Technology 5 min read By Arthur Ellington
CIOs Urged to Weigh Build-or-Buy AI Software Decisions Carefully
Enterprise technology leaders are being advised to assess core competencies, hidden costs, security risks and long-term support before building custom AI software in-house rather than buying established platforms.
Business leaders weighing whether to build their own artificial intelligence software or buy established platforms are being urged to apply a series of hard-nosed tests before committing resources, as the appeal of replacing subscription products with custom AI-built applications collides with the realities of cost, security and long-term maintenance.
The debate has gained traction in enterprise technology, where the prospect of using AI coding tools to create bespoke applications has led some chief information officers to consider abandoning platforms such as SAP, Workday or HubSpot. The pitch is straightforward: cut recurring licence fees that can run into millions of pounds a year and put engineering teams to work on tailored systems.
But experience from the CIO seat suggests the calculation is rarely that simple. One technology leader recalls being asked by a beverage company chairman to build a custom analytics dashboard. The team had the talent and the resources, yet four months of setbacks forced a harder question: even if the company could build it, should it? The project was cancelled and a commercial solution purchased instead, on the grounds that every month spent building was time not spent on priorities that could create more value.
That lesson now informs advice to enterprises facing the same choice in the AI era. The first question is whether software development is genuinely a core competency. Even sophisticated engineering organisations with hundreds of thousands of employees have concluded that building billing or HR systems in-house distracts from innovation. For most businesses, the priority is leaning into their own expertise rather than constructing something unrelated to their central mission.
The second test concerns the true scale of any saving. The business case for building often begins with the subscription fee that will be eliminated, but that figure ignores development, operation and maintenance costs. Companies building their own large language model-based software solutions have typically spent five to 10 times more than they would have using an established workflow automation platform, according to experience cited in the debate. Initial development is only the tip of the iceberg; maintenance, security updates and keeping the system current as the business changes represent the real expense.
One large financial services company considered using an LLM to rewrite a core system, a process that could easily take 18 to 24 months, all to save the equivalent of 0.5 per cent of its annual operating budget. Such examples underline how quickly the arithmetic can turn against a build decision once the full lifecycle is counted.
Security and governance form the third area of scrutiny. Getting software to work on day one is not the hard part; defending it on day 1,000 is. Cybersecurity in the AI era has become more complex and the stakes higher, particularly when autonomous AI agents can exceed their permissions. Roughly half of organisations have seen AI agents go beyond their authorised access, and one car rental management platform had its production database wiped by a coding agent in nine seconds. CIOs are therefore being advised to assess honestly whether their platform can withstand evolving threats, whether it is auditable for regulators, whether permissions can be tracked and whether it can comply with shifting rules.
A fourth question is what happens when the CIO leaves. Average tenure for CIOs is about four and a half years, while a custom platform typically takes two or more years to build and longer to mature. The person who designed it, made the architectural decisions and knows where every integration lives may be gone before the system is running well. One CIO described a previous employer where a predecessor built a custom platform the whole organisation depended on. When he left, so did the institutional knowledge: no documentation, no support team, no vendor to call, leaving senior leaders with something they could neither explain nor afford to shut down.
Yet building in-house is not always the wrong answer. A chief digital and information officer at one of the world's largest shipping and logistics companies is pursuing an in-house AI strategy, but only greenlights projects that differentiate the business and draw on its proprietary data and decades of institutional knowledge. That selective approach suggests the build-or-buy decision is less a binary choice than a test of strategic focus, financial discipline and operational readiness.
5
