Discoveries briefing

What one day and night of research and building taught us about Service Studio. The take, not the process. Every number here comes from a case file; anything we have not proven yet is marked as a guess.

19 cases, 4 red teams, 5 scale reports 13 calls waiting for you Written Oct 27, 2026, 3:30 AM MT

If you read nothing else

The plan holds, but each job is thin. A $300 job leaves about $57 after pros, dump fees and the cost of the customer. That falls to about $27 if pros get more than 63% or a customer costs more than $40.
The company cannot sell yet. The LLC was dissolved by the state on June 21, 2026, and two brand names have problems (Everline is taken; Bin Buddy is blocked in Utah as an exact name). Reinstating costs about $82; nothing gets filed before that.
Our first junk prices lost money. New prices go by crew size ($189 and $269 for one person, $329 and $449 for two), set just under the big franchises.
The engine is real and tested: 348 unit tests, 19 of 19 real-browser journeys, a full production rehearsal, English and Spanish. Still open: 28 safety builds before the first live job.
Your first call: P13. Approve a small capped launch fund (LLC, name filings, insurance, background checks), or the live launch waits. Profit will come from repeat customers, property-manager accounts and dense pro coverage, not from ads.

The 10 biggest discoveries

Ranked by how much each one changes what we do next.

The company that would own the brands is dissolved

What we found: Red Pine Creative Marketing LLC missed its January 31, 2026 renewal. The state dissolved it on June 21, 2026. The engine also had the wrong legal name in it ("Red Pine Creative LLC" belongs to a stranger's old company).
Why it matters: A brand name (DBA) is filed by a live company. Stripe, receipts, insurance and contracts all need a seller in good standing. Nothing can go live until this is fixed.
What we did: Froze all name filings and domain buys. Built your step-by-step go-live kit. Reinstatement is allowed for two years and costs about $82 (your call 10).

Source: case C10 names, go-live kit

Our first junk prices lost money

What we found: The $139 / $199 / $279 tiers fail the margin floor in every scenario once insurance and the cost of getting a customer are counted. Even $169 / $249 / $349 only worked if a customer cost under about $21 to $44.
Why it matters: Cheap owner-operators (one charges $150 for an in-home half load) will always beat us on price. Insured two-person crews cost real money.
What we did: Repriced by crew size: $189 / $269 one person, $329 / $449 two people. Pros get up to $95 / $135 / $170 / $215 plus the dump receipt. Positioned as an insured local brand just under the franchises. Software now blocks any job below the floor.

Source: C4 unit economics, red team C4/C5/C9, C11 margin

Profit per job is thin, so volume and repeat business are everything

What we found: On a $300 average job: $174 to the pro, $21 dump and supplies, $9 card fees, $6 refund reserve, $8 insurance and tools. That leaves $82, then about $25 to win the customer, so $57 per job (19%).
Why it matters: Small slips erase it. At $50M a year, paying pros 5 more points costs $2.5M; $10 more per customer costs $1.67M; refunds rising from 2% to 5% burns $1.5M.
What we did: Made the margin floor law in code. Every booking records where the customer came from. The money model gets rebuilt from real numbers at 100 and 500 jobs.

Source: round 2 scale economics, C19 synthesis section 2

The app is the small part; the company is 14 departments

What we found: Every big failure in this industry came from a missing rule, not a missing feature. DoorDash nearly died in Sept 2013 from a rush it could not switch off. HomeAdvisor paid up to $7.2M for selling weak leads. Instacart paid $60M for fees the screen hid.
Why it matters: We can copy their prevention rules on day one instead of learning them through lawsuits.
What we did: Wrote the company operating system: 14 departments, each with a daily, weekly and monthly rhythm, plus 28 "build before the first live job" items (booking cap, a defect code on every job, a claims desk, a written pro removal and appeal policy).

Source: C19a big platforms, C19 synthesis

An all-in-one app like TaskRabbit or Angi would be a mistake

What we found: Angi's lead network revenue fell 72% in 2025 after it let homeowners pick their own pro. Its managed, pre-priced line fell 23%.
Why it matters: Your instinct was right: being the brand that books, charges and stands behind the work beats running an open app.
What we did: Kept the branded-company model. You then chose many brands over one brand (call 13), so the research risks became engineering requirements (see the Model section).

Source: C19d model check (Angi 10-K for 2025)

The closest look-alikes got sued, fined or folded

What we found: Handy paid a $6M California settlement (pros must set their own rates) and $2.95M to the FTC and New York in 2025 (earnings claims, hidden fines). Homejoy died in 2015 after a $19 first clean, a 25% cut, customers going around it, and expanding abroad too early.
Why it matters: The legal line is control. If we set the pro's pay, hours and penalties, they start to look like employees.
What we did: Locked in: pros accept, decline or counter each payout (inside a ceiling the margin allows); zero fines ever; scores only change who sees offers first; no hourly or monthly earnings promises in recruiting copy.

Source: C19b home-service marketplaces

Selling extras at the door triggers the FTC 3-day cancel rule

What we found: Our first fix ("the office prices extras by phone") failed the red team: the rule covers sales made at the home even when the customer invited us.
Why it matters: Any sale made on site could be cancelled for 3 days and needs a formal notice. A plan offered by the pro at the door counts too.
What we did: Extra junk is charged only at a per-yard rate the customer approves online when booking, with a cap. Pros never quote, sell or collect on site. Plans are offered by email after the job is charged. Same-day extras stay switched off until the attorney answers.

Source: C12 payments, red team C12, C13, C19 section 7.1

Customers come from Google and partners, not from paid ads

What we found: Google Local Services Ads cost about $94 to $134 per booked junk job. Industry data: $117.92 per landscaping lead and $46.99 per cleaning lead (leads, not jobs). Facebook Marketplace bans service listings. Nextdoor junk threads in Draper were nearly empty.
Why it matters: Paid leads break the margin floor at our prices. Google allows one business profile per real location with its own staff, so map pins for their own sake are not allowed.
What we did: No paid ads for 60 days. Google Business Profile is the main engine; first jobs come from your network and $5 Craigslist posts. Planned mix at $1M: 40% property-manager and other business accounts, 25% Google, 25% repeat and cross-brand, 10% paid.

Source: C2 demand rounds 1 and 2, round 2 demand at scale, red team C4/C5/C9

Money could leak in 26 places; 8 were wide open

What we found: The engine's job ledger is solid but it is not a set of books. Of 26 leak paths: 7 caught, 11 partly caught, 8 open. Examples: the $15 card dispute fee was recorded nowhere; a won dispute stayed a loss; a failed pro payout still showed as paid; sales tax was collected but never set aside.
Why it matters: This is exactly your fear. Sales tax is the state's money held in trust, and spending it is a personal-level problem.
What we did: Designed the accounting department: 5 AI roles (Recorder, Reconciler, Controller, Analyst, Tax and Compliance), a CPA and you. No AI can move money. No role checks its own work. Two bank accounts (operating and held money). Your time: about 10 minutes a week plus month end. The leak fixes are written but not merged yet.

Source: C18 accounting parts A and B, C18 design

A human operations lead is needed from the first live job

What we found: Round 1 said "zero staff at $1M". Round 2 disagreed: every company running AI support at scale still had people for escalations, and physical jobs create damage and safety incidents. $1M needs 2 to 3 people besides you; $10M needs 5.
Why it matters: Your 10 hours a week stays the same at $1M, $10M and $50M only if someone else holds day-to-day authority.
What we did: Until revenue pays for it, the ops lead is you or someone you name. That hire is the first thing revenue funds. Claims over $250 go to a human; AI can never deny a claim.

Source: round 2 AI-native org, round 2 risk and claims, C19 section 0

Surprises and changed minds

Things we believed on Oct 25 that turned out wrong, and things nobody expected.

Entity

The LLC is dissolved

We assumed the company existed. The state dissolved it June 21, 2026 for a missed renewal. Fixable for about $82 within two years.

C10

Names

Two brand names are taken

Bin Buddy: Utah says the name is unavailable, a Bin Buddies LLC formed in West Jordan in July, and binbuddyco.com is a live bin cleaner. Everline: a federal trademark that covers snow removal, with a Salt Lake office. Grab the Gig is risky too: a pending "Grab-A-Mow" trademark belongs to a company recruiting the same Salt Lake pros.

Proposed: Roomwise Hauling, Rakewell Yard Co., Pinecrew. Salt & Sun and Blue Fern are clear.

C10, red team C10

Prices

Old prices lost money

The approval-sheet junk prices failed in every scenario. The fix was pricing by crew size, because the $50 workers' comp waiver stops covering anyone once a pro brings a paid helper.

C4, red team C4/C5/C9

Reviews

Asking only happy customers for reviews is banned

Your idea was to catch the bad ones first. Google and the FTC ban "review gating". Built instead: every customer sees the Google link; low ratings also open a recovery item for you. We still keep every rating for pro scorecards.

C2, C3, decision ledger

History

Handy and Homejoy are the warning

Handy: $6M plus $2.95M in penalties for setting pay, fining pros and inflating earnings claims. Homejoy: dead in 2015 after a $19 promo, a 25% cut and early expansion. Survivors let the pro set the price, or hire employees.

C19b

Market

Angi's lead business collapsed

Network revenue down 72% in 2025; total revenue $1,030.5M, down 13%. Selling leads to pros is a shrinking business.

C19d

Law

On-site sales and the FTC 3-day rule

Our first workaround failed the red team. Now all extras are approved online at booking, inside a cap. An attorney still has to confirm.

C12, red team C12

Money

Accounting leaks were hiding in the code

8 open leak paths, including an unrecorded $15 dispute fee and a ledger line that overstated our margin by the dump reimbursement. Also: no brand tag on card payments yet.

C18 part B

Design

Generated art beat hand-drawn art

The first Grab the Gig redesign used hand-coded drawings and scored 7 out of 10 from you: right idea, not polished. Image generation from our own sketches gave illustrator quality in minutes at $0 on the ChatGPT plan. Now the rule for every site.

Hourly log, design memory

Languages

Multilingual means rewriting, not translating

A native writer drafts from the brief, not from the English. Spanish uses "tú" for pros and "usted" for homeowners. Slang like "troca" or "jale" only in texts and social posts, because Utah's fastest-growing Hispanic group is Venezuelan. Spanish hero: "El cliente ya dijo que sí."

C14, C17

Copy

Our recruiting headline scored 7 out of 20

"Got a truck or a yard crew? Get local jobs." was generic and invited uninsured side-hustlers. New: "Junk hauling and yard jobs, already booked. See your pay before you say yes." The bottom-of-page 9-field form is gone: a 2-input step sits in the hero and a sticky phone bar.

C15, C16

Benchmarks

Franchise numbers were smaller than we thought

Junk King's 2025 average franchise did $551,635, not the $1.41M we first used (that was a 2022 top quarter). College HUNKS averages $1.555M; mature 1-800-GOT-JUNK franchisees $3.70M.

Round 2 scale economics

Tax

Utah taxes some of our services

Car detailing and interior cleaning are taxable; leaf cleanup is likely exempt; junk removal is unclear, so we charge no tax on it until a CPA confirms.

Hourly log, C4

Texts

Text opt-out only understood STOP

The law expects "cancel", "quit", "unsubscribe" and Spanish words like PARE and ALTO too. Fixed and merged with the Spanish build.

C14, hourly log 03:05

What we know about the money

All of these are planning numbers built from research, not results. They get rebuilt from the first 100 and then 500 real jobs.

$57left per $300 job after the customer cost

19%of each job is ours, after everything

63%max pro share before the model breaks

$40max cost to win one completed job

One $300 job

Line	$	Share
Customer pays (before tax)	300	100%
Pro or crew	174	58%
Dump fees and supplies	21	7%
Card and payout fees	9	3%
Refund and damage reserve	6	2%
Insurance, texts, AI	8	2.7%
Before customer cost	82	27.3%
Cost to win the customer	25	8.3%
Ours	57	19%

The $300 is a blended average: junk $380, yard $280, cleaning $220, detailing $160, bins $40.

The margin floor (your call 1)

One-time jobs: keep at least $25 or 15% of the price, whichever is more, after the customer cost. Target $40 or 20%.
Repeat visits: at least $12.50 per planned route hour or 15%, whichever is more.
Checked: a $189 junk job keeps $30.64 against a $28.35 floor. A $25 bin visit keeps $5.45 against $3.75. Both pass.
Breaks if: a bin pro is paid more than about $13.70 a stop, or a plan customer costs more than about $45.
Enforced: the software refuses any quote, pay bump, extra or refund that breaks the floor unless you log an override.

The $1M, $10M and $50M picture

	$1M a year	$10M a year	$50M a year
Jobs a month	278	2,778	13,889
Jobs a working day	13	126	631
Cities	1	5	20
Brands per city	3	4	5
Pro businesses doing 3 jobs a week	22	214	1,070
Our share after costs	$190k	$1.90M	$9.50M
Office and city costs	$140k	$1.4M	$7.0M
Profit left	$50k	$500k	$2.5M
People besides you	2 to 3	5	City leads in 20 markets, plus claims, finance, legal
Your hours a week	~10	~10	~10
Fastest pace (stretch)	Quarter 5	Quarter 12	Quarter 20

Stretch scenario, not a promise. A 1% unrecorded leak at $50M is $500,000.

Where profit really comes from

Repeat and cross-brand customers

A customer costs about $65 to win and buys about 2.67 jobs over their life. Getting that to 3.20 jobs adds about $677k at $50M. One shared household record across brands is the asset.

Business accounts

Property managers, realtors and estate-sale firms book again and again. Planned as 40% of jobs at $1M.

Dense pro coverage

One more job per crew per week means about 803 crews instead of 1,070 at $50M. Busy pros cost less to keep and say yes more.

Not ads, not price cuts

Paid leads cost $94 to $134 per booked junk job. A promo far below market is what killed Homejoy.

Source: round 2 scale economics, round 2 demand at scale, C4.6 floor red team, C19 section 2.

The business model

What survived the research, what you decided, and the risks we are building around.

What holds

We are the brand. The customer books and pays us; an insured local pro business does the work. We stand behind it.
One back office, many brands. Grab the Gig (pros, pay, dispatch, claims, books) is shared. Each brand is a thin copy: name, phone, site, Google profile, voice.
Same offer everywhere. A one-time job, or a repeat plan every 4 or 12 weeks on trash day, paid per visit, cancel online in one step.
Pros are businesses, not workers. Owner-led crews of 1 to 5 with their own liability insurance, commercial auto, and workers' comp or the waiver for everyone on the job.
Junk first. In-home carry-out from photos, $99 to $399 range. We do not fight $79 curbside pickups or Sandy's free bulk pickup.

What you decided

Many brands for volume (call 13, Oct 27). The research recommended one brand first; you rejected it. Volume wins.
Think big. Every structure is sized for $1M, $10M and $50M from day one. Sequence the build, never shrink the vision.
Copycat brands. A brand may switch an option off, never add its own.
Every job meets the margin floor.
English and Spanish everywhere, more languages later.
$0 until revenue. This is now in tension with launch costs (see P13).

Risks we are engineering around

Risk the research raised	How we build around it
Every brand starts cold in every city	A brand cannot take bookings in a zone until signed, insured pros cover it. Each brand gets its own demand lane and a dated pass or kill test.
Reviews split across brands	One shared review engine, plus a family page showing the brands are one company.
Your hours grow with every brand	A hard owner-hours budget is a company metric. Going over it creates an automation task, not more work for you.
Several Google profiles at one home address can get suspended	One profile per real, verifiable location with its own crew. More locations only when the money supports a real base.
No cross-selling between brands	One household record with consent, and a cross-brand offer only after a job is charged.
Pros look like employees	Pros accept, decline or counter pay. No fines, no schedules, no uniforms. Removal only on written grounds, with notice and appeal.
A demand rush we cannot serve	A booking cap from day one: only show time windows that 2 qualified pro businesses can cover.
Damage and claims	Brand pays the customer first by fixed rules, then recovers through the pro's insurer. Never an automatic pay deduction.

Source: C19d, C19 sections 5 to 7, decision ledger.

Built and proven vs still a draft

"Proven" means checked by Claude in a clean copy of the code and in a real browser, not just "tests pass".

Proven

The engine: 348 unit tests, 19 of 19 real-browser journeys, accessibility checks passing in English and Spanish.
Production rehearsal in Docker: booking to payout, then backup, wipe and restore.
115 bug-bash findings fixed from 4 real-browser testers, including 3 blockers. A second round found 53 more, none blockers, all majors fixed.
Money adds up to the cent across 5 closed test jobs. 2,164 garbage requests caused zero server errors. The mail worker was killed 8 times and still sent 66 emails exactly once.
Money rules in code: saved card, holds, $39 and $75 fees, per-yard extras approved at booking, margin floor guard.
Security: rate limits, session controls, upload checks, security headers, 0 known package vulnerabilities.
English and Spanish across sites, portals, emails and texts, with a gate that blocks paid Spanish bookings until legal wording is reviewed.
9 AI agents pass 82 of 82 practice scenarios, running in shadow mode only.
Grab the Gig redesign (generated art, step-by-step sign-up, EN and ES): 332 unit tests, 16 of 16 journeys. Waiting for your design review before merge.

Still a draft

Prices and floors: in code, but provisional until you approve calls 1 and 4.
Legal pages (customer terms, pro agreement, privacy, texts): drafts for a Utah attorney.
Spanish legal and money wording: about 100 lines need a native Utah reviewer.
Accounting leak fixes: written, not yet committed. Opt-in backend and agent rules: committed, not yet merged.
28 safety builds before the first live job: booking cap, defect codes, accept or counter pay, claims desk, crew roster, pro removal and appeal, W-9 tracking and more.
Bin Buddy site redesign: built on a branch, but the brand name must change.
Switched off on purpose: texts (until a Twilio account), live AI agents (no API key), same-day extras (until the attorney), real payments (Stripe test mode not set up).
Company OS and accounting department: designs, not running departments.

Source: hourly log Oct 26 to 27, C19 section 8. Engine main branch at 53ec01a.

What we still do not know

Each unknown has a cheap test with a date. If a test fails, the decision flips.

By	Test	What it decides
Nov 9	Survey 15 to 20 local haulers: their prices, insurance, would they work under our brand	Whether insured pros say yes at our payout caps
Nov 9	Two insurance broker quotes	If our insurance costs more than about $25 a job, switch to a referral model
Nov 9	5-second test of the recruiting headline with 10 to 15 haulers	Keep the headline if 7 of 10 can say back what we offer
Nov 16	Native Utah review and test of the Spanish copy	Ship or rewrite the Spanish pages
Nov 30	Sign-up flow results; property-manager account test	Opt-in design; whether business accounts are a real lane
Dec 15	Mystery quotes from franchises, Lugg and locals	Whether our in-home price wedge holds
Jan 31, 2027	Ask pros on sign-up which other trades they do	How much one pro network really serves several brands
First 100 jobs	Track source, cost and outcome of every job	Real customer cost, repeat rate, claims
500 jobs	Rebuild the money model from real data	Replaces every guess on this page

Attorney questions (top ones)

Does the FTC 3-day rule reach extras the customer approves on their own phone at home?
Does our flow (we set the customer price, pros accept or counter) keep pros as independent businesses under Utah and IRS tests?
Is a card hold a "deposit" under Utah rules? Are the $39 and $75 fees enforceable as written?
Can we offset a proven damage claim against a future payout, with notice?
Does reinstatement cover the time the LLC was dissolved?
Utah's marketplace law only protects platforms that charge pros a fee. We assume it does not cover us and will never add a fee to qualify.

CPA questions (top ones)

Is residential junk hauling taxable in Utah? Are cancel and trip fees?
Do we report revenue as the full job price (we are the seller) or only our cut?
Do dump fee reimbursements and tips go on pros' 1099s?
One operating account plus one held-money account: enough for several brands?
Reinstate the old LLC ($82) or form a new one ($59): any tax reason to prefer one?

Also unknown until real data: signed pro rate cards, quote-to-paid rate, real dump and claim costs, whether a second brand at the same base can get its own Google profile, and how many customers agree to hear from a second brand. Sources: go-live kit part 5, C19 section 11, case board flip tests.

Decisions waiting for you

In priority order. Reply "approve P13, 10, 8" or "change 4: ..." and the board updates. Full cards on the main board.

P13: Launch-gate fund vs the $0 rule (new)

The first insured live job needs cash first: LLC reinstatement (~$82), name filings ($22 each), a company insurance binder, and background checks. Recommendation: a small capped, itemized fund, booked as a shared company cost or your contribution, repaid first from job profit. Pros carry their own insurance. We never take an uninsured job to keep the rule. If you do not answer: software, testing and recruiting continue at $0, and the live launch waits.

Call 10: Reinstate the LLC (~$82)

Unblocks name filings, the bank account, live Stripe, Google verification and insurance.

Calls 8 and 9: New names

Roomwise Hauling (junk), Pinecrew (pro network), Rakewell Yard Co. (yard). All checked clear. Blocks domains, filings, the Google profile and the hauler survey messages.

Name the operations lead

You or someone you name holds day-to-day authority for incidents and pros from the first live job, until revenue pays for the first hire.

Calls 1, 2 and 4: Margin floor, positioning, junk prices

$25 or 15% floor; insured local brand just under the franchises; $189 / $269 / $329 / $449 by crew size.

Calls 6 and 7: Money rules and pro requirements

No deposit, 48-hour hold, charge on "Looks good", $39 and $75 fees. Pros carry their own liability, auto, and workers' comp or the waiver.

Calls 3 and 5: Pilot scope and the shared offer

Junk only this fall, yard in spring, no snow. One-time job plus a repeat plan every 4 or 12 weeks for every brand.

Calls 11 and 12: Hosting and company insurance

Oracle Pay As You Go (still $0). Insurance (about $138 or more a month) decided with two broker quotes by Nov 9.

Design review: Grab the Gig redesign

Built and tested on a branch; merges after your look.

The playbooks we now own

This research was built once and is kept for every future venture: FreeGuy spawns, The Scout Report, client sites, Volt and the rest.

Copy and conversion playbook

From cases C14 to C17 plus client-site lessons.

A 10-minute start: brief, customer's own words, 12 headlines, score, red team, 5-second test.
A headline scoring sheet (clarity, specificity, credibility, difference).
Sign-up and form design: short first step, buttons where people decide, sticky phone bar.
Languages: native rewriting, review gates, legal and money lines held to a stricter standard.
Voice guides for our workers and homeowners in English and Spanish, and a swipe file of rival headlines.
9 house rules, including: never invent proof, no A/B tests without about 400 visitors per version.

Venture operating playbook

From cases C18, C19 and the 5 round-2 scale reports.

A 15-minute start: pick the model, copy the 14-department map, run the prevention checklist, set the calendar.
How each model wins or dies: lead gen, managed, bidding, employees, franchise, house of brands.
Prevention rules from DoorDash, Uber, Instacart, Handy, Homejoy, Angi and franchises, with sources.
The AI-run accounting department, ready to copy.
Legal traps: classification, tips, hidden fees, earnings claims, reviews, texts, door-to-door sales.
What changes at $1M, $10M and $50M.

How they compound

They load themselves. Both are Claude skills that switch on whenever copy or a new venture comes up, in any project.
Each venture starts where this one ended. The next business skips a day of research and starts from cited answers.
Each venture adds to them. Every launch, stage change or test result is appended to the playbook's change log, so the 10th venture runs on 9 ventures of lessons.
The method is part of it: research as cases, a red team before deciding, dated tests that can flip the call.