WalkBid

1. One-liner

Walk the building filming on your phone; get a measured, fixture-counted, ISSA-loaded janitorial bid back in minutes.

2. Trend signal — why now?

Three things landed at once.

First, the measurement half went free. Apple’s RoomPlan framework and video-based 3D scanning now hit ½-inch to 2cm accuracy on any iPhone Pro (12 Pro through 17 Pro), and a wave of 2025–2026 apps (MakePlan, Manifold, RoomScan Pro) are built straight on it. Measuring a building’s cleanable square footage — the single hardest, slowest manual step in a janitorial bid — no longer needs a wheel, a tape, or a blueprint you can’t trust.

Second, the classification half went cheap. 2026 multimodal LLMs can take the same walkthrough footage and do what a wheel can’t: tell an office floor from a restroom from a break-room-with-a-kitchen, count toilets and urinals and sinks, flag hard-floor vs carpet, and spot the soil-load tells (high-traffic lobby, medical waiting room) that change the production rate. That’s the part owners currently eyeball and get wrong.

Third, money is pouring into vertical service AI. Avoca raised $125M at a $1B valuation in April 2026 answering missed calls for trades. QuoteIQ, Hosta.ai, and MeasureSquare are funded around photo→estimate — but every one of them is aimed at residential / one-time jobs (lawns, pressure-washing, house cleaning). The recurring commercial janitorial contract — the 3–5× bigger, stickier revenue — is still being bid in Excel and 2010-era calculators.

The pain is loud and specific. Industry guidance is blunt: “You cannot price a building you haven’t seen. Photos lie. Floor plans miss details. The 8,000 sq ft office with ‘light cleaning needs’ turns out to have 14 restrooms and a break room with a full kitchen.” And underbidding isn’t a rounding error — “The price is too low, and you bleed money within 90 days,” “the lowest bidder often quits within 6 months because they’re losing money on every visit.”

Provenance:

• Signal 1 (Demand): 1.25M US janitorial firms, $112B market; “Photos lie… turns out to have 14 restrooms”; underbidding “bleed money within 90 days”; incumbents (CleanGuru $59–135/mo, free calculators) all require manual area/fixture entry — https://fieldcamp.ai/blog/commercial-cleaning-pricing-guide/ , https://www.ibisworld.com/united-states/number-of-businesses/janitorial-services/1496/ — 2026-05
• Signal 2 (Feasibility): Apple RoomPlan + video 3D scan now ½-inch–2cm accurate on any iPhone Pro (free framework); multimodal LLMs classify zones + count fixtures from same footage — https://www.scanmanifold.com/blog-posts/floor-plan-app-iphone-guide , https://volpis.com/blog/apple-roomplan-overview/ — 2026
• Signal 3 (Economic): Avoca $125M at $1B for vertical trade AI; QuoteIQ/Hosta/MeasureSquare funded for photo→estimate but aimed at residential, not commercial ISSA recurring contracts — https://fortune.com/2026/04/27/avoca-ai-agents-missed-calls-hvac-plumbing-roofing-kleiner-perkins-chen-shrivastava-braswell/ , https://myquoteiq.com/best-ai-tools-for-home-service-businesses/ — 2026-04 Category: Tech-unlock

3. The opportunity

The incumbent isn’t really CleanGuru — it’s the Excel sheet and the 1–2 hour manual walkthrough. CleanGuru, JanitorialManager, BidLoom, the dozen free “janitorial bid calculators” — they all assume you already walked the building with a clipboard and know the cleanable square footage, the restroom count, the floor types, the soil load. They turn those numbers into a price. They do not get you those numbers. That data-collection step is exactly where owners burn time and where they guess wrong and underbid.

WalkBid attacks the data-collection step. You film the building once on your phone. WalkBid measures the zones, classifies them, counts the fixtures, applies the right ISSA production rate per zone, loads labor at burden (the 50–70% of cost owners forget), adds supplies/overhead/margin, and hands back a defensible bid and a branded PDF proposal — in minutes, not the 5–7 days that “80% of the competition” takes.

The 10× isn’t “AI on top of a calculator.” It’s collapsing the part of the workflow that the calculators never touched. A focused team can own the commercial-recurring niche before the residential-estimate players bother to retool for ISSA zone math, frequency schedules, and multi-year contracts.

4. Target market

• Primary customer: Owner-operators of small commercial cleaning / janitorial companies in the US, 1–25 employees, roughly $200K–$3M annual revenue, who bid recurring office / medical / retail / school / gym contracts. The owner usually does the walkthroughs and bids personally.
• Why they buy: “Every bid is a gamble. I walk the building, guess the hours, and I don’t know if I made money until 60 days in.” They’ve seen the lowest bidder win and quit; they’ve eaten a money-losing contract themselves. A bid that’s measured instead of guessed, and out the door in 48 hours, wins more work and protects margin.
• Rough TAM reasoning: 1.25M US janitorial firms (IBISWorld, 2026), heavily small. Even if only the ~15–20% that actively chase commercial recurring contracts and are smartphone-comfortable are reachable, that’s 180K–250K firms. Capture a few thousand at $99–199/mo and you’re at $5M ARR.
• Why now for them: They already own an iPhone. The scan tech just became accurate and free, and AI can finally read the building. Margins are thin and getting thinner — the cost of one underbid contract dwarfs a year of subscription.

5. Product sketch (MVP)

• Phone walkthrough capture — open the app, walk the building filming; on-device LiDAR/video scan measures each room’s area as you go.
• AI zone classification — auto-tags each space (open office, private office, restroom, break room, lobby, corridor, hard-floor, carpet, medical) so the right production rate applies per zone, not one blended guess.
• Fixture & feature count — counts toilets, urinals, sinks, kitchens, stairwells, entry mats from the footage; flags high-soil zones that slow production rates.
• ISSA-loaded bid engine — applies per-zone production rates, frequency schedule (e.g. 5×/wk vs 3×/wk), fully-burdened labor, supplies, overhead, and target margin to produce labor hours and a price.
• Branded proposal PDF — generates a clean, send-ready proposal within minutes of the walk; owner edits assumptions and ships it same-day.
• Bid-vs-actual feedback loop — after the contract starts, log actual hours; WalkBid flags where the estimate drifted so the next bid is sharper (and the data compounds into the moat).
• Assumptions are always visible and editable — every number (rate, frequency, margin) is shown and overridable. No black box; the owner stays in control of the price.

6. AI angle — what’s load-bearing

Two AI jobs, both load-bearing. Remove either and the product collapses back into the existing manual-entry calculators that already exist.

1. Spatial measurement from a phone walkthrough — turning raw video/LiDAR into per-room cleanable square footage. This is what kills the 1–2 hour clipboard walk.
2. Scene understanding — classifying each zone and counting fixtures/features from the same footage. A tape measure gives you area; it cannot tell you “this is a 12-restroom medical suite with heavy soil load,” which is precisely what determines the production rate and where owners underbid (restrooms alone add 3–5 hours/week to a 12-restroom building).

The bid math itself (production rates × area × frequency + burden) is deterministic and public — that’s deliberately not the AI. The AI’s whole job is to populate the inputs that humans currently collect by hand and get wrong.

7. Localization angle (if any)

N/A — this is a US-first play. The wedge is the English-language ISSA production-rate standard, US labor-burden math, and the smartphone-LiDAR install base, all of which are strongest in the US commercial-cleaning market. A UK/AU/Canada expansion is straightforward later (same standards, same devices), but there’s no India/LatAm pricing or payment-rail angle that improves the wedge — forcing one would weaken it.

8. Business model — path to $1M–$5M ARR

• Pricing: $99/mo Solo (1 user, unlimited bids), $199/mo Team (up to 5 estimators + bid-vs-actual analytics + custom proposal branding). Benchmarked just above CleanGuru’s $59–135 because WalkBid removes the labor step CleanGuru leaves to you. Optional $5–10 per finished walkthrough-bid for very low-volume occasional bidders.
• ACV: ~$1,500 (blended toward Solo early, Team later).
• Rough math to $1M ARR: ~650 customers × ~$130/mo × 12 ≈ $1.0M. Out of 180K+ reachable firms, that’s a ~0.4% capture.
• Rough math to $5M ARR: ~2,800 customers blended at ~$150/mo, OR ~2,000 customers + a per-bid usage layer + a route into adjacent recurring-bid trades (landscaping, window/pressure washing, facility maintenance) that share the “walk it, price it” shape.
• Expansion path: Solo → Team seats as the company grows estimators; bid-vs-actual analytics as a retention/upsell hook; later, a “win-rate” pricing-optimization tier and adjacent-trade modules.

9. Go-to-market wedge — first 100 customers

• Facebook/owner groups, directly. There are large, active commercial-cleaning owner communities (e.g. “Cleaning Business Owners,” methodcleanbiz/Jan-Pro/BSCAI orbit, multiple 10K+ member groups). Post a 60-second screen recording: phone walk of a real office → finished bid. The hook writes itself (“stop guessing your bids”). Offer the first 50 a free measured bid on their next real prospect.
• Bid-loss interception via free tool. Ship a free “Is your bid leaving money on the table?” calculator that takes their last bid’s price + square footage and shows the ISSA-implied hours gap. It’s the lead magnet and a live demo of the bid engine; the upsell is “do this from a walkthrough automatically.”
• The training/coach channel. A handful of well-followed janitorial-bidding educators (TheJanitorialStore, MethodCleanBiz, CMM) sell bidding courses to exactly this audience. Affiliate/co-sell: WalkBid is the tool that does what they teach. One coach endorsement = dozens of logos.
• Cold outreach to recent bidders. Scrape public RFP/bid-board postings for janitorial contracts (municipal, school district, property-management portals); the cleaning companies responding are bidding right now. Send a personalized 90-second walkthrough demo of a similar building type.

10. Build complexity — justification

Medium. The measurement layer is off-the-shelf (Apple RoomPlan / a video-scan SDK), the zone-classification and fixture-counting layer is off-the-shelf multimodal LLM + light fine-tuning on cleaning-specific scenes, and the bid engine is deterministic public-formula code. The real work is the iOS capture UX (must be dead-simple for a non-technical owner walking a building) and getting zone classification + fixture counting accurate enough on messy real footage to be trusted. A small team ships a credible v1 in ~3–4 months; the accuracy tuning is the long pole, not the architecture.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	No regulated data; recording your own client’s building during a sales walk is standard practice.
Ethical — no harm / dark patterns	✅	Helps owners not underbid; assumptions transparent and editable.
Market exists (evidence above)	✅	1.25M firms, $112B market, named incumbents charging money, loud bidding pain.
1–5 person team can build this	✅	iOS app + off-the-shelf AI + deterministic bid engine; 3–4 mo to v1.
Launchable with <$50K / ₹40L	✅	Solo/pair build; inference + scan SDK costs are modest at launch volume.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Underbidding is a money-bleed felt on every bid; but it’s “guess and find out in 60 days,” not literally hair-on-fire daily. Strong, not maximal.
Demand evidence	15	12/15	1.25M firms, paying incumbents, explicit “photos lie / 14 restrooms” pain, funded adjacent players. Docked because I couldn’t pull verbatim owner quotes (Reddit unfetchable).
Build feasibility	15	11/15	Measurement + classification are off-the-shelf; the long pole is fixture/zone accuracy on real messy footage and a simple capture UX.
Distribution clarity	15	12/15	Named owner groups, named coaches, public bid boards, free-tool lead magnet. Conversion math is reasonable but unproven.
Revenue mechanics	15	12/15	Pricing benchmarked to CleanGuru with a clear “we do the step you don’t” justification; ACV and customer counts realistic.
Time to first revenue	10	7/10	3–4 mo to v1, then self-serve trial-to-paid; first dollars plausible within weeks of launch but not pre-sellable in week one.
Defensibility	10	4/10	Bid-vs-actual data compounds and the commercial-ISSA focus is a real wedge, but a funded residential player (QuoteIQ et al.) could retool. Execution + niche focus, not a hard moat.
Total	100	74/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy (iOS capture + computer-vision accuracy tuning is the core risk) · domain-expertise-required (must encode ISSA production rates, frequency schedules, and labor-burden math correctly, or the bids are wrong and trust evaporates on day one).

Key assumptions to validate (3–5)

1. Assumption: Owners trust an AI-measured/classified bid enough to send it to a client. How to test: Run 20 supervised real walkthroughs; compare WalkBid’s area/fixture/hours output to the owner’s own manual numbers. Need agreement within ~10% and owner saying “I’d send this.”
2. Assumption: Fixture counting + zone classification is accurate enough on real, cluttered, occupied buildings (not staged demos). How to test: Ground-truth 30 diverse buildings (office, medical, school, retail) and measure classification precision/recall and fixture-count error.
3. Assumption: Owners will pay ~$99–199/mo, above CleanGuru. How to test: 30 owner interviews + a pre-sale landing page; need >20% of demo-watchers to start a trial.
4. Assumption: The walkthrough UX is simple enough for a non-technical owner to capture a usable scan first try. How to test: Hand the app to 10 owners cold, no instructions; measure first-scan success rate.

Risk flags

1. Platform dependency: Heavy reliance on Apple RoomPlan / iOS-Pro LiDAR. Android and pre-Pro iPhones are weaker; mitigate with a video-photogrammetry fallback, but capture quality varies.
2. Accuracy-trust cliff: One badly-measured bid that loses an owner a contract destroys trust permanently. The product must fail safe (flag low-confidence scans, force human review) rather than confidently output a wrong price.
3. Competitive timing: A funded residential photo→estimate player (QuoteIQ, Hosta) could add commercial ISSA logic. The defense is depth in commercial-recurring workflow (frequencies, burden, bid-vs-actual) and owning the niche channel first.
4. Market timing (mild): Owners are not all smartphone-confident; the laggard segment may resist. Targets the modern, growth-minded operator first.

14. Structured verdict

Score:                  74/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder (iOS + computer vision) paired with a commercial-cleaning domain advisor who has bid contracts
Time to revenue:        3–4 months to v1, first paying customers within weeks of launch
Capital to launch:      $15–30K ($ for inference, scan SDK, and a domain advisor's time)
Top 3 assumptions to validate first:
  1. AI area/fixture/hours output lands within ~10% of an owner's manual bid — 20 supervised real walkthroughs
  2. Owners will pay $99–199/mo above CleanGuru — 30 interviews + pre-sale page, >20% demo-to-trial
  3. Non-technical owners can capture a usable scan first try — 10 cold-start usability tests
Kill criteria:
  - Abandon if zone classification + fixture counting can't clear ~90% accuracy on real occupied buildings after tuning (bids become untrustworthy)
  - Abandon if <20% of demo-watchers start a trial across the first two channel tests
  - Abandon if a funded residential player ships commercial ISSA walkthrough-to-bid before v1 launch

15. Next step — 1-week validation sprint

• Day 1–2: Recruit 8–10 commercial-cleaning owners from the big Facebook owner groups. Get them on a call; have each describe their last 2 bids and what they guessed vs. what actually happened. Confirm the money-bleed pain is real and frequent for them, not just the blogs.
• Day 3–4: With 3 willing owners, walk a real prospect building together. Capture footage on a phone, and in parallel have the owner do their normal manual bid. Hand-process the footage (even manually + an off-the-shelf scan app) into area/fixture/zone numbers and compare to the owner’s.
• Day 5: Decide go / no-go on a hard number: did the phone-walk-derived bid land within ~10% of the owner’s manual bid on at least 2 of 3 buildings, and did ≥2 of 3 owners say “I’d send that bid to a client”? If yes, build. If the footage can’t reliably yield trustworthy area/fixture data, the whole thesis is dead — kill it before writing iOS code.