Lead Funnel Optimizer
Find Where Your Funnel Leaks, Then Fix It
Your agent reviews a live lead funnel the way an experienced conversion and lead quality operator would. It finds where the funnel leaks and why from its own data, ranks the fixes, then makes them or tests them. Every claim in its report comes from a number it read, with the source.
Free with a LeadCapture account (trial or paid). Already connected? Your agent already has this skill.
The diagnosis approach
Evidence first.
Opinions never.
Gathering the data is read only and free. The agent reads everything before it concludes anything.
- 1
Scopes in one message
Which funnel and the goal: more leads, better quality, fewer rejected or returned leads, or lower cost per lead. Plus the traffic source, recent changes and any buyer complaints. It compares the last 30 days with the 30 before.
- 2
Gathers the data
Structure, analytics by day, drop-off per step, Funnel Forensics conversion signals, step friction, real session recordings on the worst step, settings and styles, delivery health and lead quality.
get_funnel_analyticsget_step_metricsget_recording_insightsget_step_frictionlist_recordingsget_integration_healthlist_leads - 3
Diagnoses
- Compares the funnel with itself first: last period, mobile vs desktop, step vs step
- Separates traffic problems (people who were never going to convert) from funnel problems (people who tried and got stuck)
- Checks delivery and lead quality before conversion: a funnel feeding a failing integration is the most urgent problem there is
- Small numbers lie: on a low-traffic step it says the signal is weak instead of ranking it
- 4
Reports on one screen
The headline problem first, every finding with its number and source, then the ranked fixes.
- 5
Fixes or tests, with your yes
Lists exactly what it will change and waits for your yes. Bugs and best-practice gaps get changed; judgment calls get an A/B test.
- 6
Follows up
Checks running tests and never calls a winner early. If you want, it sets up a weekly check on your own agent that flags only what changed.
What it looks for
What it sees, what it means,
what to do.
| What the data shows | Likely cause | What the agent does |
|---|---|---|
| High first-step drop-off, mostly quick exits | Traffic, not the funnel: wrong audience, ad promise does not match, slow page | Checks the ad message and page speed with you. Says plainly that funnel changes will not fix it |
| High first-step drop-off, visitors read then left | First question too hard or too personal, or the offer is unclear | Tests an easier first question and adds a first-step callout |
| One step much worse than its neighbors | Confusing wording, too many options, or a question that feels too personal | Reads answer clicks and changes, adds a WHY WE ASK callout, rewords or merges options |
| Validation errors on a typed step | Format too strict, unclear placeholder, or a bug | Reads the messages and recordings; errors on valid input go to LeadCapture support |
| Mobile far worse than desktop | Layout, keyboard covering the button, long option text | Reviews mobile recordings on the worst step |
| Buyer rejecting or returning leads | Fields or values off spec, duplicates, bad phones | Compares a delivered payload with the spec field by field and checks validation and duplicate settings |
| Junk details at bot-like speed | Bot or incentivized traffic | Fraud scoring first, then a talk about the traffic source |
The session evidence comes from Funnel Forensics, and judgment calls run through A/B testing.
Audit checklist highlights
Every gap is a finding.
Each “no” on the checklist is a finding. Most are fixes, not tests.
Structure
- Easy first step, never contact details
- Disqualifying questions early, routed to a polite DQ step
- One question per step, buttons over dropdowns
- Dynamic transition before the name step
- Thank-you step says what happens next
Reassurance
- Callout on the first step
- Callout on every personal step
- WHAT HAPPENS NEXT on the contact step
- No callout claims what the buyer cannot back
Styles
- Progress bar hidden on step 1
- Brand color on buttons, progress and selections
Compliance
- Complete TCPA consent above the button
- TrustedForm or Jornaya if the buyer requires them
- Vertical footer lines where they apply
- No promises of approval, rates, savings or outcomes
Quality and tracking
- Email and phone validation on
- Duplicate Lead Check for sold leads
- UTM Auto Capture for paid traffic
- Lead event on one step only
- Delivery healthy
How testing works
Change the bugs.
Test the judgment calls.
Change now
- A validation error on valid input
- A broken route
- A missing consent line
- Progress bar on step 1
- No callout on a personal question
- Delivery failing
A/B test first
- The first question
- Removing a question the buyer does not need
- Form only vs a landing page
- The headline and offer
- The contact step headline, button and callout
How a test runs
- Creates a draft experiment; the variant is a clone, integrations included
- Makes one change on the variant, and shows you both
- Starts it only after your yes
- No winner until the sample is ready, and at least a full week
- Judges on leads the buyer accepts, not just raw leads
The report
One screen. Worst problem first.
Every number sourced.
Every finding names its source: analytics, recordings, change history, leads or delivery. At most five findings.
Example prompts
Ask what is wrong.
Get evidence back.
- Reads analytics, drop-off and Funnel Forensics
- Watches real sessions on the worst step
- One-screen report with ranked fixes
- Checks delivery health and the latest error
- Compares a delivered payload with the spec
- Reviews phone, email and duplicate settings
- Runs the audit checklist
- Lists each gap as a finding
- Offers to fix the gaps after your yes
- Reads the experiment results
- Says plainly if the sample is too small
- Applies the winner only when it is decided
How to get it
Connect LeadCapture and you
already have the skills.
The LeadCapture connection serves all three skills to your agent, always the current version. Installing your own copy is optional.
Connect LeadCapture
Connect at Agents & Skills in LeadCapture, or add the MCP URL as a custom connector.
- Claude: Settings > Connectors, add the URL and sign in.
- Claude Code: run the command below.
- Other agents: add the MCP server from Agents & Skills.
No account yet?
The skills are free and need a LeadCapture account. Start a 7-day free trial.
Then just ask
“Build me a final expense lead funnel” or “Why is my Medicare funnel not converting?” Your agent picks the right skill.
Prefer the app?
Choose Build with your own agent, the first option when you create a new funnel.
Optional: install your own copy
Only if you want to keep or edit your own. An installed copy always takes priority over the one the connection serves.Claude Code
pluginClaude
web and desktop- Download the plugin zip from Agents & Skills.
- In Claude, go to
Plugins > Add > Upload plugin. - Turn on Create and edit files in
Settings > Capabilities.
Codex and other agents
skills folder- Copy the folders under
plugins/leadcapture/skills/from the repo. - Paste them into your agent's skills folder (Codex:
~/.codex/skills/). - Keep LeadCapture connected so the agent can do the work.
The plugin does not add the LeadCapture connection, so it never duplicates one you already have.
Lead Funnel Optimizer FAQ
What does the Lead Funnel Optimizer do?
It reviews a live LeadCapture funnel using its own analytics, Funnel Forensics session insights, lead quality and delivery health, finds where people drop and why, and then fixes the problem or sets up an A/B test.
Will it change my live funnel without asking?
No. Before any change it lists exactly what it will do and waits for your yes. Changes to a live funnel reach real visitors right away, so it never switches on anything that turns visitors away without an explicit yes.
Does the analysis cost anything?
Reading your funnel data is free. A fresh Funnel Forensics AI run uses an AI run, so the agent only offers one when the latest snapshot is missing or old, and only runs it after you say yes.
What if my funnel has little traffic?
The agent says when a signal is too weak to rank, and it never calls an A/B test winner before the sample is ready.
Will it break my buyer delivery?
It checks the integration mapping before renaming fields, removing required questions or changing answer values that delivery depends on.
Does it show what visitors typed?
No. The report uses ids and counts, never what a visitor typed.
Find the step that wastes your ad spend.
Connect LeadCapture and ask your agent why a funnel is not converting.
Free skills. A LeadCapture account (trial or paid) is required.