The ARR that is not ARR

ARR is not an accounting standard. It is a management convention, and management sets the convention.

The inclusions we look for first. One-year contracts with no auto-renewal, counted at full annual value, where the customer decides again every 12 months. Implementation and onboarding fees, which are non-recurring by definition. Managed services and support retainers priced separately from the subscription. Training and professional services. Usage overages annualized from a strong month.

Every one is legitimate revenue. None of it belongs in the number a buyer is paying a multiple of. Which of them are present, and in what proportion, differs enough between companies that guessing is pointless.

The margin data is where the mix tends to surface first. The 2026 Aleph and Benchmarkit benchmarks put 2025 median software gross margin at 80% and median total gross margin at 76%, and attribute the 4-point gap to the dilutive effect of professional services and other non-recurring revenue. A company whose blended margin sits well below its stated software margin is telling you something about the mix, whether or not the ARR slide does.

Where the reconstruction actually happens

Not in the ARR schedule. In the billing system.

Invoice-level data separates recurring from non-recurring cleanly, because the billing engine had to make that distinction to raise the invoice. Contract terms tell you which revenue renews automatically and which requires a decision. Together they produce a recurring base that is often smaller than the headline, and, more usefully, a base whose cohorts can be tracked.

The second question follows immediately: who pays it. Consider a company with 40 logos where the top 3 customers fund the business and the other 37 are broadly cost-neutral to serve. The logo count says diversified. The cash flow says 3 relationships. Concentration on the revenue line and concentration in the cash flow are different measurements, and the second decides whether a bad quarter is survivable.

Why this is worth doing properly

Retention is the whole argument, and retention is only meaningful when it is calculated on a base that means something. SaaS Capital’s 2026 benchmarking, drawn from more than 1,000 private B2B SaaS companies, puts median net revenue retention for bootstrapped companies between 3 and 20 million dollars of ARR at 103%, with the 90th percentile at 117.9%, and describes median retention as essentially flat against the prior year.

Those figures are calculated on genuinely recurring revenue. Run the same calculation on a base that includes services and one-year cancellables and the result is not comparable to anything, including the same company a year earlier.

Which is the real risk. Not that the number is wrong, but that it is incomparable. A buyer who accepts the seller’s ARR definition is benchmarking against a peer set that used a different one, and every conclusion downstream inherits the error.

This is the workstream that most often changes a price, and it cannot be done without the seller’s data.