Renting software was the whole argument

The case against the box

In 1999, enterprise software arrived in a box, or rather on a stack of CDs, accompanied by an implementation project measured in months and a licence fee measured in hundreds of thousands of dollars. The model was settled. Oracle sold database licences. SAP sold ERP licences. The customer bought the right to run the software on hardware the customer owned, in a data centre the customer maintained, administered by staff the customer hired. The annual maintenance contract — typically eighteen to twenty-two percent of the licence cost — bought patches and the right to call a support line. This was on-premise computing, and it was not framed as a choice: it was simply what software was.

Marc Benioff had spent thirteen years at Oracle before he left to found Salesforce. He had watched customers sign large cheques and then spend the next eighteen months failing to get the software working the way the salesperson had described. The core product he chose to disrupt was sales-force automation — what the industry called CRM, customer relationship management — because it was a category where the gap between the promise and the delivered reality was unusually wide. Siebel Systems dominated the space. Siebel's implementations were notoriously long and expensive, and the software was frequently abandoned by the sales teams it was meant to serve, because those teams found it easier to keep their own spreadsheets.

Benioff's argument was simple and deliberately theatrical. Salesforce launched with a "No Software" logo — a circle-and-slash over the word — and a pitch that the application would live on Salesforce's servers, be accessed through a web browser, and be billed monthly per user. No implementation project. No hardware. No upgrade cycle managed by the customer's IT department. The term "software as a service," which Salesforce did not invent but did more than any other company to establish as an industry category, captured the model: the software was a utility, like electricity, and you paid for what you used. The acronym that stuck was SaaS, a usage model whose billing logic would reshape every major vendor's pricing within fifteen years.

A decade of resistance

The incumbents did not immediately concede the argument. Their position, stated with varying degrees of candour, was that enterprise software was too complex and too mission-critical to be trusted to a third party's infrastructure. Security objections were raised loudly. A database containing a company's customer records and pipeline data, stored on shared hardware operated by a startup, was presented as an obvious risk. The objection was not entirely bad faith: in 1999, the infrastructure for delivering reliable software over the internet at enterprise scale was genuinely immature. Salesforce itself suffered several high-profile outages in its early years that competitors cited as evidence.

The more durable resistance, though, was financial. The on-premise licence model was extraordinarily profitable for the vendors. A customer who paid a large upfront licence fee and then an annual maintenance contract was a customer who was committed, whose switching costs were enormous, and whose relationship with the vendor was governed by contracts that took years to unwind. The SaaS model replaced that structure with a monthly subscription that was, in principle, cancellable. Analysts and CFOs who looked at SaaS renewals pointed out that a vendor whose customers could leave had a very different business than one whose customers were structurally locked in by sunk cost, data gravity, and the sheer weight of customisation built on top of the original installation.

What the incumbents could not argue away was that Salesforce worked, and that sales teams used it. By the mid-2000s, Salesforce had passed a hundred thousand subscribers. Benioff had understood something the on-premise vendors were slow to credit: the person whose adoption actually mattered for CRM was not the CIO who signed the procurement contract, but the sales representative who was expected to enter data into the system. A web application that worked from a laptop in an airport without a VPN had a fundamentally different adoption profile than a client-server application that required a configured corporate network. The end-user argument was not a marketing position; it was an architectural one.

What copying it required

By 2010, the major incumbents had conceded the category. Oracle began acquiring SaaS companies — RightNow in 2011, Taleo and Eloqua in 2012 — rather than building the capability organically, because organic build was slow and the customer base was moving. SAP acquired SuccessFactors in 2012 for approximately $3.4 billion, a figure that signalled how seriously it was taking the threat to its human capital management business. Microsoft, which had its own on-premise CRM product, pivoted Dynamics toward a cloud delivery model across the 2010s. The no-software argument had won, and the vendors who had resisted it longest were now paying acquisition premiums to buy their way into it.

Workday, founded in 2005 by former PeopleSoft executives including Dave Duffield and Aneel Bhusri, was the clearest case of the next generation of enterprise software being built cloud-native from the start, specifically for HR and finance. NetSuite, which had been founded even before Salesforce and was eventually acquired by Oracle in 2016 for approximately $9.3 billion, had made the same bet on ERP. The incumbents were acquiring rather than building because the SaaS model was not simply a deployment change — it required a fundamentally different approach to data migration, versioning, and the multi-tenant architecture that let a single codebase serve thousands of customers simultaneously without giving any of them access to each other's data.

The thing the SaaS transition did not eliminate was lock-in. It transformed it. An on-premise customer was locked in by the cost and complexity of moving data and customisations off a system installed in their own data centre. A SaaS customer was locked in by the same data gravity — their historical records, their configured workflows, their integrations with other systems — combined with the new reality that the underlying infrastructure was entirely in the vendor's hands. Salesforce customers who wanted to leave Salesforce faced data export challenges, API dependency problems, and the now-familiar discovery that the customisations they had built on the platform were not portable to a competitor. The pricing had changed from upfront to recurring; the structural position of the customer had not changed as fundamentally as the original argument implied.

What Benioff had correctly identified in 1999 was not that renting software would set customers free. It was that the friction of deployment and administration was high enough, and the failure rate of large on-premise implementations visible enough, that a model which moved those problems to the vendor would find a market. The FoxMeyer collapse in 1996, Hershey's fulfilment failure in 1999, and a long list of less publicised disasters had established that on-premise enterprise software was not simply complex but reliably, expensively dangerous to install. Against that record, renting seemed reasonable. It turned out to be a different set of risks, with a different set of beneficiaries.