The 'build vs buy' question is outdated
For the better part of two decades, 'build vs buy' meant: write it yourself with an engineering team (expensive, slow, your headache forever) or subscribe to a SaaS (fast, cheap upfront, someone else's problem). That framing assumed building was hard and buying was simple. Neither is fully true any more. Agentic engineering has compressed build timelines dramatically. And 'buying' a SaaS in 2026 means signing up for an open-ended per-seat rent contract with annual increases, no exit clause, and no ownership of the outcome. The real question is: rent forever, or own it once? And for a specific class of tools, the answer is increasingly obvious.
Four signals that you should replace a SaaS
First: your team uses fewer than half the features, yet pays for the whole product. If your field-service team uses job scheduling, mobile dispatch, and invoice generation but ignores the asset-management module, the IoT integrations, and the customer portal, you are paying a seat-weighted bill for software you do not use. Second: the annual bill exceeds £100 per month and grows with headcount. At that level a five-year projection almost always crosses a co-funded build cost. Third: you have had a data-portability scare — a vendor acquisition, a pricing restructure, or a feature you relied on being moved to a higher tier. Fourth: you have been a customer for more than two years with no end in sight. If the subscription has no natural finish line, you are renting forever. That is the clearest signal of all.
Four signals that you should NOT replace it
Be honest about when the SaaS is genuinely irreplaceable. Payment rails — Stripe, GoCardless, Adyen — are not just software; they are licensed financial infrastructure with compliance certifications, banking relationships, and fraud-detection models that took years to build. Accounting and ERP systems like Xero or Sage touch your regulatory filing obligations and carry audit trails that must meet legal standards. Any tool where the vendor network itself is the value — LinkedIn for recruitment, a job board, an industry-specific marketplace — cannot be replicated because the data and the audience live on the platform. And tools carrying SOC 2, ISO 27001, or FCA compliance certification have cleared regulatory bars that an agentic build would need its own programme to achieve. For all of these, keep paying the rent.
Why co-funding changes the economics
The traditional objection to replacing a SaaS with a custom build was the upfront cost. A bespoke rebuild of a field-service management tool might cost £35,000–£60,000 in senior engineering time. No single 30-person company wants to absorb that alone, especially when the SaaS is 'good enough.' Co-funding dissolves that objection. Five companies with the same core use case — say, FM contractors all using the same job-scheduling tool — each contribute a founding share. The build cost per company drops to £7,000–£12,000. That is less than two months of what many companies already pay the SaaS vendor. The result is a tool shaped around the actual workflow, owned collectively in perpetual licence, with no per-seat bill.
The AI factor
It is worth being clear about what 'AI-fast' means and what it does not mean. Agentic engineering — senior engineers directing code-generation agents, reviewing output, and taking accountability for delivery — is genuinely faster and cheaper than a traditional development engagement. A rebuild that might have taken six months in 2022 takes six to eight weeks in 2026. That reduction in engineering hours is what makes the per-company build cost low enough to reach crossover in year one. What it does not mean is vibecoding: a prototype assembled by a solo operator with no tests, no architecture, and no one accountable when it breaks at 9 a.m. on a Monday. Delivery is senior-led. The agent is the tool, not the team.
How to start
The first step is not committing to a build — it is a scoping conversation about which tool you would replace if the economics made sense. Rollout IT maps the features you actually use, produces a build estimate, and matches you with other companies in the same situation. If a founding-five cohort forms, the build starts. If not, you have a clear picture of the maths with no obligation. The migration happens in parallel: you keep using the old tool until the replacement is live and you have verified the data import. Then you cancel. The whole process, from first conversation to go-live, typically runs eight to fourteen weeks.