A financial model can be perfectly coherent and still describe a company that does not exist: 150% growth with a gross margin of 90%, marketing at 3% of revenue and a CAC payback of four months, every year, for five years. Every formula resolves. The plan is still not credible.
Benchmarks are how you catch that. Not as targets, but as a reality check: does this plan look like any company we know of, and if not, why not? This article covers the five ratios worth computing, the ranges to hold them against, and the four mistakes that make the comparison meaningless. All the figures, with their samples and dates, are on the SaaS spending benchmarks page.
Before comparing: four mistakes that void the check
1. Mixing denominators. Spending benchmarks are published as a percent of ARR, a percent of GAAP revenue, or a share of operating expenses, depending on the source. "About 25% of opex on sales and marketing at $0 to $1M of ARR" (Scale Venture Partners) and "8% of ARR on marketing" (SaaS Capital) measure different things. Compute your ratio with the same denominator as the benchmark you compare it with, or do not compare.
2. Summing medians. The median company on marketing is not the median company on R&D. In SaaS Capital's 2026 survey, the nine department medians add up to 86% of ARR, while the median total spend is 96% for bootstrapped companies and 101% for equity-backed ones. Compare each line on its own, and the total with the total.
3. Choosing the wrong peer group. Bootstrapped and equity-backed companies are different populations. At the median, equity-backed private SaaS companies spend 10% of ARR on marketing against 5% for bootstrapped ones, 17% on sales against 10%, 25% on R&D against 16%. A bootstrapped plan judged against venture medians looks underinvested; a venture plan judged against bootstrapped medians looks reckless. Pick the group that matches how the plan is funded, and write it down.
4. Using one band for every year. A five-year plan crosses several ARR bands. A growth rate that is ordinary under $1M of ARR is exceptional at $20M. Compare year 1 with the band the company is in during year 1, and year 4 with the band it will be in then.
The five ratios
1. Growth rate, by ARR band
Compute year-over-year ARR growth for each year of the plan, and note which ARR band the company sits in that year.
Ranges from High Alpha's 2025 SaaS Benchmarks (800+ companies, 50th and 75th percentiles):
| ARR band | Median | 75th percentile |
|---|---|---|
| Under $1M | 100% | 300% |
| $1M to $5M | 50% | 100% |
| $5M to $20M | 30% | 70% |
| $20M to $50M | 30% | 40% |
The pattern that should raise a question is not a high growth rate in year 1. It is a growth rate that stays in the top quartile for five years while the company moves through three bands. For bootstrapped companies at $3M to $20M of ARR, SaaS Capital puts median growth at 15% and the 90th percentile at 42.3%.
2. Gross margin
Revenue minus cost of revenue (hosting, third-party software and AI costs embedded in the product, support, professional services), as a share of revenue.
High Alpha's medians run from 74% under $1M of ARR to around 80% at $5M to $20M, with the 75th percentile between 80% and 86%. Two things to check in the model. First, that the cost of revenue exists and grows with customers: a model with no hosting line, or a flat one, will show a margin that improves for free. Second, if the product calls AI models, that their cost is in there: High Alpha reports early-stage margins down nearly 10 points in a year and attributes it, as a likely cause, to AI costs.
3. Total spend and its split, as a share of ARR
Total operating and delivery spend divided by ARR, then each department on its own. SaaS Capital's 2026 medians, as a percent of ARR:
| Department | Bootstrapped | Equity-backed |
|---|---|---|
| Sales | 10% | 17% |
| Marketing | 5% | 10% |
| R&D | 16% | 25% |
| G&A | 11% | 18% |
| Customer success | 5% | 10% |
| Total spend | 96% | 101% |
Read the total first. A plan that reaches profitability in year 2 with venture-level spending on every line has a contradiction somewhere. Then read the lines one by one, against the group that matches the plan.
4. CAC payback
The number of months of gross profit it takes to recover the cost of acquiring the new revenue:
CAC payback (months) = sales and marketing spend of the prior period ÷ (new ARR in the period × gross margin) × 12
Two details change the result: use gross margin, not raw ARR, and lag the spend (the sales and marketing that closed this quarter's deals was mostly spent before). The 2025 median was 16 months across the B2B SaaS companies in the Aleph and Benchmarkit 2026 benchmarks, with large differences by segment: 11 months below $5k of annual contract value, 22 months at $50k to $100k.
The check: compute it from the model's own lines. If the plan shows a payback well under the median for its deal size, the model probably has either too little sales and marketing spend or too much new ARR per euro spent. One of the two is the assumption to revisit.
5. Net revenue retention
Revenue at the end of a year from the customers who were there at the start, divided by their revenue at the start. Above 100% means expansion outweighs churn.
High Alpha's medians sit between 100% and 104% depending on the band; SaaS Capital puts bootstrapped companies at $3M to $20M at 103% (91% gross retention). A plan with 130% NRR from year 1 is betting on expansion from a customer base that is still small. It may be right; it should say why.
Reading the gaps
A ratio far from the median is not a mistake. It is one of two things:
- A reason. The plan has a structural difference from the median company: a product-led motion with very low CAC, an enterprise motion with long payback, a regulated market that needs more R&D. Write the reason in the model, next to the assumption that produces the gap. Scale Venture Partners gives two examples from its own portfolio: a healthcare company spending more than 40% of opex on R&D because of regulatory approvals, and one building its own foundation model spending more than half.
- A wrong assumption. If there is no reason, the gap points to the input to revisit. Low CAC payback points to the funnel. High gross margin points to a missing cost line. High growth for five years points to the new-customer assumption.
Either way, the outcome of the check is a list of written answers, not a model edited until every ratio sits on the median. A plan identical to the median company is not more credible; it is less informative.
Running the check with an AI assistant
The check is mechanical once the model is structured, which makes it a good job for Claude or another assistant, with two conditions: it computes from the model rather than estimating, and it reports rather than edits.
A prompt that works:
Compute these five ratios for each year of the model: ARR growth, gross margin, total spend as a share of ARR (and each department: sales, marketing, R&D, G&A, customer success), CAC payback (prior-year S&M ÷ (new ARR × gross margin) × 12) and net revenue retention. For each year, give the ARR band. Compare each ratio with the benchmarks below, using the [bootstrapped / equity-backed] column. List every ratio more than 50% away from the median, with the model lines that produce it. Do not change any assumption.
[paste the tables from the benchmark page, with their source and year]
The last sentence matters. An assistant asked to "make the plan more realistic" will move numbers towards the median, and a plan quietly rewritten to match a survey is worse than the original: it has lost the reasons that made it yours.
Where Layerz sits in this
Layerz does not hold benchmarks, and it does not judge whether a plan is realistic. What it checks is coherence: every formula resolves, every line is linked to what it depends on, and nothing is a hidden hardcoded value that breaks when an assumption changes. That is the precondition for this check, because a ratio computed on a model with silent hardcodes measures nothing.
Layerz is a structured spreadsheet built for AI. Claude reads the model over MCP, computes the ratios from the actual lines, and can trace each gap back to the assumption that produces it. The reason for a deliberate gap goes in that assumption's note, so it is still there the next time someone, or the next session, runs the check. The model exports to Excel with live formulas when an investor wants to recompute it themselves.
Further reading: SaaS Spending Benchmarks (2026) · Financial Projections Before Revenue · The Verification Tax · How to Stop AI From Hardcoding Values in a Financial Model
Layerz keeps a financial model as structure separate from data. Claude writes the formulas, the engine calculates, and every number traces back to its formula. Excel export is clean, standard, and never paywalled. Explore Layerz →