Chris M. / 14 July 2026 / Reviewed 11 August 2026

Is SaaS Dead? Off-the-Shelf Platforms and UK Construction

The question arrives in a particular form. Someone forwards a clip in which a technology chief executive appears to announce the end of business software, and asks whether the firm should stop renewing its platforms and have something built instead. It is a reasonable question asked on a false premise, and the false premise is worth clearing out of the way before anything useful can be said.

Satya Nadella did not say SaaS is dead. That phrase is a headline. What he said, on the BG2 podcast in December 2024, is that business applications are "essentially CRUD databases with a bunch of business logic", that "the business logic is all going to these agents", and that the notion business applications exist as separate things "is probably where they'll all collapse" in what he called the Agent Era ¹. That is an argument about where logic sits, not a prediction that firms will stop buying software. Marc Benioff, who has an obvious interest of his own, called the death-of-SaaS argument "so much nonsense" on a Salesforce earnings call in September 2025 ².

Neither man is disinterested and neither should be taken at face value. The useful position sits underneath both, and for a UK contractor it is this: off-the-shelf platforms remain the right answer for most of what a construction business does, the pricing model underneath them is genuinely changing, and the case for building something of your own is narrower and more specific than the current noise suggests.

The case study everybody quotes, and how it actually ended

Klarna is the example reached for whenever someone argues that AI has made business software optional, so it is worth following all the way through rather than stopping where the argument is convenient. In February 2024 Klarna announced that its AI assistant was handling two thirds of customer service chats in its first month, doing "the equivalent work of 700 full-time agents", cutting resolution from 11 minutes to under 2, and driving what the company projected as "$40 million USD in profit improvement" for that year ³. That is the company's own figure, self-reported, and it is where most retellings stop. In August 2024 came the line that made the rounds: "We just shut down Salesforce. Within a few weeks, we will shut down Workday."

Then the correction. In March 2025 the chief executive, Sebastian Siemiatkowski, said plainly: "So no, we did not replace SaaS with an LLM." Klarna had built an internal stack using Neo4j and other components, which is a different claim entirely ⁴. In the same period he told TechCrunch "I don't think it is the end of Salesforce; might be the opposite", and, asked whether other firms should follow, "Will all companies do what Klarna does? I doubt it" ⁵. By May 2025 Klarna was recruiting human customer service staff again, with Siemiatkowski conceding that "cost unfortunately seems to have been a too predominant evaluation factor" and that "what you end up having is lower quality" ⁶.

A company with a large in-house engineering function, running standardised digital processes, tried the most aggressive version of this and walked a material part of it back within fifteen months. That is not an argument against ever building. It is an argument for being honest about what building costs after the launch announcement.

What is genuinely changing: the meter, not the model

The substantive shift is in how software is charged for rather than whether it is bought. IDC's European software research lead put it as "SaaS, as we know it, is being disrupted, not by decline but by evolution", and IDC projects that by 2028 pure seat-based pricing will be obsolete, with 70 per cent of vendors moving to alternative models ⁷. Bessemer's 2025 Cloud 100 analysis records the same move in practice, citing Intercom shifting to outcome billing where customers pay only when the AI resolves a conversation, and notes the average Cloud 100 revenue multiple falling to 20x from 34x in 2021 ⁸. Both are published by investors with positions in the outcome, which is worth holding in mind, though the direction is corroborated across independent sources.

Spending itself is not falling. Battery Ventures' survey of large-enterprise technology leaders found 70 per cent reporting increased 2025 budgets against 2024, with only 17 per cent expecting any decline, and 76 per cent citing vendor consolidation as the lever they pull when they do cut ⁹. Consolidating vendors is not the same as leaving SaaS. It is buying more from fewer people.

Two cautions on that source. It surveys around one hundred executives at large enterprises, 89 per cent of them at firms of a thousand people or more, and it is published by a venture capital firm. It says very little about a fifty-person subcontractor in the West Midlands, and it should not be stretched to.

The test the UK government already publishes

The most useful decision framework here is not from a vendor at all. The Government Digital Service publishes a build-or-buy standard, and it is unusually direct.

Buy, it says, when "there is a commercially available way to meet most of your user needs" and "you do not need a high level of customisation or bespoke changes". Build when "your user need is unique or rare", or when "you cannot scale, adapt or integrate available commercial products to meet your core needs" ¹⁰.

Then the sentence that should be read twice by anyone about to commission a heavily configured platform: "Even small modifications to OTS software can remove most of the benefits of using it." The guidance lists what customisation does, which is increase costs, make maintenance harder, reduce the ability to scale and adapt, and restrict future upgrades ¹⁰. A related GDS post adds the point most build proposals skate over: "You cannot outsource risk. It's important to make sure you have the resources, insight and knowledge to manage and oversee your products in the long-term" ¹¹.

What building actually costs when it goes wrong

The evidence on bespoke software risk is unusually good, because governments are obliged to publish their failures. The National Audit Office puts UK public sector digital and technology spend at a minimum of £14 billion a year, and found more than £3 billion of cost increases across five digital change programmes it examined, alongside delays totalling at least 29 years and cost growth of at least 26 per cent against original forecasts ¹².

One of its earlier examples reads like a parable for this exact decision. On the Ministry of Justice's electronic monitoring programme, "the bespoke requirements proved too ambitious to implement, resulting in serious delays and parting company with two successive suppliers". The resolution was to buy: the department "decided to procure existing GPS tags that were already available in the market", five years later than planned ¹³.

The academic picture is consistent. Flyvbjerg and Budzier's study of 1,471 IT projects found an average cost overrun of 27 per cent, but the average conceals the real danger: one project in six was what they term a black swan, with cost overruns averaging 200 per cent and schedule overruns near 70 per cent ¹⁴. A one in six chance of a project running to three times its budget is not a tail risk a contractor with a 3 per cent margin can absorb casually.

Does AI make building cheap enough to change the answer?

This is the load-bearing assumption in most build proposals written since 2024, and the evidence for it is weaker than the confidence around it. A randomised controlled trial by METR, using sixteen experienced open source developers across 246 real issues, found that when developers were allowed AI tools they took 19 per cent longer, and that they believed they had been sped up ¹⁵. METR is careful about the scope of that result and does not claim AI fails to help most developers, and the tools tested were those of early 2025, so it should be read as a caution rather than a verdict. Google's DORA research across roughly five thousand respondents found 90 per cent using AI at work, and that AI adoption "does continue to have a negative relationship with software delivery stability", with 30 per cent reporting little or no trust in AI-generated code ¹⁶.

Faster first drafts are real. Faster maintenance, stability and institutional knowledge are not the same thing, and it is the second set that determines whether a bespoke system is still working in year four.

Where this actually lands for a UK construction business

Construction is not an average industry in this respect, and the UK data is specific enough to be worth stating. Construction is the largest sector by number of SMEs, with 885,000 businesses, 16 per cent of the total ¹⁷. It is also close to the bottom for technology adoption. The Department for Business and Trade's Longitudinal Small Business Survey found 69 per cent of SME employers using technologies or web-based software overall, with construction at 58 per cent ¹⁸, and among businesses with no employees construction was the lowest sector of all at 29 per cent ¹⁹. ONS data records construction at 13 per cent AI use, and notes that construction businesses "are more likely to access AI through the purchase of external software or ready-to-use services" ²⁰.

A respondent to the government's AI adoption research, a small construction business in Scotland, made the practical case better than any consultancy could: "With any software development there will be fairly significant cost, whereas if you buy something off the shelf, you can pick it up and drop it" ²¹. That instinct is correct far more often than it is wrong. The honest answer to the question in the title is that SaaS is not dead, its pricing is being rebuilt, and for most UK contractors the right move is to buy the commodity and reserve building for the places where the business genuinely differs from every other contractor. Those places exist, and they are usually narrower than they first appear: the specific approval routes, the inspection and test plans, the quality gates and the way a particular firm records what happened on site. That is the subject of the rest of this series.

A closing note on the sceptical case, because a consultancy that builds bespoke systems has an obvious interest in the answer being "build". Battery's mid-2026 survey found only 16 per cent of organisations seeing positive return on more than half their AI projects, and 31 per cent seeing return on less than a quarter, with 94 per cent lacking any consistent framework for measuring it at all ²². Anyone selling you a build, including us, should be asked which of those two groups their last three clients ended up in.

Three questions follow directly from this one and are answered elsewhere in this library. The narrower version of the same decision, where the off-the-shelf option is already installed and paid for, is set out in Microsoft Copilot versus a custom construction AI assistant. The arithmetic for comparing a licence bill against a build, using your own numbers rather than a vendor's, is in how to calculate the return on construction AI. And the cheapest way to find out which side of the GDS test your firm actually falls on is to test it for a month before committing, which is the subject of running a safe 30 day pilot.

What could not be established for this guide

Two things, stated because a guide that only reports what supports its argument is marketing. Reliable figures on licence utilisation and shelfware, meaning paid seats that nobody uses, could not be verified from any primary source. The published numbers in this area come from SaaS management vendors whose product is sold on the size of the problem, and the specific reports sought were not retrievable. The shelfware argument is therefore made in this series from vendor pricing structures rather than from adoption statistics.

Gartner's worldwide IT spending figures could not be retrieved from Gartner directly, and the secondary reporting located for them contained an evident unit error. No Gartner figure is quoted here as a result.