Built for Central Florida service businesses. Most agencies sell you a video and a dashboard. This is different: fourteen specialist agents, one shared brain, and a closed loop from your discovery call to your closed jobs. Here is the entire thing, client-facing and internal, nothing left out.
No black box. This is exactly what happens after you book a discovery call, in order.
The call runs through Granola. A watcher checks for the transcript every 15 minutes and writes it into the shared brain, tagged to you. Your numbers, your goals, your exact words. Nothing you say gets re-asked or lost in someone's notes.
transcript → brain, 15-min cycleNot a templated pitch deck. The proposal engine takes your average job value and your close rate and shows Starter, Growth, and Pro side by side, with the math done on your business. Then it makes one recommendation by a fixed rule: the cheapest tier that actually fits your stated goal. Not the most expensive one we can talk you into.
deterministic rule, not a sales scriptLive Stripe payment links, one per package, built straight from the catalog. The per-job fee, where it applies, is reconciled monthly against real closed jobs and billed by invoice. Pricing: Starter $7,500 pay-in-full or 4 × $1,875. Growth $10,000 or 4 × $2,500. Pro $12,000 or 4 × $3,000. Plus your Meta ad spend, paid direct to Meta, never marked up through us.
stripe catalog, live linksper-job fee only on real closed jobsYou log in once with a magic link. The portal walks you through one step at a time: connect your Meta ad account as a partner (about 5 minutes, we initiate, you approve), book your kickoff call straight from the portal, then the shoot gets scheduled. Each step unlocks when the last one clears.
meta connect ~5 minkickoff booked via cal.comYour kickoff answers feed nine scripted concepts: six for people who don't know you yet, one proof piece, one offer explainer, one guarantee video in your own words. The shoot is an interview, not a teleprompter. We record the body once and stack 10 to 20 different 3-second openers on top of it. One body, many hooks. Three hook variants per concept means 27-plus tested ad units from a single day.
You get a shot list 48 hours before the shoot. Your only job is to show up and talk about your work.
9 concepts · 3 hooks eachshot list 48h aheadThe full campaign is built as a spec: structure, targeting, budgets, creative, copy. It goes into Meta paused. We verify your pixel and Conversions API are reporting before a dollar moves. We never launch pixel-dark, because numbers you can't trust are worse than no numbers. Then it flips live, and the 14-day guarantee clock starts.
Honest note: for our first clients the launch step is constructed by hand from the agent-built spec while the direct Meta connector finishes. shipping next
pixel/CAPI verified before spendguarantee clock starts at launchThe MIT lead-response study on 15,000-plus leads: contact inside 5 minutes versus 30 makes you roughly 100x more likely to connect and 21x more likely to qualify. So the moment a lead lands, Concierge sends a branded email with a booking link instantly, and a human calls within 5 minutes. Day or night, the email never waits.
Texting goes live about 2 weeks after signup, once carrier registration for your business clears. shipping next · ~2 weeks per client Automated voice is on the roadmap; until then the call is a real person, assigned at kickoff. shipping next
instant email · call <5 minquiet hours + STOP compliance built inConfirmation at booking, a reply-Y reminder at 48 hours, a final reminder same-day. Healthcare trials put this pattern at a 30 to 50 percent no-show reduction. A missed appointment drops into a same-day "want another slot?" sequence, not the trash. Then the booked, qualified appointment is handed to you. You close. We don't pretend to.
24h / 3h / 15m remindersLeads, cost per lead, booked jobs, spend, and the verdict in one honest line. If the week was slow, the email says slow, names the reason, and states the fix. Never spun. The portal answers the same four questions any owner actually has: is it working, what's working, why, and what we're doing next.
weekly email + live portalslow weeks reported straightWhere a per-job fee applies, it's reconciled monthly against jobs you actually closed and confirmed, with a dispute window. The lead count is kept by a dedicated CRM agent whose entire job is honest accounting: dedup, source tags, qualified-versus-raw. Your leads stay your data, exportable any time.
qualified-lead system of recordWe only publish results that happened, from the same funnel data you watch in your portal. Attribution is shown as arithmetic, never as a magic platform number. We present your CRM funnel as the result and Meta's count as Meta's claim, because that gap is exactly what a skeptical owner should see.
One orchestrator, thirteen specialists. Every agent reads the shared brain before it acts and writes back when it's done. Here is the full roster, grouped by what it does for you.
Humans still own what humans should: every sales call, every approval gate, every client-facing send during ramp, and the final cut on anything that matters. Every escalation path ends at the founder.
This half of the page is the internal map. We publish it anyway, because the machine is the product and hiding it would be the opposite of the brand.
Every transcript, decision, mistake, and result lives in one markdown knowledge base. Agents read it before acting and write back after. Transcripts file in automatically on the 15-minute Granola cycle. A nightly consolidation pass at 2 AM merges the day's raw captures into the canonical notes, flags conflicts for human review, and regenerates the index. Nothing starts from a blank slate, ever.
Design rule worth stating: plain markdown and grep over a vector database, at this scale. Full provenance, human-readable diffs, zero infrastructure that can silently disagree with itself.
markdown brain · nightly 2am consolidationAngles come from voice-of-customer language: your reviews, competitors' 1-to-3-star reviews, and competitor ads that have survived 6-plus months in the Ad Library. Five pains times five lenses is 25 candidate angles before a word gets written. Then the fan-out: one strong model writes the spec, 8 to 12 cheap workers each write exactly one script in parallel, and one strong judge scores them all adversarially and kills the weak ones before any money touches Meta.
Every batch must pass a freshness gate as a set: at least 3 angles, 2 awareness stages, 2 formats, 2 emotional tones. Meta's current algorithm penalizes sameness; diversity is the lever, and the gate enforces it.
plan → fan out → judge → gateweak variants die before spendThe doctrine, in order: quality first. The strong model is every agent's default. A task may only move to a cheaper model after it passes a golden-set eval proving parity, case by case, with zero regression allowed. No eval gate, no downgrade, no exceptions. Client-facing and judgment work never moves, at any savings.
Why bother at all: tokens moved to the grunt tier after passing their gate cost about 90 percent less, prompt caching bills repeated context at roughly a tenth (verified in telemetry, not assumed), and the whole stack sits under a hard budget cap with per-agent slices so no single bug can run the bill. The honest version: at current scale the absolute savings are single-digit dollars a month. The point is flat per-client cost at scale and that a cheaper model can never quietly wreck a deliverable.
eval-gated downgrades onlyhard cap + per-agent budget slicesPer action: read the brain, act, log the result. Nightly: the 2 AM consolidation crystallizes the day into tomorrow's context. Weekly, two rhythms: the Monday loop pulls your last 7 days from Meta and the CRM, diagnoses the funnel (low hook rate means fix the first 3 seconds; good clicks but no bookings means the problem is downstream, not the edit), and briefs the next batch on a 60/30/10 split of proven winners, iterations, and fresh swings. Friday, the orchestrator reads every agent's signals and writes the decisions the next batch is built on.
On top of that sits the opportunity engine: a daily pass where every agent flags chances to move your number, scored against one gate: does this help the client make more money from the ads. If it can't be said in one sentence, it gets dropped. Over-delivery that doesn't touch revenue is scope creep wearing a costume.
monday loop · friday synthesis · nightly ingestPixel and Conversions API verified before launch, always. UTM source tags stamped at CRM entry and persisted lead to booked to showed to closed. The portal renders the same numbers the agents write, never a re-entered copy. Return is shown as arithmetic the owner can audit: spend divided by closed jobs, job value times jobs closed. And the limit stated plainly: source tags are attribution, not proven causation. We say that out loud because the agencies that don't are the reason owners stopped trusting agencies.
never launch pixel-darkarithmetic, not platform magicEvery completed job feeds the machine: the before-and-after footage becomes next week's ad, the review ask fires at the satisfaction peak, the close data tunes what Meta optimizes toward. Creative fatigues in roughly two weeks on Meta right now; a system that manufactures fresh, real footage from your own delivery is the only durable answer to that.
Video ads can't fix a broken reputation, save a business that can't take more work, or close jobs for you. If that's the real problem, we'll say so on the first call, before you've spent a dollar. Disqualifying bad fits is policy, not politeness.
The clock starts the moment your ads go live, after the pixel is verified, never before. We take two new clients a month so every one gets the full machine, not a slice of it. If the system above sounds like what your business is missing, the first step is a 30-minute conversation.
Book a discovery call →Scale By Video · Central Florida · honest video ads for service businesses · scaleby.video