Criteria and points
Criteria say what makes a lead good for your business. For every lead, the AI answers each criterion's question about the message; Leadveyor then adds up the points of the answers to get the score. You find them on the Criteria tab of a scoring profile.
How the score is calculated
Each scored criterion has a list of possible answers, and each answer has points. For every message, the AI picks the answer that fits; the lead's score is the sum of the points of the chosen answers.
Example — a catering company with three criteria:
| Criterion | Answers and points |
|---|---|
| Wedding | 9+ months away +30 · 3–9 months +20 · under 3 months +10 |
| Guests | more than 200 +25 · 100–200 +15 · fewer than 100 +5 |
| Near the kitchen | in Collingwood or next to it +20 |
A message "Our wedding is in 10 months, about 120 guests, in Collingwood" scores 30 + 15 + 20 = 65.
- The message says nothing about a criterion? It gives 0 points. A message without a date simply doesn't get the wedding points.
- The score is capped at 100. If your best case adds up to more, every lead above 100 shows 100. The Criteria tab shows the best possible total next to the list.
- Nothing could be answered at all? The lead is unscored instead of getting 0, so you can tell "bad lead" from "we know nothing".
- The tier (for example critical / high / normal / low) comes from the score and the tiers you set.
Profiles can also combine criteria as an average of grades or as a share of the maximum points; see How criteria combine. Points, described here, are the recommended way.
Types of criteria
The Criteria tab groups criteria by type:
| Type | When it fires | Effect |
|---|---|---|
| Scored | The AI picks one of its answers | Adds that answer's points to the score |
| Exceptional | The answer to its question is "yes" | The lead goes straight to the top tier (or the tier you choose), with score 100, whatever the other criteria say |
| Disqualify | The answer to its question is "yes" | The lead becomes disqualified with score 0 |
| Rule | A condition on the extracted fields is true | Adds points without asking the AI (see Rules) |
Exceptional criteria are for leads you must never miss — a funeral, an existing key client, a date you're short of bookings. Disqualify criteria are for enquiries you never take — job seekers, areas you don't serve. If both fire, disqualify wins; if several of one kind fire, the first in the list wins. The lead shows the reason of the criterion that decided.
Adding a criterion with AI
Click Add criterion, choose the type and describe in your own words what to look for. Click Draft with AI: in a few seconds you get the full criterion — name, question, instructions and answers with points — in the edit form. Check it, change anything, and click Add to profile.
Write the description as you would explain it to a new employee:
- Scored: "Bigger events are better: more than 200 guests +25, 100–200 guests +15, fewer than 100 +5." If you give points, they're kept; if you don't, the best answer gets 20 and the rest less.
- Exceptional: "Funerals: the customer mentions a funeral, wake or memorial service. They need a call back right away."
- Disqualify: "Job seekers: the person is applying for a job or asking about work, not ordering anything."
Prefer Write it myself to fill in an empty form instead. Drafting costs a fraction of a cent and nothing is saved until you save or publish the profile.
Name
A short name shown on the criterion card, in the lead's score breakdown and in webhooks. "Wedding lead time", "Guest count", "Near the kitchen".
Dimension
Optional. The group the criterion belongs to in the lead's breakdown — for example When for a date criterion, Where for distance. Dimensions don't change the score; they show which side of an enquiry is strong or weak (good "What", unknown "When"). The list of groups is set in Settings → Dimensions.
Question and instructions
The question is what the AI asks about every message: "Is it a wedding, and how far away is it?", "How many guests are expected?". Ask one thing, about the message.
Instructions tell the AI how to answer: keywords and synonyms, how to read dates, what to do in edge cases. For example:
Look for wedding, reception, ceremony, bride, groom, engagement party counts as a wedding.
Count the time from today to the event date.
If the date is a range ("next spring"), use its earliest day.
If there is no date at all, answer unknown — don't guess.
Good instructions matter more than anything else for accuracy. A few rules of thumb:
- Say what to do when the message doesn't say: usually "answer unknown". Unknown gives 0 points, which is what you want for missing information.
- Never ask the AI to assume or invent facts the message doesn't contain.
- For every "yes" the AI must quote the message; a quote that isn't really in the message is thrown away. You see the quote as evidence in the lead's score breakdown and in the Test panel.
Answers and points
The possible answers of a scored criterion, each with its points. The AI reads the descriptions and picks the one that fits; the description is all it knows, so make each one clear and make sure they don't overlap:
| Points | Answer |
|---|---|
| +30 | Wedding 9 or more months from today |
| +20 | Wedding 3 to 9 months from today |
| +10 | Wedding less than 3 months from today |
- Add answer adds a line (up to 10). Order doesn't matter: answers are sorted by points when you save.
- An answer can give 0 or negative points — for example "−10: the customer only wants a price list" lowers the score.
- If no answer fits, the criterion gives 0 points.
- A yes/no criterion is just one answer: "+20: the event is in Collingwood or a neighbouring suburb".
Reason and tier
For Exceptional and Disqualify criteria:
- Reason — shown on the lead when the criterion fires: "Funeral — call back within the hour", "Job application". Your team sees it in the console and your CRM gets it in the webhook.
- Tier (Exceptional only) — where the lead goes. By default the highest tier; pick another to send, for example, VIP regulars to "high" rather than "critical".
ID
Under Advanced: the criterion's identifier, used in the API, in webhooks and in the score breakdown. It's created from the name; change it only if your integrations rely on a specific ID.
Rules
Rule criteria are decided without AI, by a condition on what was extracted: "guest count is 100 or more", "distance is under 30 km", "message arrived outside business hours". They cost nothing per message and never misjudge, so use them whenever the answer is a plain comparison of a field.
In the console a rule's condition is edited as JSON, for example:
{ "field": "fields.guest_count", "op": "gte", "value": 100 }
Rules can also have several answers, each with its own condition and points. The full list of fields and operators is in the scoring guide; you can also edit rules on the JSON tab.
Tips
- Start with 4–7 criteria. Every criterion is a question per message; a handful of strong ones beats twenty weak ones.
- Give the most points to what really predicts a sale. Once you record won and lost outcomes, compare them with the criteria to see which ones matter.
- Test before you publish. Pick a few real messages in the Test panel — a great lead, an average one, a poor one — and check that the scores and tiers come out the way you'd judge them yourself.