The two axes
Every skilled trade is AI-resilient in absolute terms. That is the thesis: AI cannot swing the wrench. So the two axes measure it directly, then the quadrant chart plots each trade’s relative position among its peers so an owner can see where their trade actually stands.
AI-Resilience
How protected the trade’s core value is from AI/automation.
| Sub-score | Weight | What it measures |
|---|---|---|
| Physical/on-site core | 35% | How much of the paid work requires hands-on, on-site skilled labor an AI agent cannot perform. |
| Disintermediation resistance | 25% | How hard it is for an AI agent or platform to bypass the contractor and go direct. |
| AI-native entry barrier | 20% | Licensing, trust, relationships, and physical presence that block AI-native new entrants. |
| Customer lock-in | 20% | Service agreements, recurring maintenance, and relationships that retain customers. |
AI-Leverage
How much unrealized profit AI can unlock by orchestrating the work around the labor.
| Sub-score | Weight | What it measures |
|---|---|---|
| Coordination/back-office drag | 35% | Share of cost/time in coordination, estimating, scheduling, billing, the work AI can orchestrate. |
| Operating-leverage step-change | 25% | Margin recoverable from mis-costed jobs, rework, billing lag, and cash trapped in open work. |
| Revenue expansion | 20% | Upside from better bidding, pull-through, and new service lines AI can unlock. |
| Domain-data amplification | 20% | How much the trade's proprietary job data compounds in value with AI (Level's core thesis). |
On the weights (and why the ranking is robust)
The sub-score weights (0.35 / 0.25 / 0.20 / 0.20) are a deliberate, transparent judgment: the first sub-score of each axis is the load-bearing one (can AI do the physical work; how much of the profit is trapped in coordination), so it carries the most weight, and the rest taper. They are round on purpose, because false precision would be its own dishonesty. The important claim is not the exact decimal , it is that the ordering is stable: shifting any weight by ±0.10 does not change which quadrant a trade lands in, and barely moves the ranks. Read the composite as a tier and a quadrant, not as a claim that a trade scoring 82.3 is meaningfully ahead of one at 81.7. The raw sub-scores and the full data are published so anyone can reweight and check.
The four quadrants
Plotting relative AI-Resilience (x) against relative AI-Leverage (y) splits the trades into four positions. Position, not color, carries the meaning. Because every trade is AI-resilient in absolute terms, treat the quadrant and tier as the signal, not small decimal gaps between neighbors.
Transform & Win
Protected from AI and full of upside. The best place to be. Press the advantage.
Safe Haven
Well protected, with upside already partly captured. Defend the moat.
Race to Modernize
Real AI upside, but more exposed. Speed of adoption decides who wins.
Under Pressure
Lower relative protection and slower to gain. The hardest corner.
Dimensions published alongside the AI scores
The AI-Resilience and AI-Leverage model stays a clean, defensible thesis. Four further dimensions are published alongside it, not folded into the scores, so each can be cited on its own.
| Dimension | What it is |
|---|---|
| ai scores | AI-Resilience x AI-Leverage, the core AI thesis (see axes). |
| reputation | Reputation & professionalization from public review listings, a market-quality dimension published alongside (NOT inside) the AI scores, so the AI model stays a clean, defensible thesis. |
| firmographics | Company size, revenue & age per trade from a national multi-source business firmographic dataset, a market-structure dimension, also published alongside the AI scores. |
| market structure | National permit/contractor-density aggregates + Level CFO operating benchmarks. |
| geography | State wage differentials + licensing/bonding regimes and metro-level contractor density / permit trend per trade (50 states + top-50 metros), with citations retained on published facts. |
| building stock | Commercial & industrial building-stock age and size by state. Age mix is descriptive; replacement implications depend on the trade and are not measured by this cross-sectional file. |
| validation | Public-equity comp returns as an external market-validation axis. |
Data provenance
Public sources
BLS/OEWS labor stats, IBISWorld/CFMA market and financial benchmarks, public county tax-assessor and building-permit records, public business registries, public business-review listings, and public-equity filings. Research observations identify their sources; aggregate tables disclose their table-level provenance.
Proprietary aggregates
Aggregate operating benchmarks provided by Level CFO (n disclosed). Aggregate-only; no individual company identified.
Compiled by Sam Yang: CFO across trade contractors, ex-contractor-software product director, Stanford MBA, ex-PE/IB.
On independence: this index is compiled by Sam Yang at Level CFO, which serves contractors in several of these trades. Level owns some of the aggregate operating data cited here and says so plainly. Provenance and a published methodology come before authority. Nothing here is dressed up as third-party research.
Attribution for reuse
Trade Economy Index original analysis and compilation are licensed under CC BY 4.0. When republishing them online, use: Source: TradesIndex by Level. Include a clickable link to https://tradesindex.org. Quoting specific wording is not required.
Building-stock provenance
The state building-stock dimension aggregates public state and county tax-assessor and property records. Level analyzed an aggregated compilation of those records and does not claim a direct download from every government website. The underlying stacked weekly snapshots were deduplicated to one record per parcel, then grouped by state, construction-era band, and commercial or industrial use.
A property here is an assessor parcel or property record, not necessarily one physical building. One parcel can contain multiple structures. Construction era is unknown for 39.4% of records, and those records are excluded from the pre-2000 percentage denominator. This is not the FEMA/ORNL structure-footprint dataset used by Level CFO to count serviceable buildings by state and metro.
Research passes & confidence
Numeric figures are reviewed in repeated research passes, then reconciled into one published value. Published fields retain citations, the research-pass count, and a confidence tier. A pass count measures repeated review runs. It does not establish that the underlying evidence sources are independent.
| Badge | What it means |
|---|---|
| N research passes · high confidence | Reviewed repeatedly with close agreement across the recorded passes. This is not an independence count. |
| verified | Reconciled by a second-stage adjudicator when initial passes diverged. |
| medium | Supported, but with some spread across sources. |
| limited support | The recorded evidence is limited and should be treated cautiously. |
How each trade was scored
The one-line basis behind each trade’s sub-scores. Full per-trade detail lives on each trade page.
| Trade | Index | Basis |
|---|---|---|
| HVAC & Refrigeration | 82.3 | On-site repair/install AI cannot perform; heavy service-agreement lock-in; strong recurring revenue. High coordination drag (dispatch, parts, warranty) = large AI upside. Level core vertical. |
| Plumbing | 76.5 | Highly physical, emergency-driven, licensed. Lower recurring lock-in than HVAC. Solid but less service-agreement upside. |
| Electrical | 80.2 | Licensed, physical, code-bound. Controls/low-voltage adds tech-enabled revenue expansion. Strong AI upside in estimating/design. |
| Roofing | 74.8 | Physical work AI cannot do, but low lock-in and easy new-entrant flow (storm chasers). Big operating-leverage upside: insurance/retail cash-cycle, closeout lag, mis-costed jobs. |
| Glass & Glazing | 76.8 | Custom fabrication + install, project-driven, PM-referral relationships. Job-costing complexity = strong AI-leverage. A vertical Level works in. |
| Doors & Access | 78.6 | Install + recurring service/maintenance (dock, access control) = good lock-in. Commercial payment-term cash lag = operating-leverage upside. A vertical Level works in. |
| Landscaping & Grounds | 74.9 | Physical, but highly fragmented and low entry barrier. Recurring maintenance contracts help lock-in. Crew-productivity + design-build job costing = large AI upside. A vertical Level works in. |
| Commercial Cleaning & Janitorial | 74 | Physical/fine-motor work robots struggle with, recurring contracts, but very low entry barrier and commodity pricing. Labor-heavy, thin margins = big operating-leverage + true-cost upside. A vertical Level works in. |
| Painting & Wall Finishing | 69.3 | Physical, but lowest entry barrier and lock-in of the set. Commercial vs residential margin gap + job costing = operating-leverage upside. A vertical Level works in. |
| Concrete & Masonry | 75.5 | Heavy physical, capital + skill barrier. Project-based, WIP/retainage complexity = strong operating-leverage upside. Low recurring lock-in. |
| Fire & Life Safety | 81.7 | Code-mandated recurring inspection = exceptional lock-in + entry barrier. Compliance recurring revenue. Strong AI upside in scheduling/compliance tracking. |
| Low-Voltage & Security | 76.3 | More tech-exposed than pure trades (some remote/monitoring), but recurring monitoring contracts = strong lock-in and revenue expansion. High data-amplification. |
| Restoration & Remediation | 76.8 | Emergency physical work, insurance-driven. Very high coordination drag (insurance docs, adjusters, closeout) + cash-cycle pain = large AI + operating-leverage upside. Low lock-in. |
Limitations
- Judgment weights. The sub-score weights are a transparent editorial judgment, not an empirically fitted model. We publish the raw sub-scores so you can reweight; treat tiers and quadrants as the signal, not decimal gaps between neighbors.
- Occupation mapping. Public labor codes (BLS SOC) are broader than these trade categories, so some employment counts are approximate and a few trades share an occupation code. Where two trades would show an identical inherited count, the count is shown only on its primary trade.
- Historical public-company context. Company prices, returns, margins, and valuation figures retain their stated July 23, 2026 snapshot and fiscal periods. They are excluded from current comparisons until independently dated, value-bound market evidence is available.
- Small comparison sets. Some trades have only a handful of public-company comparables, so the validation axis is directional, not definitive, for those trades.
- Evidence families are not yet classified. Research-pass counts describe repeated review runs and must not be interpreted as independent-source counts. Citations should be evaluated directly. Reputation and firmographic dimensions are only published where coverage is sufficient.
- Not third-party. Level CFO compiles this and advises contractors in several of these trades. Provenance and a published method come before authority; nothing here is dressed up as independent research.