Blog

LevelBench-100 Research (v3.0)

An unassisted AI model is a probabilistic reader, not an accounting team. When we tested 27 raw AI models on 100 messy SMB QuickBooks takeover checks, accuracy topped out at 71% and every single model posted bad book-entries when records conflicted or check images were missing.

The between-meetings test

The value of a fractional CFO is not what gets said in the meeting. It is what changes before the next one.

The honest CFO readiness test

A CFO can help you make a hard decision. They cannot make your inputs trustworthy or carry the decision through the business by giving you another slide deck.

Benchmarks

Everyone assumes an older building costs more to service. Our older comparison (original version and sample not confirmed) measures billed revenue: in office buildings, routine service revenue was nearly flat with age, while replacement revenue rose about 1.85x. It does not measure service cost. Age earns a multiplier in only 3 of 12 segments and inverts in retail. The pooled first pass across building types hid these segment differences.

Published multiples from named M&A advisors, plus our own operating data

Two hypothetical HVAC companies at $5M EBITDA and assumed 7x versus 11x multiples imply $35M versus $55M enterprise value. Recurring work is one diligence question, alongside scale, mix, management depth and deal structure.

From Level's proprietary contractor research

Your field software says the job made 38%. Your P&L says 24%. Is it timing, a different cost definition, or missing records? Reconcile the same jobs before choosing which number to trust.

Benchmarks

A national review-count benchmark can hide geographic differences. National and state review-count percentiles can differ materially. Compare the same trade and geography before treating a national rank as local market position; the directory does not establish what causes the difference.

From Level's proprietary contractor research

The median recorded collection ratio is 85.1% across 464 contractors. It is cash collected divided by billed revenue in an unpublished comparison window, not a permanent loss, a receivable balance or the gap between profit and bank cash.

Benchmarks

Dollars-per-square-foot figures quoted for commercial mechanical service span about 14x, from $0.15 to $2.15. That is not a range. That is what a number looks like when the data behind it is rarely shown.

Benchmarks

Level's working range, not a measured industry percentile, is 45-55% of revenue. The useful answer is where YOUR number sits by account, because in cleaning one underpriced contract can drag the whole book, and the benchmark taped to the wall will not tell you which one.

Construction Accounting

Ask the vendor to show the report, its source records and the handoff to your books. A feature label does not establish that the workflow fits your jobs, payroll requirements or current edition.

Business Growth

Two hypothetical electrical contractors at the same EBITDA and assumed 6x versus 11x multiples differ by 5 turns of EBITDA. Specialty mix is one diligence question, alongside management depth, cash conversion, buyer fit and deal structure. This hypothetical comparison does not establish a premium for an actual contractor.

Business Growth

Two HVAC shops with similar revenue can get very different offers. Deal size often matters. Within a size band, the levers you control are how much of your revenue is recurring service agreements and whether a buyer's accountant can trust your numbers without a long cleanup.

From the founding team's contractor analysis

Most contractors track revenue and jobs booked. Neither tells you if you are making money. These are the 8 numbers that do, in the priority order we use when judging profit leverage.

Illustrative job-costing walk-through

The margin on your bid sheet is not the margin you earn. The gap is labor burden you didn't fully load, callbacks nobody charged to the job, material that drifted after you quoted, and receivables you're financing for free.

From the founding team's operator analysis

A route estimate is only comparable with actual margin when both include the same labor burden, drive time, equipment cost, and seasonal capacity.

Illustrative leakage waterfall

A full schedule is not the same as a collected schedule. No-shows leave fixed-cost slots empty, denials force rework and partial write-offs, and undercoding gives away revenue you already earned. A full-slot forecast can show 20% while a fictional realized scenario is 10%; that is not a measured typical practice outcome.