Why aren't leads reaching me? a triage playbook

A quiet funnel is rarely a single problem. There are four common shapes, each with a different cause and a different fix. Reading the symptom correctly cuts your time-to-fix in half. This is the triage playbook.

Funkel’s discovery pipeline has three stages: a cheap filter (size, region, role), an enrichment step (profile data, company data), and an expensive filter (the AI scoring against your ICP and the signal payload). The “Why leads aren’t reaching you” panel on the agent detail page shows where candidates are dropping. Open it before you read the four symptoms below; the answer is usually visible there.

Symptom 1. Zero leads in the pipeline

Most likely cause: the cheap-filter stage is rejecting everything. Either your ICP is over-narrow (titles too specific, company size too tight, region list too short) or the signal you turned on is producing fewer candidates than you assumed.

Diagnose: open the rejection panel. If the cheap-filter line is responsible for the entire drop, the fix is the ICP. If the line is empty (no candidates entering the pipeline at all), the signal itself is the problem, either there is nothing to find right now or it is misconfigured.

Fix:

• Widen one ICP field at a time. Start with company size (move 50-200 to 50-500), then region, then titles.
• If you are running competitor_pain with a competitor that does not have public complaints right now, add a second competitor or switch to recent_job_changes as a baseline signal.
• If you are running an engagement signal (company_followers, competitor_urls), confirm the company URL or post URL you configured is correct and active.

Symptom 2. Leads enter but get rejected at the score step

Most likely cause: the AI scorer is doing its job but your score floor is set higher than your ICP can actually produce. The expensive-filter step rejects candidates whose ICP-fit score sits below the floor.

Diagnose: the rejection panel will show a spike at the score-floor line. Click into a few rejected candidates. The AI reasoning is persisted (recently improved): you can read why each one was rejected. Look for a pattern.

Fix:

• Lower the score floor by ten points first. Run for 48 hours and read the new candidates. If they are tolerable fits, keep the lower floor; if they are stretches, raise back five.
• If the AI reasoning consistently flags the same gap (“titles do not match”, “company size unclear”), edit the ICP description to spell out what you actually accept. The scorer reads the ICP as text; vague text produces strict scoring.
• The AI now receives the full ICP shape (recently fixed). If you set up your agent before that fix landed, re-save the ICP to make sure the scorer is working with the complete payload.

Symptom 3. Leads accept the invite but never reply

Most likely cause: the connection note worked but the first message after the accept is the problem. Either it is too long, too sales-y, or it does not connect to the signal that earned the connection.

Diagnose: open three or four sent first-messages from the campaign. Read them out loud. If any of them sound like the kind of message you would archive without thinking, that is your answer.

Fix:

• Cut the first message to two or three sentences. Reference the signal that triggered the original outreach, then ask one specific question. No pitch, no calendar link, no paragraph about your company.
• If you are using AI mode and the messages all read similarly, the prompt is generating to a template. Tighten the tone instructions in the agent settings (give it three concrete examples of voice you want).
• Try the free Connection Note Critique tool on a sample first message; it scores the same things that predict reply rate.

Symptom 4. Leads reply but they are the wrong people

Most likely cause: the ICP is too loose, or the signal you are running surfaces a different intent than you read into the name. top_active_profiles sounds like buyers but it surfaces content authors; recently_raised_funds finds congratulators as often as founders.

Diagnose: look at the last ten replies and sort them into “right buyer, wrong moment”, “right buyer, right moment”, and “wrong buyer entirely”. If the third bucket is more than a third of the volume, the ICP is loose. If the first bucket dominates, the signal is wrong for your funnel stage.

Fix:

• For ICP-loose: add a hard-no clause to the ICP description (“not students, not consultants, not agency owners unless they are buying for a client”). The scorer respects negative criteria.
• For wrong-signal: switch from a content-discovery signal to an intent signal. competitor_pain and your_profile_url have the cleanest buyer-intent shape.
• Reject loudly inside Funkel. The reject button on a lead is a signal back to the scorer; rejected leads inform future scoring on similar profiles.

The rule for all four

Change one thing, wait 48 hours, then read the new data. Tuning two things at once means you cannot tell which change moved the numbers. The teams that fix quiet funnels fastest are the ones that resist the urge to overhaul everything in one afternoon.

For a deeper read on which signals predict what, see our field guide to LinkedIn intent signals.