Skip to content

The Rise of SaaS

Abstract

When Marc Benioff founded Salesforce in 1999 with the slogan “No Software,” it read as a provocation. Twenty years later, renting software by the month was the normal way to sell it. This article follows software as a service (SaaS) from the time-sharing bureaus of the 1960s and the failed application service providers of the dot-com years to Salesforce, Office 365, Adobe’s Creative Cloud, and the pandemic boom of Slack and Zoom. It also covers the costs of the model: customers own no copy of the program, their data sits with the vendor, and a single mistake at the vendor, such as Salesforce’s permissions script of 2019 or Atlassian’s deleted sites of 2022, stops thousands of companies at once.

Before SaaS

In the 1990s, enterprise software was a physical product. SAP R/3 arrived on dozens of CDs. Oracle licenses were priced by CPU count and cost millions. Upgrades meant projects: months of planning, external consultants, and downtime.

The vendors’ business model was lucrative: a one-time license fee (large), an annual maintenance fee (17 to 22% of the license price), and implementation partners who often earned more than the software maker itself. Siebel Systems dominated customer relationship management (CRM) with exactly this model: complex and expensive, with no alternative for large enterprises.

From the customer’s side, the money flowed before the benefit was visible. Implementation projects frequently failed or blew their budgets, and software running on local servers aged until the next costly upgrade round.

Renting computing was not new. From the mid-1960s, time-sharing bureaus sold access to central computers by the hour over telephone lines, and they were a major part of the computer services industry until the personal computer made them obsolete in the early 1980s (see The Cloud Computing Era). What SaaS added was the web browser as a universal terminal and a single program serving every customer.

The Application Service Providers

The first attempt to rent business software over the internet came from the application service providers (ASPs) of the late 1990s. Companies such as USinternetworking (USi) of Annapolis, Maryland, and Corio took existing enterprise packages from SAP, PeopleSoft, or Siebel, installed a separate copy for each customer in their own data centers, and charged a monthly fee for running it. In 2000 USi was the market leader, with more than $100 million in revenue.

The model had the costs of both worlds. Each customer still had its own installation, with its own customizations and its own upgrade projects, so the provider gained little from scale, and it had to finance data centers and software licenses up front. When the dot-com bubble burst, the money stopped. USi filed for Chapter 11 bankruptcy protection in January 2002, and Breakaway Solutions, FutureLink, and Interliant failed in the same months; the ASP Industry Consortium ceased to function. Computerworld put USi’s failure down to slower-than-expected adoption of the ASP model and a weak economy, which left the company with more debt than its customers’ fees could carry.

Marc Benioff and “No Software”

Marc Benioff had worked at Oracle, knew the license business from the inside, and disliked it from the customer’s perspective. In 1999 he founded Salesforce.com on a simple thesis: CRM software does not need to be installed. It runs in the browser.

The difference from the ASPs was in the architecture. Salesforce ran one program, in one version, for all customers at once, with each company’s data kept apart inside a shared system. This multi-tenant design meant that an improvement or a fix went to every customer on the same night and that the cost of running the service fell as the customer count grew.

The early slogan was aggressive: “No Software,” with a crossed-out CD-ROM symbol as the logo. It was a political statement more than a technical description, and Benioff understood marketing. The model:

  • No installation, no local IT infrastructure
  • A monthly or annual subscription fee per user
  • Automatic updates, without customer projects
  • Scalable: five users today, five hundred tomorrow

The early years were hard. Large enterprises distrusted the browser as a work environment and asked how they could hand their customer data to an outside provider. Salesforce grew first through smaller companies that could not or would not build Siebel infrastructure. The dot-com bust that killed the ASPs helped it: as IT budgets shrank, the argument of no capital expenditure, predictable operating costs, and fast implementation won customers.

The Economics of the Subscription

Info

Why subscription models suit software vendors better than one-time sales:

A company selling software for €100,000 books €100,000 in revenue once. A SaaS vendor selling the same functionality for €2,000 per month has generated only €24,000 after year 1, but €120,000 by year 5, with growing predictability.

The decisive difference: Annual Recurring Revenue (ARR) is predictable and bankable. From the mid-2010s, investors valued SaaS companies at 10 to 15 times ARR, far above the multiples of traditional software companies. That financed aggressive growth. Net Revenue Retention (how much of last year’s revenue remains, plus expansion) became the central health metric: above 120% means the existing customer base grows on its own.

The SaaS vocabulary became its own dialect: MRR (Monthly Recurring Revenue), ARR, churn, CAC (Customer Acquisition Cost), LTV (Lifetime Value), NRR (Net Revenue Retention). Each of these concepts existed before in related form; SaaS made them the standard language of investors, CFOs, and founders.

The calculation changed for customers too. Software stopped being capital expenditure (CAPEX, a balance sheet item) and became operating expense (OPEX, on the profit-and-loss statement). Many finance departments welcomed the change: no depreciation schedule, no debate about capitalization.

The SaaS Stack Grows

In 2005 Salesforce opened its platform to third parties with AppExchange, an app store for business extensions, all sold on subscription.

Specialized SaaS vendors appeared for every business function:

  • ServiceNow (2004): IT service management
  • Workday (2005): HR and finance software for large enterprises
  • HubSpot (2006): marketing and CRM for mid-sized companies
  • Zendesk (2007): customer support
  • Dropbox (2007): file storage as a service

GitHub (2008), Twilio (2008), and Stripe (2010) brought SaaS to software developers. Over the following decade the software stack of a typical company shifted from purchased licenses to subscriptions.

Office and Photoshop Move to the Cloud

The largest desktop vendors followed. Microsoft had already sold hosted Exchange and SharePoint as the Business Productivity Online Suite (BPOS). On 28 June 2011 it replaced BPOS with Office 365, launched at events around the world and available in forty markets. “We’re here to introduce Microsoft Office 365, where Microsoft Office meets the cloud,” Steve Ballmer said in New York. The package combined Office with Exchange Online for mail, SharePoint Online for documents, and Lync Online for messaging and calls. On 29 January 2013 Microsoft brought the subscription to private customers with Office 365 Home Premium. On 21 April 2020 most consumer and small-business plans were renamed Microsoft 365.

Adobe went further. On 6 May 2013, at its MAX conference, it announced that there would be no further versions of Creative Suite: Photoshop, Illustrator, InDesign, and the other programs would be developed only for Creative Cloud, at about $50 a month for the full set. The programs were still installed on the user’s computer, but using them now required a running subscription.

With these moves the subscription reached people who had never heard the word SaaS. A student writing a thesis and a photographer editing pictures now depended on a license server and, increasingly, on files stored in the vendor’s cloud.

Slack, Zoom, Notion: COVID as Accelerator

In March 2020 the world sent its office workers home. What followed was the largest natural experiment in the history of SaaS.

Slack (launched 2013, public 2019) had already been growing, but the pandemic raised the number of simultaneously connected users from 10 million on 10 March 2020 to 12.5 million on 25 March. Salesforce bought it for $27.7 billion, a deal announced in December 2020.

Zoom (Eric Yuan, 2013) became a verb. “Let’s zoom” replaced “let’s have a call.” Quarterly revenues quadrupled, and its market capitalization briefly exceeded that of all US airlines combined.

Notion (founded 2013, breakout from 2018) became the showcase for “bottom-up SaaS”: individuals and teams adopted the tool without involving an IT department, and the invoice followed the usage. This pattern, product-led growth (PLG), became the preferred sales strategy for new SaaS products.

The SaaS Winter of 2022/23

After a decade of almost uninterrupted growth and rising valuations came the correction. The US Federal Reserve raised interest rates aggressively, and capital that had been cheap for years became expensive.

The effects on SaaS companies were direct:

  • Growth companies that depended on cheap capital saw their valuations halved or cut to a third
  • The “Rule of 40” (growth rate plus profit margin of at least 40%) replaced pure growth as the investor criterion
  • Layoffs across the industry: Salesforce 10%, Workday 3%, and Zendesk was sold to private equity

The subscription model survived. Companies with sound unit economics did well, and consolidation followed as large vendors absorbed smaller ones.

When the Service Is Down

A licensed program installed on the company’s own servers failed one company at a time. A multi-tenant service fails for all its customers together, and the customer can do nothing except wait for the vendor’s status page to change.

On 17 May 2019 Salesforce deployed a database script meant to change permissions for customers of its marketing product Pardot. It “resulted in granting users broader data access than intended”: in affected organizations, every user could suddenly read and change all of the company’s data. Salesforce’s only quick remedy was to block access to every instance that contained an affected customer, more than a hundred server instances, which locked out many customers who had never used Pardot. The outage began at 9:56 a.m. Pacific time and lasted about fifteen hours; administrators then had to restore their users’ permissions by hand, and some were still doing so after the weekend.

On 5 April 2022 Atlassian ran a script to remove a retired app, “Insight – Asset Management,” from customer sites of Jira and Confluence. The team that requested the deletion supplied the IDs of the customers’ entire cloud sites instead of the app’s IDs, and the deletion interface accepted either without checking. Between 07:38 and 08:01 UTC the script deleted 883 sites belonging to 775 customers, with no intermediate stage from which they could simply be undeleted. The deleted data included the customers’ contact details, which made it harder for Atlassian to tell them what had happened. The first sites came back on 8 April and the last on 18 April; some customers had no Jira or Confluence for two weeks. Atlassian’s post-incident review was blunt: “The buck stops with us. Full stop.” It promised “soft deletes” that could be undone.

Identity is the most concentrated point of all. When an automated job at Microsoft deleted a signing key that Azure Active Directory still needed, on 15 March 2021, users of Teams, Exchange Online, SharePoint, and every third-party application that used Microsoft’s sign-in were locked out for hours, though the applications themselves were running (see Famous Software Disasters). Large customers responded with backup communication channels, local copies of critical documents, and contractual service levels, which compensate with service credits rather than restore lost hours.

Dead End: On-Premise Enterprise Software

Warning

SAP, Oracle, and the forced transformation

SAP and Oracle are not dead; their revenues are larger than ever. But their core business has shifted. SAP’s S/4HANA program moves customers to the cloud, and Oracle followed with Oracle Cloud (Larry Ellison had been a SaaS skeptic for years). Both now sell primarily subscriptions.

The actual dead end was the perpetual license: a one-time fee for software that runs forever. Customers liked it (pay once, use forever). For vendors it was economically worse than subscriptions, and capital markets found it hard to value. Microsoft Office was a perpetual license product for decades before the subscription took over.

What disappeared was the class of mid-sized on-premise specialty vendors that lacked the resources to move their whole product to the cloud while still supporting customers on old installations. They were acquired or became irrelevant.

Who Controls the Data

SaaS changed how software is built. When customers can cancel at any time, the product has to stay good after the sale, and vendors created customer success departments, invested in onboarding, and shipped features continuously.

It also shifted power inside companies. End users could subscribe to SaaS tools without IT approval, which weakened IT departments as gatekeepers. “Shadow IT” grew until companies subscribed to SaaS management platforms to keep track of all their other subscriptions.

The shift moved the company’s records too. A customer list in Salesforce, a project history in Jira, or a mailbox in Exchange Online exists in the vendor’s data center, under the vendor’s security practices, the vendor’s outages, and the law of the vendor’s home country (see The Cloud Computing Era on the CLOUD Act and Schrems II). Exporting it is possible, but rarely in a form another product can use without work. Customers who stop paying lose the program, and their files stay available only as long as the vendor’s retention rules allow.

📚 Sources