the message for every "yes". The score is arithmetic on those answers. As a result:
- The same answers always give the same score. Changing your points re-weighs leads predictably.
- Every point is explained. The breakdown lists each criterion, its result, its points and the quote.
- The score can't be talked up. Instructions hidden in a message are ignored, and a "yes" without a real quote doesn't count.
Criteria
Your profile lists the criteria. There are four kinds:
| Kind | Decided by | Effect on the lead |
|---|---|---|
| Scored (AI) | The model picks the answer that fits | Adds that answer's points |
| Rule | Code checks a condition on the facts, e.g. guests ≥ 100, distance ≤ 30 km | Adds points — no model call, no cost, no misjudgement |
| Exceptional | A yes/no question or a condition | Sends the lead to the top tier (or a chosen tier) with score 100 |
| Disqualify | A yes/no question or a condition | Marks the lead disqualified with score 0 |
Rules are preferred whenever the answer is a plain comparison of an extracted value; AI criteria are for judgement that needs reading: "is the customer the property owner?", "is this a corporate event?".
Points and the score
Each scored criterion has a list of answers, each with points: wedding 9+ months away +30, 3–9 months +20, under 3 months +10. For every lead:
- Each criterion gives the points of the chosen answer — or 0 if no answer fits or the message doesn't say.
- The score is the sum, capped to 0–100.
- If no criterion at all could be answered, the lead is unscored rather than 0 — "we know nothing" is not "bad".
Points can be negative, to penalise: only wants a price list −10. Profiles can alternatively average grades on a 1–5 scale or count points as a share of the maximum; see How criteria combine.
Tiers
Tiers turn the number into an action: critical from 60, high from 40, normal from 20, low from 0. A lead takes the first tier from the top whose minimum it reaches. You name the tiers and set the boundaries (two to six tiers).
Two tiers are set by Leadveyor: disqualified and unscored.
Exceptional and disqualify
Some facts outweigh everything else. A funeral needs a call within the hour, whatever its size; a job application is never a customer. These criteria don't add points — they decide the tier:
- If a disqualify criterion fires, the lead is disqualified, even if an exceptional one fired too.
- Otherwise, if an exceptional criterion fires, the lead goes to the top tier (or the tier that criterion names).
- Among several of the same kind, the first in the profile wins.
The lead shows the reason of the deciding criterion — "Funeral: call back within the hour" — in the console and in webhooks. A disqualifier never fires on unknown information.
Unknown and confidence
A message that doesn't mention something leaves that criterion unknown: 0 points, never a penalty, never a disqualification. Each lead carries a confidence — the share of the profile (weighted by points) that could be answered. A score of 30 with confidence 90% is a weak lead; a score of 30 with confidence 30% is a lead that told you little — worth a call to find out more.
Dimensions
Criteria can be grouped into dimensions — Who, What, When, Where, Why or Budget, Authority, Need, Timeline. Each lead then gets a score and confidence per dimension: "What 80, When unknown" says the request is great but the date is missing. Dimensions never change the overall score.
Evidence and the breakdown
Every lead stores, for each criterion: the result (met / not met / unknown), the chosen answer, the points, and the quote. You see it as Why this score on the lead in the console; webhooks and the API deliver it with every lead, so a CRM can show it too.
Learning what matters
Recording outcomes — won, lost, no response — on leads closes the loop: comparing outcomes with the criteria shows which ones predict a sale. Raise the points of those that do, lower the rest, test on real messages, and publish a new version.