What is Leadveyor
Leadveyor turns incoming messages into scored leads. It receives enquiries from every channel a business uses — email, website forms, form builders, its own apps — sets aside what isn't a customer, works out who the customer is and what they want, recognises people who write again, and scores every lead by criteria the business defines, with the reason and a quote from the message for every point.
What it does
A business that gets enquiries from several places faces the same problems every day: real enquiries are buried among ads, newsletters and notifications; the details are spread over free text in different formats; the same person writes three times; and nobody can say quickly which enquiry deserves a call first. Leadveyor does that first pass automatically and the same way every time:
- Filters junk — ads, sales pitches, newsletters, receipts, auto-replies and spam never reach your team as leads, but stay visible and recoverable.
- Extracts the lead — name, email, phone, company, what they want, where, and the fields that matter to you (event date, guests, budget…), as typed values a computer can compare.
- Merges repeat messages — a reply or a second form from the same person adds to their lead instead of creating a duplicate.
- Scores by your criteria — each criterion gives points; the score and tier say how good the lead is, and the breakdown says why.
- Delivers the result — in the console, through signed webhooks to your CRM, through the API, and to AI assistants such as Claude or Cursor through the MCP server.
Leadveyor doesn't replace your CRM or your follow-up: it decides what is a lead and how good it is, and hands that to the tools where your team works.
The path of a message
- In / out1. SourcesEmail, web forms, your apps through the API
- Code2. IntakeStored, de-duplicated, queued; never lost
- AI3. Junk filterDomain rules, then a fast model: lead, junk or not a lead
- AI4. ExtractionWho, what, where, your fields; answers to your questions with quotes
- Code5. EnrichmentDistance and drive time; repeat messages merged; contact linked
- Code6. ScoringPoints from your criteria → score, tier, reasons
- In / out7. DeliveryConsole, signed webhooks to your CRM, API
Every message takes the same seven steps:
- Sources — a message arrives through an email address, a form link or the API. Each source belongs to a workspace and says which scoring profile to use.
- Intake — the message is stored first and queued for processing. The same message arriving twice is recognised and processed once.
- Junk filter — your domain rules, then a fast language model decide whether the message is a customer enquiry. Only a confident "not a lead" stops a message.
- Extraction — a language model reads the message once and returns the lead(s) in it: contact, request, location, your fields as typed values, and its answers to your criteria — each "yes" with a quote from the message.
- Enrichment and duplicates — distance and drive time from your business are added; if the same person wrote recently, the message is merged into their lead; the lead is linked to the person's contact.
- Scoring — code adds up the points of the answers into a score from 0 to 100, picks the tier, and records the reason for every point.
- Delivery — the lead appears in the console, and webhooks deliver it to your systems, with retries.
Principles
Models read, code decides. Language models do what they are good at — understanding messy human text: telling an enquiry from an ad, finding the date in "the Saturday after Easter", answering "is this a wedding?". The score itself is never asked from a model: it is computed by code from the model's answers and your points. The same answers always give the same score, and you can see exactly how it was reached.
Every point has a reason. Each criterion records its result (met, not met, unknown), the points it gave, and for every "met" an exact quote from the message. A quote that can't be found in the message is rejected and the answer becomes unknown — which also protects you from messages written to talk a model into a higher score.
Unknown is not "no". When a message doesn't say something, the answer is unknown: it earns no points and lowers the lead's confidence, but never counts against the customer and never fires a disqualifier. A short form and a poor lead are different things, and the confidence shows which is which.
Doubt counts as a lead. Losing a customer costs far more than reading one extra ad, so the junk filter only sets a message aside when it is confident; everything set aside stays visible and can be sent back with one click.
You define what "good" means. Leadveyor has no built-in idea of a good lead. Your scoring profile — criteria in plain words, points, tiers — is the only source of the score, and works for any business in any language.
Changes never rewrite history. Profiles are published as frozen versions. Every lead remembers the version that scored it, so changing criteria today doesn't silently change yesterday's scores, and a bad change is undone with one click.
Nothing is lost. A message is stored before it is processed. Failures are retried; a message that still fails is kept, marked failed and can be run again. Webhook deliveries are signed, logged and retried too.
One set of rules for people and programs. The console, the API, the MCP server and integrations use the same operations and the same permissions. Anything you can do in the console, a program can do through the API, and an AI assistant through MCP.
What you configure
| What | Where | Purpose |
|---|---|---|
| Business description, address, time zone, currency, hours | Workspace settings | Context for reading messages; origin for distance; local dates and times |
| Sources | Sources | Where messages come from, which profile scores them |
| Scoring profile | Scoring profiles | Criteria and points, tiers, fields to extract, junk rules |
| Webhooks | Webhooks | Where results are sent |
| API keys | API keys | Access for your own programs and AI assistants (MCP) |
Where to go next
Read on through this chapter for each step in detail — sources, junk filtering, extraction, contacts and duplicates, scoring, profiles and versions, reliability and security. Then the console guide shows how to set everything up, the API guides show how to connect your own systems, and the MCP guide shows how to work with Leadveyor from an AI assistant.