Skip to content

AI Startup Opportunities: Market Validation & Unit Economics

Analysis Date: March 9, 2026 Methodology: Bottom-up market sizing with publicly sourced data. Base cases are pessimistic; upside cases assume favorable execution.

Table of Contents

  1. Privacy-First Edge Camera Analytics
  2. AI RFP/Proposal Response Engine
  3. AI CRE Deal Screening & Underwriting
  4. AI Visual Inspection for Food/Pharma Manufacturing
  5. AI Healthcare Voice Agent (Prior Auth)
  6. Comparative Summary & Ranking

1. Privacy-First Edge Camera Analytics

Product: Jetson box ($500-$1,200) + $80/camera/month SaaS

1.1 TAM / SAM / SOM

TAM (Total Addressable Market): The global video analytics market was valued at ~$12.4B in 2025, growing at 22% CAGR. The on-premise/server segment held 67% share ($8.3B). The U.S. represents ~35% of this market = ~$2.9B for on-premise video analytics in the U.S.

SAM (Serviceable Addressable Market): Facilities with regulatory or policy mandates for edge-only processing:

Segment # Facilities Cameras/Facility (est.) Monthly SaaS Potential
Hospitals 6,093 50-200 $4,000-$16,000
Law firms (mid-size+) ~25,000 (of 463,600 total, ~5% need edge) 10-30 $800-$2,400
Cannabis facilities ~37,800 licenses (incl. ~7,500 dispensaries) 15-40 $1,200-$3,200
Government buildings ~9,000 federal buildings (GSA managed) 30-100 $2,400-$8,000

SAM = facilities genuinely needing edge-only (HIPAA, FedRAMP, state cannabis regs, attorney-client privilege): - Hospitals requiring on-prem: ~3,000 (50% of 6,093, many use hybrid cloud) - Cannabis: ~5,000 cultivation/processing facilities (state regs often mandate on-site video retention) - Government: ~3,000 sensitive facilities (classified, law enforcement) - Law firms: ~2,000 (large firms with strict data governance) - Total SAM facilities: ~13,000 - Average 30 cameras x $80/mo = $2,400/mo per facility - SAM = 13,000 x $2,400 x 12 = ~$374M/year

SOM (Serviceable Obtainable Market) -- Year 3: Realistic penetration of 1-2% of SAM = 130-260 customers - SOM = 130 x $2,400 x 12 = ~$3.7M ARR (base case) - SOM = 260 x $2,400 x 12 = ~$7.5M ARR (upside)

1.2 Unit Economics

Metric Value Notes
ARPC (Annual Revenue Per Customer) $28,800 30 cameras x $80/mo x 12
Hardware revenue $800 (net of $500-1,200 COGS ~$300-700) One-time, low margin
Hardware COGS $350-$700 per box Jetson Orin NX module $199-$399 + enclosure + assembly
SaaS Gross Margin 75-82% Cloud infra minimal (edge processing), support costs
Blended Gross Margin (Year 1) ~70% Hardware drag in Y1, improves to ~78% at renewal
CAC (estimated) $8,000-$15,000 Vertical SaaS enterprise sale, compliance-driven
LTV (5-year, 10% annual churn) $28,800 x 4.1 x 0.78 = ~$92,000
LTV:CAC 6:1 to 11:1 Strong
Payback Period 3-6 months High ARPC drives fast payback

1.3 Path to $1M ARR

  • Customers needed: 35 (at $28,800 ARPC)
  • Timeline: 12-18 months
  • Strategy: Focus on cannabis facilities first (clear regulatory mandate, concentrated buyer, lower sales complexity than hospitals). 5,000+ facilities with mandatory video surveillance and retention requirements.
  • Bottleneck: Hardware deployment logistics, on-site installation

1.4 Path to $10M ARR

  • Customers needed: ~350
  • Timeline: 24-36 months from launch
  • Scaling assumptions: Expand from cannabis to hospitals (longer sales cycle but 3-5x ARPC). Add channel partners (security integrators, VARs). Need 3-5 sales reps covering verticals. Assumes 15% logo churn offset by expansion (more cameras per site).

1.5 WTP Evidence

  • Verkada charges $199-$1,799/camera/year for cloud-based analytics ($17-$150/camera/month)
  • Rhombus and Spot AI use similar per-camera subscription models
  • $80/camera/month is competitive with Verkada and reasonable for compliance-mandated edge-only processing
  • Edge-only premium is justifiable: customers in regulated industries are already paying similar amounts but getting cloud solutions that may not meet their compliance requirements

1.6 Key Assumption That Could Kill This

Regulatory risk cuts both ways. If HIPAA, state cannabis, or FedRAMP rules are relaxed or if major cloud providers achieve sufficient compliance certifications for these verticals, the "must be on-premise" mandate weakens. Also, Verkada and Rhombus could add edge-only modes. The moat is narrow -- it is a deployment architecture choice, not a fundamentally different product.

1.7 Bottom-Up Math

Year Customers ARPC ARR
Y1 35 $28,800 $1.0M
Y2 120 $30,000 $3.6M
Y3 280 $32,000 $9.0M
Y4 400 $35,000 $14.0M

2. AI RFP/Proposal Response Engine

Product: $299-$499/month SaaS for SMBs, higher for enterprise

2.1 TAM / SAM / SOM

TAM (Total Addressable Market): The AI RFP response automation market was $1.1B in 2025, projected to reach $2.43B by 2029 (21.7% CAGR). The broader RFP software market is ~$4.5B.

SAM (Serviceable Addressable Market): Target: SMBs responding to 5-50 RFPs/year, especially government contractors.

Segment # Companies Notes
SAM.gov registered entities 674,000 Not all actively bid; many are grant recipients
Actively bidding federal contractors (est.) ~100,000-150,000 Subset that regularly responds to opportunities
State/local government vendors ~200,000-300,000 Across 50 states, thousands of municipalities
Private-sector RFP responders (IT, consulting, construction) ~150,000-250,000 Companies regularly responding to corporate RFPs

Companies responding to 5-50 RFPs/year (SMB sweet spot): estimated ~80,000-120,000 firms.

At $299-$499/month ($3,588-$5,988/year): - SAM = 100,000 x $4,800 avg = ~$480M/year

SOM (Year 3): - Base case: 0.5% of SAM = 500 customers = $2.4M ARR - Upside: 1.5% = 1,500 customers = $7.2M ARR

2.2 Unit Economics

Metric Value Notes
ARPC $4,800/year Average of $400/mo
SaaS Gross Margin 78-85% LLM API costs are primary COGS (~$5-15/RFP response)
CAC (SMB) $1,200-$2,500 Content marketing + free trials; GovCon community is tight-knit
CAC (Enterprise) $8,000-$15,000 Longer sales cycle, demos, pilots
LTV (4-year, 15% annual churn) $4,800 x 3.1 x 0.80 = ~$11,900 SMB churn is higher
LTV:CAC (SMB) 5:1 to 10:1 Healthy
Payback Period 3-6 months (SMB)

LLM Cost Sensitivity: At 20 RFPs/month x 50 questions/RFP x $0.01-0.05/question = $10-$50/customer/month in API costs. This is manageable at $400/mo price point (2.5-12.5% of revenue).

2.3 Path to $1M ARR

  • Customers needed: ~210 (at $4,800 ARPC)
  • Timeline: 12-15 months
  • Strategy: Target government contractors first. The GovCon community is tight-knit with clear watering holes (GovCon conferences, SAM.gov adjacent communities, APMP chapters). Offer free tier for 1-2 RFPs/month to build word of mouth.
  • Competitive moat risk: Loopio and Responsive (RFPIO) charge $15,000-$150,000/year -- they serve enterprise. The SMB gap ($299-$499/mo) is underserved but AI-native startups (AutoRFP at $899/mo, Iris AI, Tenderbolt) are moving fast.

2.4 Path to $10M ARR

  • Customers needed: ~1,400 SMBs + ~100 enterprise ($2,000-$5,000/mo)
  • Timeline: 30-42 months
  • Scaling assumptions: Layer on enterprise tier at $2,000-$5,000/mo. Add win-rate analytics and CRM integration upsells. Build content library network effects. Need 5-8 sales reps + robust content marketing engine.

2.5 WTP Evidence

  • AutoRFP.ai: $899-$1,299/month (mid-market)
  • Loopio: $25,000-$150,000/year (enterprise)
  • Responsive (RFPIO): $15,000-$25,000/year (enterprise)
  • Average SMB RFP team spends 20 hours/response at ~$50/hr loaded cost = $1,000/response
  • At 10 RFPs/month, that is $10,000/month in labor -- $400/month software is a no-brainer if it cuts time by even 30%

2.6 Key Assumption That Could Kill This

Commoditization risk is severe. RFP response is a natural LLM application. Every major proposal software vendor (Loopio, Responsive, Qvidian) is adding AI features. GPT wrappers with a content library are not defensible. Differentiation must come from deep vertical specialization (e.g., government FAR/DFAR compliance, specific proposal templates) or proprietary win-rate data. Also, the $299-$499 price point may be too low to support the CAC needed to reach small government contractors.

2.7 Bottom-Up Math

Year SMB Customers Enterprise Customers Blended ARPC ARR
Y1 210 5 $5,200 $1.1M
Y2 650 30 $5,800 $4.5M
Y3 1,400 100 $6,500 $11.6M

3. AI CRE Deal Screening & Underwriting

Product: $500-$2,000/month SaaS for syndicators and small PE firms

3.1 TAM / SAM / SOM

TAM (Total Addressable Market): The U.S. CRE technology market is part of the broader PropTech sector (~$18B globally). CRE analytics and underwriting tools specifically represent ~$2-3B.

SAM (Serviceable Addressable Market):

Segment # Firms (est.) Notes
Real estate PE firms (all sizes) ~2,500-3,500 Based on ~11,000 total PE/hedge fund firms, ~25-30% focused on real estate
Real estate syndicators (active, doing Reg D offerings) ~3,000-5,000 Based on SEC Form D filings; market consolidated post-2022 rate hikes
Small CRE investment firms / family offices ~5,000-8,000 Includes regional operators, family offices with CRE allocations
CRE brokerages doing investment sales ~2,000-3,000 Mid-market brokerages, not CBRE/JLL/Cushman tier

Target: mid-market syndicators and small PE firms doing $5M-$50M deals. - Estimated active target firms: ~5,000-8,000 - At $500-$2,000/mo ($6,000-$24,000/year), average $12,000/year: - SAM = 6,500 x $12,000 = ~$78M/year

SOM (Year 3): - Base case: 2% = 130 customers = $1.6M ARR - Upside: 5% = 325 customers = $3.9M ARR

3.2 Unit Economics

Metric Value Notes
ARPC $12,000/year Average of $1,000/mo
SaaS Gross Margin 80-85% Data feeds (CoStar, Reonomy) are significant COGS
CAC $3,000-$6,000 CRE community is tight-knit; conferences, podcasts, LinkedIn
LTV (5-year, 12% annual churn) $12,000 x 3.6 x 0.82 = ~$35,400
LTV:CAC 6:1 to 12:1 Strong
Payback Period 3-6 months

Critical COGS consideration: If the product requires CoStar or similar data feeds, those licenses can run $10,000-$50,000/year per seat, which would crush margins. The product must either use free/low-cost data sources or negotiate bulk API access.

3.3 Path to $1M ARR

  • Customers needed: ~84 (at $12,000 ARPC)
  • Timeline: 12-18 months
  • Strategy: Target multifamily syndicators first (largest, most active sub-segment). Content marketing via CRE podcasts (Best Ever CRE, etc.), conference sponsorships. Founder-led sales leveraging CRE network.
  • Key advantage: ARGUS Enterprise is the incumbent at ~$1,500+/year but is complex, outdated UX. Mid-market firms often use Excel. The "better than Excel, simpler than ARGUS" positioning is strong.

3.4 Path to $10M ARR

  • Customers needed: ~600-700 at blended $14,000-$16,000 ARPC
  • Timeline: 30-42 months
  • Scaling assumptions: Add deal marketplace / matching features. Expand to lenders (banks underwriting CRE loans). Layer on portfolio monitoring SaaS. Need 5-8 sales reps plus partnerships with CRE associations.

3.5 WTP Evidence

  • ARGUS Enterprise: ~$1,500+/year (industry standard, complex)
  • Archer.re: modern alternative, pricing not public but estimated $500-$1,500/mo
  • The CRE Suite: cloud-based, estimated similar range
  • Analyst salary for underwriting: $70,000-$120,000/year
  • A tool that saves 10 hours/week of analyst time = $15,000-$25,000/year in labor savings
  • $1,000/mo is very justifiable against hiring an additional analyst

3.6 Key Assumption That Could Kill This

Market size may be too small. The number of mid-market syndicators actively doing deals has shrunk significantly since 2022 rate hikes. Many syndicators have paused acquisitions or gone out of business. If the addressable market is really 3,000-4,000 firms (not 6,500+), the ceiling is ~$48M SAM, making $10M ARR achievable but leaving limited room for a venture-scale outcome. Also, CRE is deeply cyclical -- a prolonged downturn freezes deal flow and reduces WTP.

3.7 Bottom-Up Math

Year Customers ARPC ARR
Y1 84 $12,000 $1.0M
Y2 230 $13,000 $3.0M
Y3 500 $14,500 $7.3M
Y4 750 $16,000 $12.0M

4. AI Visual Inspection for Food/Pharma Manufacturing

Product: $4,000 hardware + $1,500/month SaaS per production line

4.1 TAM / SAM / SOM

TAM (Total Addressable Market): The AI visual inspection system market is growing at 19-25% CAGR, projected to reach $74.6B globally by 2029. The U.S. food/pharma subset is estimated at $2.7B in 2024, growing to $13.7B by 2029.

SAM (Serviceable Addressable Market): Target: food/pharma manufacturers with 50-500 employees.

Segment # Facilities Lines/Facility (est.) Notes
Food manufacturing (NAICS 311, 50-500 employees) ~6,000-8,000 3-8 Of ~29,500 total food mfg establishments, ~25% have 50-500 employees
Pharma manufacturing (50-500 employees) ~1,500-2,500 2-5 Of ~4,700 FDA-registered drug mfg plants, subset in target size
Total target facilities ~7,500-10,500 3-6 avg

Per production line: $4,000 hardware + $1,500/mo SaaS = $22,000/year/line Average 4 lines per facility: $88,000/year per facility

  • SAM = 9,000 x 4 lines x $22,000 = ~$792M/year

But realistic penetration (many lines don't need AI inspection): - Assume 50% of facilities have suitable lines = 4,500 facilities - Adjusted SAM = 4,500 x 3 lines x $22,000 = ~$297M/year

SOM (Year 3): - Base case: 1% = 45 facilities x 3 lines = 135 lines = $3.0M ARR (SaaS only, excluding hardware) - Upside: 2.5% = 112 facilities x 3 lines = 336 lines = $6.0M ARR

4.2 Unit Economics

Metric Value Notes
ARPC (per facility, 3 lines) $54,000/year SaaS + $12,000 hardware Hardware is one-time
Hardware COGS ~$2,500-$3,000 per unit Cameras, compute, enclosure, mounting
Hardware Gross Margin 25-33% Low, but one-time
SaaS Gross Margin 70-78% Higher COGS than pure SaaS due to model training, edge compute management
CAC $15,000-$30,000 Enterprise manufacturing sale; long cycle (3-9 months)
LTV (5-year, 8% annual churn) $54,000 x 4.0 x 0.74 = ~$160,000 Manufacturing customers are sticky
LTV:CAC 5:1 to 11:1 Good for hardware+SaaS
Payback Period 3-7 months High ARPC

Customer ROI: - Manual inspection: 2 inspectors x $50,000/year = $100,000/year labor per line - AI system: $22,000/year per line - Net savings: $78,000/year per line (78% cost reduction) - Recall avoidance: A single food recall costs $10M+ on average; even preventing one in 5 years is massive ROI - Payback for the customer: 3-4 months

4.3 Path to $1M ARR

  • Customers needed: ~19 facilities (at $54,000 ARPC, SaaS only)
  • Lines deployed: ~57
  • Timeline: 15-24 months (long sales cycle in manufacturing)
  • Strategy: Start with food manufacturing (less regulatory friction than pharma). Target bakeries, snack manufacturers, meat processors -- high defect rates, labor-intensive inspection. Pilot with 1 line, expand to full facility.
  • Bottleneck: On-site deployment, model training per product type, 3-9 month sales cycles

4.4 Path to $10M ARR

  • Customers needed: ~185 facilities at $54,000 ARPC
  • Lines deployed: ~555
  • Timeline: 36-48 months
  • Scaling assumptions: Build library of pre-trained models for common food/pharma products. Add pharma vertical (higher WTP, stricter requirements). Channel through food safety consultants and equipment distributors. Need field engineers for deployment (capital-intensive scaling). May need $5-10M to fund hardware inventory and field team.

4.5 WTP Evidence

  • Entry-level vision inspection systems: $15,000-$40,000 one-time (no AI, fixed rules)
  • Mid-range AI inspection: $40,000-$100,000 per line
  • Cognex, Keyence, SICK: major incumbents at $50,000-$250,000+ per line
  • $4,000 hardware + $1,500/mo is dramatically cheaper than incumbents
  • But incumbents have decades of reliability data and integration expertise
  • Customer comparison: $22,000/year vs. $50,000-$100,000 upfront = lower barrier to entry

4.6 Key Assumption That Could Kill This

The $4,000 hardware price point may be unrealistically low for food/pharma-grade equipment. Food manufacturing environments require IP67/IP69K ratings, washdown capability, stainless steel enclosures, and FDA-compliant materials. A ruggedized Jetson-based system meeting these requirements could easily cost $8,000-$15,000 per line, squeezing hardware margins to zero or negative. Additionally, each food product type requires custom model training -- the unit economics of model development per customer could be prohibitive until the product library is deep. Sales cycles of 6-12 months in manufacturing are capital-intensive.

4.7 Bottom-Up Math

Year Facilities Lines Hardware Rev SaaS ARR
Y1 19 57 $228K $1.0M
Y2 60 200 $572K $3.6M
Y3 130 430 $920K $7.7M
Y4 220 730 $1.2M $13.1M

5. AI Healthcare Voice Agent (Prior Auth)

Product: $500-$1,500/month for small practices (1-5 providers)

5.1 TAM / SAM / SOM

TAM (Total Addressable Market): The AI voice agents in healthcare market was $468M in 2024, projected to reach $3.18B by 2030 (37.8% CAGR). The broader healthcare administrative automation market addresses $450B in annual administrative waste.

SAM (Serviceable Addressable Market): Target: independent medical and dental practices (1-5 providers).

Segment # Practices Notes
Medical practices (independent, <10 physicians) ~120,000-160,000 ~395,000 total physician group practices; ~42% of physicians in private practice; ~73% are small practices
Dental practices (independent) ~108,000-115,000 ~135,000 total dental establishments; ~84% owner-operated
Total target practices ~228,000-275,000

At $500-$1,500/month ($6,000-$18,000/year), average $10,000/year: - SAM = 250,000 x $10,000 = ~$2.5B/year

But not all practices handle enough prior auths to justify the cost. Practices averaging 20+ prior auths/week are the sweet spot: - Adjusted SAM = ~80,000 practices x $10,000 = ~$800M/year

SOM (Year 3): - Base case: 0.5% = 400 practices = $4.0M ARR - Upside: 1.5% = 1,200 practices = $12.0M ARR

5.2 Unit Economics

Metric Value Notes
ARPC $10,000/year Average of ~$833/mo
SaaS Gross Margin 65-75% Voice AI API costs, telephony costs ($0.05-$0.15/min), LLM inference
COGS per call $0.50-$2.00 10-20 min call x $0.05-$0.10/min telephony + LLM costs
CAC (SMB practice) $1,500-$3,000 Content marketing, medical association partnerships, referrals
LTV (4-year, 18% annual churn) $10,000 x 2.8 x 0.70 = ~$19,600 Small practices churn higher
LTV:CAC 7:1 to 13:1 Very strong
Payback Period 2-4 months

Customer ROI: - Prior auth staff cost: $35,000-$50,000/year per FTE (typically 0.5-2 FTEs per practice dedicated to PA) - Staff time on PA: 853 hours/year per physician = ~$83,000/year in labor - Manual PA cost: $11-$100 per authorization (avg ~$35) - Volume: 39-45 prior auths/week per physician = ~2,000/year - Annual PA labor cost per physician: ~$70,000-$83,000 - AI system cost: $10,000/year - Net savings: $60,000-$73,000/year per physician (85%+ reduction) - Payback for customer: < 1 month

5.3 Path to $1M ARR

  • Customers needed: 100 practices (at $10,000 ARPC)
  • Timeline: 9-14 months
  • Strategy: Target specialty practices first (orthopedics, cardiology, oncology) -- highest PA volume and pain. Integrate with top 5 EHRs used by small practices. Offer free 2-week pilot with guaranteed call recording for quality proof.
  • Key advantage: Prior auth is universally hated by physicians. 47% of physicians rank administrative automation as top investment priority. The value proposition is visceral.

5.4 Path to $10M ARR

  • Customers needed: ~1,000 practices at $10,000 ARPC
  • Timeline: 18-28 months from launch
  • Scaling assumptions: Expand from prior auth to benefits verification, claims follow-up, appointment scheduling (increases ARPC to $15,000-$20,000). Add dental practices. Build integrations with 20+ EHRs. Need 10-15 sales reps. Potential channel through medical billing companies and practice management consultants.

5.5 WTP Evidence

  • Prosper AI: Y Combinator-backed, raised $5M; charges per-call or subscription; claims 50% cost reduction
  • Infinitus: 5M+ calls automated, serves large health systems (not small practices)
  • Outsourced prior auth services: $15-$35 per authorization
  • Dedicated PA staff salary: $35,000-$50,000/year
  • $500-$1,500/month is 70-85% cheaper than a dedicated PA employee
  • 82% automation rate demonstrated by Orbit Healthcare

5.6 Key Assumption That Could Kill This

Regulatory change could eliminate the problem. CMS finalized rules requiring electronic prior authorization by 2027, and several states are passing PA reform legislation. If payers are forced to automate PA on their end, the provider-side automation market shrinks dramatically. Additionally, large EHR vendors (Epic, athenahealth) are building native PA automation -- if they bundle it free, small practices won't pay for standalone tools. The competitive landscape includes well-funded startups (Infinitus, Prosper, Cohere Health, AKASA) targeting the same space, though mostly focused on large health systems, not small practices.

Voice AI reliability risk: Healthcare calls require near-perfect accuracy. A single incorrect authorization could result in a denied claim worth thousands. Error rates and hallucination risk must be extremely low, requiring significant investment in quality assurance and monitoring.

5.7 Bottom-Up Math

Year Practices ARPC ARR
Y1 100 $10,000 $1.0M
Y2 400 $11,000 $4.4M
Y3 1,000 $12,500 $12.5M
Y4 2,000 $14,000 $28.0M

Comparative Summary & Ranking

Side-by-Side Comparison

Metric 1. Edge Camera 2. AI RFP 3. CRE Underwriting 4. Visual Inspection 5. Voice Agent (PA)
SAM $374M $480M $78M $297M $800M
ARPC $28,800 $4,800 $12,000 $54,000 $10,000
Gross Margin 70-78% 78-85% 80-85% 70-78% 65-75%
LTV:CAC 6-11x 5-10x 6-12x 5-11x 7-13x
Payback 3-6 mo 3-6 mo 3-6 mo 3-7 mo 2-4 mo
Customers to $1M 35 210 84 19 100
Time to $1M 12-18 mo 12-15 mo 12-18 mo 15-24 mo 9-14 mo
Customers to $10M 350 1,500 700 185 1,000
Time to $10M 24-36 mo 30-42 mo 30-42 mo 36-48 mo 18-28 mo
Defensibility Low-Med Low Medium High Medium
Capital Intensity Medium Low Low High Low
Execution Risk Medium Low Medium High Medium

Ranking by Path to $10M ARR (Risk-Adjusted)

1. AI Healthcare Voice Agent (Prior Auth) -- BEST OPPORTUNITY - Fastest path to $1M and $10M ARR - Largest adjusted SAM ($800M) - Strongest customer ROI (85% cost reduction vs. hiring PA staff) - Lowest capital intensity (pure software) - Best LTV:CAC ratio - Risk: Regulatory change (CMS electronic PA mandate by 2027), EHR vendor bundling - Verdict: Best risk-reward ratio. Clear pain point, massive market, fast payback.

2. AI RFP/Proposal Response Engine -- STRONG BUT CROWDED - Lowest CAC and capital requirements - Large SAM ($480M) - But: severe commoditization risk from AI-native competitors and incumbent feature additions - Needs deep vertical specialization to win (e.g., GovCon-only) - Verdict: Good market, but the window for differentiation is closing rapidly.

3. Privacy-First Edge Camera Analytics -- SOLID NICHE - High ARPC ($28,800) means fewer customers to reach milestones - Clear regulatory-driven demand in cannabis, healthcare, government - But: narrow moat (deployment architecture, not product), incumbents can add edge modes - Medium capital intensity (hardware inventory) - Verdict: Strong unit economics in a niche. May cap out as a $20-50M ARR business.

4. AI CRE Deal Screening & Underwriting -- SMALL MARKET, CYCLICAL - Best gross margins (80-85%) - Strong domain expertise moat - But: smallest SAM ($78M), highly cyclical (rate-dependent deal flow) - Market may be too small for venture-scale outcome - Verdict: Great lifestyle business or acquisition target, questionable venture fit.

5. AI Visual Inspection for Food/Pharma -- HIGH CEILING, HIGH RISK - Highest ARPC ($54,000) and strongest customer ROI - Highest defensibility (custom models per product type create switching costs) - But: highest capital intensity, longest sales cycles, hardware complexity in harsh environments - $4,000 hardware price may be unrealistic for food-grade equipment - Verdict: Biggest potential outcome but requires most capital and longest timeline. Better suited for teams with manufacturing domain expertise and $5M+ to deploy.

Key Data Sources