Case Study / Built in-house
Every agency says it is different. We were following up leads by hand, same as everyone. So we became the first tenant of our own operating system and pointed Joseph at our funnel. This is what we built, what it does, and where we held back on purpose.
The problem
Joseph started as a dashboard. A nice one, but a dashboard. It showed the pipeline. It did not move it. Leads came in through a scan, a brief, or a chat, and then a human had to remember to look, score in their head, and follow up before the interest cooled.
When I grew ILAC enrollment eight times over, I was in the room. I was the system. I caught the signal, I made the call, I closed the loop. That works right up until it does not, because a founder in the room does not scale, and the leads that arrive at 11pm on a Friday do not wait for Monday.
The honest version of the problem: we had real intake and no real follow-up discipline. The fix could not be another tool we forget to open. It had to run on capture, on its own, and tell us only when a human was actually needed.
The room does not scale. The system has to be the room.
How Joseph was wired in
Step 1
Most agencies have one form. We had three doors: a public site scan, a Nadia brief, and a start-chat. The trap is three doors with three inboxes. So all three call one function and write one record keyed by the email. Scan someone, then have them chat, and it is still one lead, not three.
Step 2
The record is scored on capture, not on a Monday review. Domain quality, source intent, how weak their current site scored, how severe the gap is, and any prior engagement. That produces a number and a temperature. Hot, warm, or cold is decided by code, not by mood.
Step 3
A hot lead is not a row that waits. It fires a realtime alert to the operator within seconds and kicks off an enrichment pass. The alert is deduplicated for thirty minutes, so one excited prospect refreshing the page does not turn into ten pings.
Step 4
Every captured lead is mirrored into the revenue view the same instant it is created. Before this, a chat lead that never escalated to a brief was invisible on the money board. Now the board is the truth, and Joseph reads from it.
How a lead gets scored
The number is not a guess. It is the sum of five bands, capped at 100. Seventy and up is hot, forty and up is warm. These are the real weights the code uses.
Domain quality
0 to 25 pts
Real company domain, business email over a free inbox.
Source intent
0 to 20 pts
A booking beats a scan beats a chat beats a newsletter subscribe.
Scan signals
0 to 30 pts
The weaker their current site scores, the higher the opportunity.
Gap severity
0 to 15 pts
Positioning, conversion, and lead-generation gaps weigh the most.
Engagement
0 to 10 pts
Opens and clicks on prior outreach add late signal.
Results
One honest caveat first. Beonbrand runs this as a solo-founder tenant, so the headline is not a sales-team ARR chart. It is follow-up discipline: how fast a lead is caught, scored, and contacted, and how little of that needed a human.
Apr to Jun 2026 · illustrative sample · live Firestore figures replace these at launch
63
Leads captured
Across all three intake paths. Split: 28 scan / 21 brief / 14 chat.
37%
Scored hot on arrival
Decided by the intake score, the moment the lead landed.
100%
Hot leads contacted under 24h
Without manual triage. The alert and enrichment pass ran on capture.
Under 9s
Capture to operator alert
From lead written to operator pinged. Deduped for 30 minutes.
24
Dropped-intent clicks caught
Clicked the diagnostic CTA and did not book. Previously invisible.
Show your work
None of this is a slide. Every claim above maps to a file you could open. Here is the mechanism and where it lives.
scan-site.ts · nadia-brief-lead.ts · _leads/priyaPipeline.tsA site scan, a Nadia brief, and a start-chat all call the same upsertLead() and converge on one deterministic lead record. No duplicate leads under load.
_leads.ts · calculateIntakeScore()Each lead gets a 0 to 100 score from domain quality, source intent, scan signals, gap severity, and engagement. 70 and up is hot, 40 and up is warm.
_runtime/hotLeadAlert.ts · alertHotLead()When a lead crosses into hot, the operator is pinged in seconds. The same alert is deduplicated for 30 minutes so one lead never spams the channel.
_workflowLeads/capture.ts · writeCapturedWorkflowLead()Every captured lead is mirrored into the money pipeline view, so a chat lead is never invisible on the revenue board.
cta-lead-trail.tsClicked the diagnostic CTA but never booked? That intent is written to a trail, so the gap between click and booking stops being a blind spot.
useFounderHomeSnapshot.ts · JosephMoneyPanel.tsxAsking Joseph who is in the pipeline pulls live data from the same store, not a fabricated number. When data is absent, it says so.
Live walkthrough
A short recording: a real scan, the lead getting captured and scored, the hot alert firing, the revenue board updating, and Joseph answering who is in the pipeline. No edits, no narration tricks.
Walkthrough pending
The recorded walkthrough is being captured. It will appear here as a Loom embed: scan to capture to hot alert to revenue board to Joseph.
What is on, what we held back
The most honest thing a system can do is tell you what it is not doing yet. Here is the line, drawn on purpose.
On
The full intake-to-visibility path runs in production. Leads are captured, scored, alerted, and mirrored to the revenue board today.
On
Joseph answers in natural language about the live pipeline. This is on, reading from the same store the board reads.
Held
The runtime tool-loop and live drip email sends are built and instrumented but held behind a founder switch. The infrastructure is ready. We have not opened the live-send tap. That restraint is deliberate, not a gap.
What we would do differently
The score is a strong first cut, but the weights are a hypothesis until enough deals close to confirm them. The fix is to tune the bands against which leads actually became revenue, not against what felt important on day one.
The most useful thing we added was not the alert. It was capturing the click that never became a booking. That gap was completely invisible before, and it is where the easiest follow-ups live.
Automating intake was the right call. Automating the proposal would not be. The system should clear the noise so a person spends their judgment on the moment that needs it, which is the offer, not the inbox.
Start with the diagnostic, or run a scan on your own site and watch the first step happen to you.