A potential lead is a reason to check the need. A detailed task or a high AI score does not guarantee a purchase.
Separate demand from supply
A request describes a task, seeks a specialist or asks about the conditions for doing some work. An advertisement offers the author's own service. A budget and deadline clarify a request, but their absence does not make it worthless.
A link, emoji or the word “free” does not prove spam. A buyer may attach a reference website, and a provider may legitimately answer a pricing question. Read the meaning rather than using one visual feature as a verdict.
| Illustrative message | Assessment | Reason |
|---|---|---|
| Need a website with a menu and booking | Potential lead | A task for a provider |
| We build websites, launch discount | Service offer | The author is selling |
| What does a site like this usually cost? | Borderline | Interest, but unclear scope |
| The website is down again | Needs context | A complaint is not necessarily a request |
| Thanks, we found someone | Closed request | Another pitch may be unnecessary |
Keep room for ambiguity
“Does anyone know a good photographer?” can be a useful request even without a date or budget. A recommendation request should not automatically be dismissed. Check the fit and clarify what is missing.
Vacancies, forwarded posts and recommendations depend heavily on context. A job advert may be irrelevant to an agency but useful to an independent specialist. A forwarded request may be old: inspect the original and its date.
Understand what AI can see
LeadJet sends keyword matches to Jev for assessment. Suitable and borderline results can be passed on to the customer, while definite non-leads are filtered out. This leaves the final judgement on ambiguous opportunities with a person.
One message cannot reveal the whole discussion, the author's budget or your ability to fulfil the job. AI can misread negation, irony and short replies. Treat the notification as something to review alongside the original, not as a verified order.
A score is not a sales probability
A high score represents the classifier's assessment of its task. It does not mean that the same percentage of people will buy. That interpretation would require separate validation against actual sales outcomes, not just labelled texts.
A stricter filter reduces noise but may lose useful requests. A more inclusive filter preserves opportunities while leaving more borderline messages for review. There is no single threshold that works equally well for every service.
Illustrative calculation: 5 of 8 selected messages are relevant, so precision is 5/8. If the reviewed stream contained 12 relevant messages in total, recall is 5/12. Neither measure is a payment conversion rate.
Qualify the opportunity before replying
Check that the request is current and fits your service, language, location and availability. Ask one important missing question rather than sending a long questionnaire as your opening message.
When reviewing search quality, record the reason for a mismatch: advertising, another service, a discussion or a closed task. This is more useful than counting notifications alone. A good filter saves attention; it does not replace professional judgement.
Sources
Examples and calculations are illustrative unless explicitly stated otherwise. They do not promise a particular number of customers or sales.