Lead scoring and deduplication, explained
A long list is not a useful list. LeadMind drops rows that cannot be contacted, removes duplicates within the job and against the last 30 days of your organisation’s leads, applies your filters, and scores what is left from 0 to 100 with a reason you can read.
Product · Updated
The fit score
A language model reads each lead’s name, category, website and location against your original request and returns a score from 0 to 100 with a one-line reason, such as “Dental clinic in Jumeirah, matches category and district; no booking widget found”. The list arrives sorted by score, so the best matches are at the top of the CSV.
- It measures fit: how closely the lead matches what you described.
- It does not measure intent: nobody has told LeadMind they are shopping.
- It does not measure deliverability: that is what email verification is for.
If the model cannot score a lead it gets a neutral 50 with the reason “Not scored by model”, never a made-up judgement.
Deduplication
| Lead type | Duplicate when these match |
|---|---|
| Business | Website domain; otherwise email; otherwise name and city |
| Person | Email; otherwise LinkedIn profile URL; otherwise name and company |
Duplicates are removed inside each job and against everything your organisation received in the previous 30 days, so running “gyms in Dubai Marina” twice in a month does not bill you twice for the same gyms. Dedupe keys are stored as hashes and contain no contact data.
Before scoring: the quality gates
- Reachable: a row needs a name and at least one way to reach it (email, phone, website, or a LinkedIn profile for a person).
- Suppressed: people and businesses who asked not to appear in lists through the data request page are removed.
- Filtered: signals in your prompt, such as “with WhatsApp” or “no online booking”, are checked on each website, and leads that fail are dropped.
All three happen before billing: you pay one credit per lead that is actually delivered. See pricing.
Using the score well
- Work the list top-down and sample the bottom tenth before discarding it; a low score with a sensible reason is often just a category mismatch.
- Put the signals that matter into the prompt as filters rather than hoping the score reflects them.
- Re-run the same prompt monthly: the 30-day dedupe means you only pay for businesses you have not seen.
Frequently asked questions
- What does the lead score measure?
- Fit: how well a lead matches what you asked for, judged by a language model from the business name, category, website and location, from 0 to 100 with a one-line reason. It is not a buying-intent score and not an email deliverability score.
- How does LeadMind deduplicate leads?
- Each lead gets a key: the website domain for a business (or its email, or its name and city when there is no site), and the email, LinkedIn URL or name and company for a person. Repeats inside a job are dropped, and so is anything your organisation already received in the last 30 days.
- Am I charged for duplicates?
- No. Duplicates are removed before delivery, and you pay one credit per delivered lead only.
- What happens to leads that fail my filters?
- They are dropped before delivery and never charged. A job that ends with nothing to deliver says why (no matches, all duplicates or all filtered) and costs nothing.
Related
- Live lead sourcingThe public sources behind every job, how a source is chosen for each prompt, and what the website crawler checks on each lead.
- Email verifierSyntax, MX, catch-all and mailbox checks, in bulk or from your assistant.
- Prompt libraryCopy-ready prompts grouped by what you sell, every filterable website signal, and the anatomy of a prompt that returns a clean list.
- PricingFree to start, plans from $29/mo, pay per delivered lead.
- Live demoA finished job replayed end to end, no signup or credits.