What Is a SaaS Company? The Test Is What Happens When You Stop Paying
Recurring revenue is the load-bearing test for what counts as a SaaS company — and the interesting part is where it stops being obvious. Nine companies, with dated figures, at the edges of the definition.
On this page
PAR Technology has been selling restaurant point-of-sale systems since 1968. GetLatka’s database has its revenue at $245M for December 2012 and $283M for December 2021, with $187M in December 2019 sitting in between. Odoo, over roughly the same stretch, went from $4.9M recorded December 2011 to $160M recorded 29 July 2021, and after 2012 never recorded a figure lower than the year before.
Both companies sell software to businesses. Only one of those revenue lines behaves the way people mean when they say “SaaS company,” and the difference is not in the product. It is in how the money arrives.
SaaS is a revenue shape, not a product category. A company is a SaaS company when its customers rent access to software the vendor operates, and pay again for as long as they keep the access. The browser, the cloud, the monthly invoice are symptoms of that arrangement rather than the definition of it.
The roster of household names that fit comfortably is a separate page. This one is about the companies sitting near the line, because that is where a definition either works or doesn’t.
The three clauses
- Rented, not owned — the customer buys a right to use, not a copy. Nothing survives the end of the relationship.
- Operated by the vendor — one codebase, on infrastructure the vendor controls, upgraded for everyone at once. The customer’s IT department is not in the loop.
- Paid again, for as long as access lasts — the invoice is a condition of the product working, not a fee for support and improvements to something already bought.
The third clause is load-bearing, and it comes with a test that settles most arguments in one sentence: what happens to the customer if they stop paying next month?
You sold a licence. A perpetual licence with an annual maintenance contract produces recurring revenue — genuinely recurring, sometimes for decades — but the renewal buys upgrades and support for an asset the customer already owns. Declining it is an inconvenience, not an outage.
You sold a subscription. The renewal is the product. Which is why churn governs a SaaS company and merely annoys a licence business: one is losing the customer, the other is losing the upsell.
Recurring revenue = the revenue that arrives next period without a new saleMaintenance fees clear that bar too. The distinction is what the customer loses by not paying.
Held to the test, PAR’s decade reads differently. A point-of-sale system that has already shipped owes nobody a renewal, so a revenue base assembled that way can go backwards without anything going wrong inside the company.
Between December 2012 and December 2019, PAR’s recorded revenue fell by $58M. Subscription revenue only does that when customers actually leave. Hardware and project revenue does it every time a buying cycle pauses, and no product description would have warned you which one you were reading.
The edge cases
The marketplace that takes a cut
Magic Eden’s revenue in the database is a single row: $96M, recorded 1 January 2022. It is a marketplace, earning a percentage of the trades it hosts. Software is what it runs on, but nothing is subscribed to — the revenue is a function of somebody else’s volume, and it recurs only as long as that behaviour does. A take rate is a different revenue model carrying a different risk. A subscription business loses money one cancellation at a time; a take-rate business can lose most of its revenue in a quarter without a single customer leaving.
The agency with a dashboard
Directive Consulting grew from $1M recorded October 2016 to $20M recorded 21 October 2021, on retainers that renew much like subscriptions do. The database has 135 people on that same October date, which works out at about $148,000 of revenue per employee. Inflectra, selling software tools, had $8.6M recorded May 2021 against 23 people recorded 21 November 2021 — roughly $374,000 each.
$374Krevenue per employee at Inflectra, on rows recorded May and November 2021, against $148K at Directive Consulting in October 2021
That ratio is the cheapest structural tell available, because it measures the thing the definition cares about: whether the next dollar of revenue needs the next person. Software copies itself; service delivery does not. An agency that builds its clients a dashboard has added software to a services business, and a software company that sends an implementation team has added services to a subscription. Both are ordinary, and the ratio usually says which one you are looking at.
Usually, not always — and the counter-example is the next boundary case.
Recurring, monthly, and not software at all
SafetyWing went from $1M recorded June 2019 to $12M recorded 21 November 2021, sold monthly, cancellable any time, bought entirely through a website. It had 24 employees on that November date, which is around $500,000 of revenue a head — a better ratio than Inflectra’s. It is an insurance company. The policy is the product and the website is the counter.
Wing Assistant runs the same shape around human labour: $3M recorded 1 March 2021, $10.5M recorded 1 March 2022, sold as a monthly plan for a dedicated assistant. Pricing page, billing cycle and renewal logic are indistinguishable from a software product’s. What arrives when you pay is a person.
None of those three is a rounding error and none of them is a SaaS company. Recurring revenue is necessary and it is not sufficient; the clause about renting software the vendor operates has to hold as well.
Usage-based pricing nobody would call a subscription
Wasabi sells cloud object storage priced by the terabyte stored rather than by the seat, so the bill moves with what the customer keeps. Revenue went from $18M recorded 11 February 2020 to $54M recorded 20 May 2021, with 128 employees on the later date. Nothing about that pricing looks like a plan, and it still satisfies all three clauses: rented access, operated by the vendor, paid again next month or the product goes away. The argument it invites is about the layer rather than the model — storage sits closer to infrastructure than to an application, which is the distinction SaaS, PaaS and IaaS exists to draw. What metered pricing genuinely lacks is contracted revenue: a customer who deletes half their data halves the bill without cancelling anything, and no churn report will ever show it.
The line GetLatka has to draw thousands of times
Verblio, which sells written content produced by a network of freelance writers, went from $3.12M recorded 25 November 2018 to $10M recorded 3 October 2021. Its headcount rows disagree with each other, and they alternate: 333 recorded 1 January 2021, 22 on 9 February 2021, 339 on 1 August 2021, 31 on 3 October 2021, 349 on 1 January 2022. Nobody hires and fires three hundred people twice in a year. Two defensible answers to “how many people work here” are both in the record, depending on whether the writer network counts as staff.
That is not a data-quality footnote. It is the definition failing to be obvious, showing up in the one place it has to be resolved anyway. Every row in the GetLatka database of software companies is a judgment about which side of the line a business sits on, and the judgments that matter are never made about Slack or Salesforce. They are made about the content marketplace, the staffing platform, the insurer with an API, and the agency whose retainer looks like a subscription in every respect except what the customer is buying.
Why the definition is worth defending
Odoo shows what it buys. Its core software is open source and can be self-hosted for free, so no customer is ever compelled to pay, and 11,000 of them were recorded in December 2018 anyway. The revenue ladder that produced is the reason anyone bothers naming the category at all.
The first six years of that ladder look like a small business. The database flags a $44M figure dated 1 January 2018 as an estimate; the confirmed rows resume at $47.5M in December 2019, and the company had 2,025 employees by December 2021. Compounding is what recurring revenue does when it is left alone, and it does not happen to a company that sells its product once.
Nearly ten years separate Odoo’s $4.9M from its $160M, and every dollar in between came from customers free to stop paying who did not. PAR recorded more revenue for December 2012 than for December 2019, because a terminal already installed in a restaurant owes nobody a renewal. Ask what your customer loses by cancelling next month. That answer, not the product, is what makes a company SaaS.
SourcesCompany figures are dated rows from the GetLatka database, quoted with the date each was recorded: PAR Technology (December 2012 to December 2021), Odoo (December 2011 to 29 July 2021, with the January 2018 revenue figure flagged there as an estimate), Magic Eden (1 January 2022), Directive Consulting (October 2016 and 21 October 2021), Inflectra (May 2021 and 21 November 2021), SafetyWing (June 2019 and 21 November 2021), Wing Assistant (1 March 2021 and 1 March 2022), Wasabi (11 February 2020 and 20 May 2021) and Verblio (25 November 2018 to 1 January 2022). Revenue-per-employee figures are our arithmetic on those rows.
