{
  "slug": "contractor",
  "label": "Contractor financial & operating benchmarks",
  "publisher": "Level",
  "publisher_url": "https://levelcfo.com",
  "dataset": "The Level Index, contractor financial & operating benchmarks",
  "contractor_count": 2242,
  "attribution": "The Level founding team's analysis of 2,200+ contractors representing $13.25B in job revenue and 2.5M invoices, built across operating, private-equity, and CFO roles (financial reviews, operator interviews, and PE due-diligence), spanning HVAC, plumbing, electrical, mechanical, refrigeration, and fire protection, and layered with named public sources.",
  "methodology": "Figures are aggregated and anonymized. Distributions are reported as percentiles (p10/p25/median/p75/p90). No individual company is identified or identifiable. Per-metric sample sizes and definitions are attached to each metric.",
  "segments": [
    "HVAC",
    "plumbing",
    "electrical",
    "mechanical",
    "refrigeration",
    "fire protection"
  ],
  "last_updated": "2026-10-07",
  "date_range": "2022/2026",
  "license": "CC BY 4.0. Cite Level and link to levelcfo.com.",
  "metrics": [
    {
      "key": "service_agreement_gross_margin",
      "pillar": "Earnings",
      "label": "Service agreement gross margin",
      "unit": "percent",
      "n": 259,
      "definition": "Gross margin on recurring service-agreement (maintenance) revenue. n=259 companies with at least $10K in SA revenue.",
      "p10": -30.4,
      "p25": 20.5,
      "median": 37.9,
      "p75": 53.5,
      "p90": 70.3,
      "top_quartile": 53.5,
      "note": "Below 25% calls for a cost and scope review. Pricing, covered work, cost allocation and customer mix are candidate explanations; the distribution does not identify a cause.",
      "basis": "contractors analyzed",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "service_agreement_pull_through",
      "pillar": "Accounts",
      "label": "Annual service agreement pull-through",
      "unit": "percent",
      "n": 386,
      "definition": "Additional repair, replacement, and project revenue from service-agreement customers divided by annual service-agreement revenue. n=386 eligible companies.",
      "p10": 0.3,
      "p25": 1.5,
      "median": 8.7,
      "p75": 29.6,
      "p90": 93.4,
      "note": "P90 includes relationship-driven project work attributed to service-agreement accounts, not only work discovered during maintenance visits. P75 is a distribution comparison point, not a proven attainable target. Company-level records are not published because the underlying roster is re-identifiable; this public file exposes only the aggregate distribution.",
      "basis": "386 companies with eligible annual agreement and follow-on revenue",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "service_agreement_revenue_per_employee",
      "pillar": "Earnings",
      "label": "Annual service agreement revenue per employee",
      "unit": "thousand_usd_per_employee",
      "n": 217,
      "definition": "Annual service-agreement revenue divided by employee count. This is a directional Level engagement observation across 217 eligible companies, not a universal staffing target.",
      "p10": 5.9,
      "p25": 11.4,
      "median": 28.5,
      "p75": 64.1,
      "p90": 150.5,
      "note": "Use the distribution as recurring-revenue intensity context. Trade mix, use of subcontractors, and which employees are included can materially change the ratio.",
      "basis": "217 companies with eligible annual service-agreement revenue and employee counts",
      "source": "The Level Index (directional contractor benchmark observation)"
    },
    {
      "key": "collection_rate",
      "pillar": "Cash",
      "label": "Collection rate",
      "unit": "percent",
      "n": 464,
      "definition": "Cash collected as a share of billed revenue.",
      "p10": 38.8,
      "p25": 70.7,
      "median": 85.1,
      "p75": 92.7,
      "p90": 96,
      "target_range": [
        92,
        96
      ],
      "target_range_basis": "Legacy comparison span retained for compatibility, not a universal target or exact P75-P90 endpoints. Collection ratio measurement window is not published.",
      "note": "A collection-rate difference describes billed revenue not converted to cash in the comparison. It is not automatically point-in-time A/R, recoverable cash, permanent loss or recurring profit. Reconcile invoice aging, credits and cutoff timing.",
      "basis": "464 contractors analyzed",
      "source": "The Level Index (contractor benchmark research)",
      "external_corroboration": "Related but differently defined external references: CFMA 2024 Construction Financial Benchmarker (56.6 days in accounts receivable, n=1,290) and Levelset 2022 Construction Cash Flow Report (fewer than 40% of contractors paid within 30 days). Neither validates the collection-ratio denominator or window."
    },
    {
      "key": "billing_speed_days",
      "pillar": "Cash",
      "label": "Billing speed (days from work complete to invoice)",
      "unit": "days",
      "n": 733,
      "definition": "Days from job completion to invoice issued. Median is 1 day when progress billing is included (~25% of companies); among post-completion invoicers the adjusted median is 7 days.",
      "median": 1,
      "median_post_completion": 7,
      "bottom_decile": 30,
      "lower_is_better": true,
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "billing_capture",
      "pillar": "Cash",
      "label": "Billing capture rate (hours invoiced / hours logged on jobs)",
      "unit": "percent",
      "n": 963,
      "definition": "Total hours invoiced to customers divided by total hours LOGGED ON JOBS. The denominator is job-assigned hours only, which is why the median is high. This is NOT technician utilization: Level's working range for utilization (65-80%) divides billable hours by TOTAL PAID hours, including drive time, admin, training, and idle time. Some recorded ratios exceed 100%; their causes are not measured here. Check billed-hour units, period alignment, job matching and logging completeness. A higher dollar rate alone does not increase an hours ratio. n=963 companies with at least 100 logged hours.",
      "p10": 66.9,
      "p25": 89.2,
      "median": 97.1,
      "p75": 100,
      "p90": 102,
      "note": "State the denominator whenever this is cited. Without it the figure reads implausible, because readers compare it to utilization, which measures something else. First derived 2026-04-07 as 'billable hour ratio' (median 96.7%, n=654) and re-run to 97.1% on the larger n=963 cohort.",
      "basis": "companies with 100+ logged hours",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "average_bill_rate",
      "pillar": "Labor",
      "label": "Average billed labor rate by company",
      "unit": "usd_per_hour",
      "n": 1770,
      "definition": "Company-level average billed labor rate across 1,770 contractors. The underlying rate-card field is used inconsistently, with some companies entering customer billing rates and others entering loaded cost or base wage, so this distribution describes mixed source fields, not independently comparable customer selling prices, wages or profitability.",
      "p10": 39,
      "p25": 55,
      "median": 79,
      "p75": 116,
      "p90": 148,
      "note": "These are labor-component rates per technician hour, not the total customer invoice after trip charges, materials, equipment, and overhead markup.",
      "basis": "1,770 companies with a calculated average billed labor rate",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "quote_conversion_decided",
      "pillar": "Accounts",
      "label": "Quote conversion rate (on decided quotes)",
      "unit": "percent",
      "n": 794,
      "definition": "Won divided by (won + lost); excludes still-pending quotes. n=794 companies with at least 20 decided quotes. Distinct from all-quote conversion below.",
      "p10": 47.1,
      "p25": 61,
      "median": 73.9,
      "p75": 83.2,
      "p90": 90,
      "note": "Won divided by won plus lost for eligible decided quotes. Pending, draft and cancelled records are excluded. This distribution does not measure estimating effort or prove a conversion lift from a follow-up cadence.",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "quote_conversion_all",
      "pillar": "Accounts",
      "label": "Quote conversion rate (all quotes issued)",
      "unit": "percent",
      "n": 794,
      "definition": "Won divided by all quotes issued, including quotes that never reach a decision.",
      "p25": 28,
      "median": 38.1,
      "p75": 47.7,
      "note": "Won divided by all issued quotes includes pending opportunities and differs from decided conversion. Quote counts do not measure estimating hours, quote value or permanent lost revenue.",
      "basis": "companies with 20+ quotes",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "days_to_convert",
      "pillar": "Accounts",
      "label": "Days to convert (winning quotes)",
      "unit": "days",
      "n": 794,
      "definition": "Days from quote sent to accepted, for quotes that convert.",
      "median": 2,
      "lower_is_better": true,
      "note": "Recorded winning quotes have shorter decision/status intervals in this aggregate. Predictive strength and causal effects were not tested.",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "days_to_loss",
      "pillar": "Accounts",
      "label": "Days to loss (lost quotes)",
      "unit": "days",
      "n": 794,
      "definition": "Days from quote sent to marked lost, for quotes that do not convert.",
      "median": 29,
      "p90": 165,
      "note": "The median recorded days-to-loss is 29; customer decision timing and status-update lag are not separated.",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "customer_concentration_top1",
      "pillar": "Risk",
      "label": "Revenue from largest single customer",
      "unit": "percent",
      "n": 959,
      "definition": "Share of total revenue from the top 1 customer. n=959 companies with at least $100K in annual revenue.",
      "p25": 18.5,
      "median": 31,
      "p75": 54.6,
      "p90": 82.4,
      "basis": "companies with >$100K revenue",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "job_gross_margin",
      "pillar": "Earnings",
      "label": "Job-level gross margin",
      "unit": "percent",
      "n": 1747089,
      "definition": "Gross margin per completed job (revenue minus actual cost), across all job types. Measured per JOB across 1,747,089 completed jobs from 1,791 companies (Athena verification 2026-07-25). The prior n=430 was another metric’s company count.",
      "p10": 10.8,
      "p25": 28.4,
      "median": 44.3,
      "p75": 60.2,
      "p90": 75.1,
      "pct_negative_margin": 6.2,
      "pct_under_20pct_margin": 16,
      "basis": "completed jobs analyzed",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "labor_hours_actual_vs_budget",
      "pillar": "Labor",
      "label": "Labor hours: actual as % of budgeted",
      "unit": "percent",
      "n": 315393,
      "definition": "Actual labor hours ÷ budgeted labor hours per job. 100% = on budget; >100% = over. Measured across 315,393 jobs from 1,391 companies (Athena verification 2026-07-25).",
      "p25": 70.1,
      "median": 99.4,
      "mean": 119,
      "p75": 131.1,
      "pct_over_budget": 40,
      "pct_over_150pct": 18.3,
      "note": "Actual labor hours divided by budgeted hours. The median and mean differ because the distribution has a long tail. This is an hours ratio, not dollar cost; actual wage, overtime, burden, scope and posting completeness determine contribution effects.",
      "basis": "jobs with both budgeted and actual hours",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "revenue_mix_labor_share",
      "pillar": "Labor",
      "label": "Labor share of quoted revenue",
      "unit": "percent",
      "n": null,
      "sample_size_status": "rederivation_required",
      "definition": "Labor as a share of quoted revenue (vs materials, equipment, subs). Median contractor is ~29% labor / ~33% materials, but variance is extreme. NOTE: sample size unverified. The prior n=430 was copied from another metric; this is measured on quoted revenue (quote-line data), not job_stats, so a job-level reproduction does not apply. Do not publish a sample size until re-derived.",
      "p25": 16.7,
      "median": 29.1,
      "p75": 42.2,
      "note": "Revenue-mix distributions describe reported labor, materials, membership and equipment shares. They do not establish a margin, capacity or scalability effect.",
      "basis": "quote line items (sample size unverified)",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "job_closeout_days",
      "pillar": "Cash",
      "label": "Job closeout lag (days)",
      "unit": "days",
      "n": 555,
      "definition": "Days from work complete to recorded financial job closeout. This status interval does not establish invoice or cash-collection timing.",
      "median": 1.7,
      "p75": 9.6,
      "p90": 35.8,
      "lower_is_better": true,
      "basis": "completed jobs",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "seasonality_top3_months",
      "pillar": "Risk",
      "label": "Revenue concentration in top 3 months",
      "unit": "percent",
      "n": 217,
      "definition": "Share of annual revenue earned in the three strongest months.",
      "p25": 38.3,
      "median": 47.4,
      "p75": 66.1,
      "p90": 91.1,
      "note": "The median recorded revenue share in the three largest months is about 47%. Reported peak-month counts and cutoff windows require separate interpretation; no universal quarter pattern is established.",
      "basis": "companies with >$500K revenue",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "quote_line_margin_by_cost_type",
      "pillar": "Earnings",
      "label": "Quote-line gross margin by cost type",
      "unit": "percent",
      "n": 2200000,
      "definition": "Gross margin on quoted work, split by cost type (labor, materials, equipment, subcontractor), measured across 2.2M+ quote line items. Labor is the highest-margin cost type; subcontractor pass-through is the thinnest.",
      "by_cost_type": {
        "labor": 47.7,
        "materials": 31.5,
        "equipment": 25.5,
        "subcontractor": 23.6
      },
      "note": "Recorded quote-line margin medians differ by cost type: labor 47.7%, materials 31.5%, equipment 25.5%, and subcontracted work 23.6%. These are quoted line-item observations, not realized job margins. They do not establish that changing service mix causes higher company profit; match selling prices, actual costs and the full job mix before drawing that conclusion.",
      "basis": "2.2M+ quote line items",
      "source": "The Level Index (contractor benchmark research)"
    },
    {
      "key": "bill_rate_by_role",
      "pillar": "Labor",
      "label": "Bill rate by role (rate-card field)",
      "unit": "usd_per_hour",
      "n": 18000,
      "definition": "Median rate-card rate by field role across 18,000+ tracked employees. The rate-card field is used inconsistently (some enter customer billing rates, others loaded cost or base wage), so neither the absolute level nor the spread establishes role-level pricing, wages or profitability without matching how each entry was used. Company-level avg bill rate median is $79/hr (n=1,770).",
      "by_role": {
        "helper": 34,
        "apprentice": 39,
        "technician": 54,
        "journeyman": 76,
        "mechanic": 75
      },
      "note": "The largest single recorded step is technician to journeyman (about $54 to $76 in the mixed rate-card field). Wage, loaded cost and selling-price definitions are mixed; no spread over wage can be computed from this field.",
      "basis": "18,000+ field employees with tracked rate-card data",
      "source": "The Level Index (contractor benchmark research)"
    }
  ],
  "revision_notes": {
    "note": "Forward-looking notes on how each figure below is derived and when its derivation was last sharpened. Every current value lives in metrics[] in this same file, which is the only authority for what a Level number is. This block never restates a superseded figure.",
    "as_of": "2026-10-07",
    "notes": [
      {
        "id": "collection_rate",
        "label": "Collection rate",
        "updated": "2026-10-07",
        "note": "Cash collected divided by billed revenue per company across 464 eligible contractors. Compare matching billing and payment windows. The ratio is not a point-in-time A/R balance or permanent-loss rate. Aging, retainage, credits and cutoff require a separate bridge.",
        "metric_key": "collection_rate",
        "in_this_dataset": true
      },
      {
        "id": "billing_capture",
        "label": "Billing capture",
        "updated": "2026-10-07",
        "note": "Billing capture compares hours invoiced with hours logged on jobs, across 963 companies with at least 100 logged hours. It is distinct from utilization, which uses total paid hours. Ratios above 100% require checking units, period alignment, job matching and logging completeness; causes are not measured here. Dollar billing premiums alone do not increase an hours ratio.",
        "metric_key": "billing_capture",
        "in_this_dataset": true
      },
      {
        "id": "labor_hours_actual_vs_budget",
        "label": "Labor hours against budget",
        "updated": "2026-07-26",
        "note": "Labor variance is measured per job, actual hours divided by budgeted hours, across 315,393 jobs from 1,391 companies, verified against the job level data on 2026-07-25. Jobs carrying no hour budget are excluded rather than assumed on target. The median and the mean are reported separately because they say different things: the median job lands on budget while a long tail runs well past it.",
        "metric_key": "labor_hours_actual_vs_budget",
        "in_this_dataset": true
      },
      {
        "id": "job_size_mix",
        "label": "Job size mix",
        "updated": "2026-10-07",
        "note": "Job size is grouped by logged hours. Published shares describe job count; logged hours are not elapsed project duration. A revenue-share comparison requires a reconciled eligible denominator and complete direct cost. The displayed job-count buckets are not an exhaustive partition."
      },
      {
        "id": "labor_churn",
        "label": "Workforce churn",
        "updated": "2026-10-07",
        "note": "Industry separation events, occupation tenure and a company staffing cohort measure different populations. The occupation chart is a fictional constant-hazard sensitivity using historical tenure inputs, not observed turnover or a proven lower bound. Use matched local staffing records for vacancy and replacement-cost decisions."
      },
      {
        "id": "technician_turnover",
        "label": "Annual technician turnover",
        "updated": "2026-07-26",
        "note": "Level does not publish a technician turnover rate. The underlying records carry employees but no termination dates, so no turnover rate can be computed from them, and an earlier figure attributed to Level could not be traced to a measurement. For labor stability, read the labor hours actual-versus-budget and labor churn measurements published here instead, or the Bureau of Labor Statistics JOLTS series for construction separations.",
        "metric_key": "technician_turnover",
        "in_this_dataset": false
      },
      {
        "id": "revenue_concentration_jobs",
        "label": "Share of revenue from the largest jobs",
        "updated": "2026-07-26",
        "note": "Level does not publish a revenue-share-by-largest-jobs figure. The earlier claim rested on a single job-size tier that read far out of line with the tiers around it on revenue per logged labor hour, so it was withdrawn rather than restated. For concentration risk, read the revenue share from the largest customer, which is measured per company across 959 contractors and is published here.",
        "metric_key": "revenue_concentration_jobs",
        "in_this_dataset": false
      },
      {
        "id": "labor_over_estimation",
        "label": "Labor over-estimation on bids",
        "updated": "2026-07-26",
        "note": "Level does not publish a labor over-estimation rate. The earlier claim came from a company-level COST variance in dollars that was relabelled as an HOURS variance without being re-derived, and it contradicts the labor hours measurement published here, which compares actual to budgeted hours per job. Read that measurement instead.",
        "metric_key": "labor_over_estimation",
        "in_this_dataset": false
      },
      {
        "id": "bill_rate_by_role",
        "label": "Role rate fields",
        "updated": "2026-10-07",
        "note": "Role-level rate-card entries mix selling prices, loaded costs and wages. Neither their levels nor spreads establish pricing or profit without matching the field definition in the underlying records.",
        "metric_key": "bill_rate_by_role",
        "in_this_dataset": true
      },
      {
        "id": "quote_line_margin_by_cost_type",
        "label": "Quoted line margins",
        "updated": "2026-10-07",
        "note": "Cost-type comparisons describe quoted line-item margins, not realized job margins or a causal effect of service mix. Match selling prices, complete actual costs and the whole job before changing the mix.",
        "metric_key": "quote_line_margin_by_cost_type",
        "in_this_dataset": true
      },
      {
        "id": "levelbench_ai_accounting_accuracy",
        "label": "LevelBench-100 AI accounting close accuracy",
        "updated": "2026-11-12",
        "note": "LevelBench v5.0 evaluates 58 live provider-executed AI model and playbook test runs across 25 foundation, fast-worker, and open-weight models on 60 anonymized SMB client closes (122 tasks, 924 checks) and a 374-company close vault (752 tasks, 5,923 checks), alongside 40 downloadable SMB close case packages (84 tasks, 588 checks) and 36 AI accounting error patterns."
      }
    ]
  }
}
