Skip to main content

Build or Buy in 2026: How UK Businesses Should Decide Between Custom Software and Off-the-Shelf

By Thoughtgears 9 min read
Business leaders reviewing charts and analysis in a modern meeting room while weighing a software decision

For most of the last decade, the build-or-buy question had a boring answer. Unless the software was the thing you sold, you bought it. Building your own CRM was the kind of decision that got written up as a cautionary tale.

That consensus is wobbling, and the reason is AI. Coding assistants have collapsed the apparent cost of a first version to something close to a weekend: a capable product manager can now demo a working internal tool before procurement has finished the vendor’s security questionnaire. When building looks that cheap, the old rule of thumb stops doing its job.

The trouble is that AI has changed the cost of the first 20% of a piece of software and barely touched the other 80%. Here is how to make the decision honestly in 2026: what you are really choosing between, what each route costs in the UK, and the three questions that settle most cases.

The 2026 twist: AI made building look cheaper than it is

The demo is real. The savings are mostly not.

Writing code was never the expensive part of owning software. The expensive parts arrive after the demo: security review, access control, data migration, integration with the systems you already run, test coverage, monitoring, the support rota, and the steady stream of changes that keep it useful. None of those have got meaningfully cheaper, and some have got harder, because AI-generated code arrives faster than anyone can review it properly, a problem we set out in our vibe coding security checklist for CTOs.

Gartner has predicted that more than 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls. That forecast is about AI projects specifically, but the failure mode generalises to AI-built internal tools: something reaches production quickly, nobody is accountable for keeping it alive, and eighteen months later it is either abandoned or quietly load-bearing and unmaintained.

The lesson is not that you should never build. It is that the prototype tells you almost nothing. Judge the build on the cost of year three, not week one.

What you are actually choosing between

Framing this as a binary is the first mistake. There are four routes, and the middle two are where most sensible answers live.

Buy off-the-shelf and adapt your process to it. Cheapest and fastest, and right whenever the process is not a source of competitive advantage. Payroll, expenses, helpdesk ticketing, accounting: configure, do not build.

Buy a platform and extend it. You license something with a real API and build only the thin layer specific to you. The most underrated option on the list: you inherit the vendor’s security posture, compliance work and upgrade path, and spend your engineering budget solely on what differentiates you.

Build custom, but only the differentiating slice. You build the workflow that is genuinely yours and integrate bought components for everything else. Most successful custom builds look like this rather than a grand replacement of an entire system.

Build the whole thing. Justified when the software is the product, when no adequate vendor exists, or when a regulatory or data-residency constraint rules the market out. Rare, expensive, occasionally exactly right.

If you are weighing option one against option four with nothing in between, the analysis is too coarse.

The three questions that settle most decisions

Most build-or-buy debates run long because the participants argue about cost when they disagree about something else. These three questions resolve most cases.

Is this process a source of competitive advantage, or just a cost of doing business? If a competitor could copy your process by buying the same product, buy the product. Build only where doing it differently is the point. The honest test: could you describe what makes your version better in a sentence a customer would care about?

Who owns this in three years? Not who builds it, but who is accountable in 2029, when the builder has moved on and the framework is two major versions behind. If you cannot name a team and a budget line, you are not buying software, you are accruing a liability.

How fast does this need to change? Bought software changes at the vendor’s pace. If your requirements shift quarterly and the vendor ships annually, you will spend years fighting the product. If the process has been stable for a decade, the flexibility you would gain from building is worth very little.

When the answers are close, ask what happens if you are wrong. Reversing a purchase means a migration and a cancelled contract; reversing a build means writing off the investment and doing the migration anyway. Buying is usually the more reversible mistake, which is a reasonable tiebreaker.

What each route actually costs in the UK

Treat every figure here as an indicative band for sense-checking quotes, not a benchmark to negotiate from. Rates move, and the public sources are industry aggregators rather than official statistics.

On the build side, UK custom software projects fall into recognisable bands: approximately £10,000 to £30,000 for a simple internal tool, £30,000 to £80,000 for a standard business application, £70,000 to £150,000 for a complex platform with multiple integrations, and £150,000 upwards (often well beyond £500,000) for enterprise systems. Underneath those numbers, UK contract developer day rates cluster around approximately £500 to £510 at the median, with senior engineers commonly at £600 to £800 and solution architects or specialist AI engineers reaching approximately £1,200. London typically adds 20 to 40%. Our guide to custom software engineering services in the UK unpacks the four commercial models behind those rates.

The number that never appears in the business case is maintenance. A reasonable planning assumption is 15 to 25% of the original build cost every year, indefinitely, which over five years roughly doubles the headline figure. A build that beats a licence in year one is frequently more expensive by year four.

On the buy side, the licence price is similarly misleading. Industry analysis puts SaaS total cost of ownership at roughly two-and-a-half to four times the headline price once implementation, integration, migration, administration and training are counted, and finds around 46% of licences go unused in a typical month: less an argument against buying than for buying carefully and reviewing seat counts.

The defensible comparison is five-year total cost of ownership on both routes. Comparing a build quote to an annual licence fee compares two things that are not alike.

Where each route fails

Builds fail in three predictable ways: scope that was never really agreed, where the specification described the happy path and the edge cases arrive in month four as change requests; the maintenance cliff, where the build succeeds, the team is redeployed and the software decays; and key-person risk, where the one person who understood it has left.

Purchases fail on integration you assumed was straightforward and was not, then on the customisation that pushed you onto a bespoke version you can no longer upgrade cleanly. Vendor lock-in is real but overstated; the immediate risk is configuring yourself into a corner that makes migration expensive.

Both share a root cause: the decision was made on the sticker price, and the work that determines the real cost was never scoped. That is what a short, structured discovery exercise exists to surface: we have written up how to run one in our guide to running a technical discovery sprint.

A practical way to run the decision

Give it two weeks, not two months.

Write down the outcome you want in business terms, without naming a solution. Genuinely survey what exists, including platforms you could extend, not just the two vendors someone already has a relationship with. Score the shortlist against the three questions above rather than a feature matrix, because feature matrices favour whoever has the most features rather than whoever solves your problem.

Model five-year total cost of ownership for the strongest build and buy candidates, maintenance and implementation included. Where you are uncertain, say so in the model rather than picking a confident-looking number.

Then name the owner. If nobody will put their name against operating the thing in 2029, the decision is not finished, whichever way it went.

Final Thoughts

AI has not settled the build-or-buy question. It has made the cheap-looking answer more tempting and the expensive consequences slower to arrive, which is a worse combination than it sounds. The discipline that protects you is unchanged: be honest about which parts of your business are genuinely differentiated, be honest about the five-year cost on both sides, and never approve a build without naming the team that will still own it in three years.

If you are weighing a build against a purchase and want an independent read on delivery cost and risk before you commit, get in touch with our team: we would rather tell you to buy than sell you a build you do not need.

FAQs

Is it cheaper to build or buy software in 2026?

Over one year, buying is almost always cheaper. Over five years it depends on maintenance (typically approximately 15 to 25% of the original build cost annually) and on how many licences you would actually pay for. Compare five-year total cost of ownership on both sides, not a build quote against an annual licence fee.

Has AI made building custom software cheaper?

It has substantially reduced the cost of a first working version and had much less effect on security review, integration, testing, support and long-term maintenance, where most of the lifetime cost sits. Treat an AI-built prototype as evidence the idea is feasible, not that the total cost has fallen.

When should a business build custom software rather than buy?

Build when the process is a genuine source of competitive advantage, when no adequate product exists in the market, or when regulatory and data-residency constraints rule out the available options. For anything a competitor could replicate by buying the same product, buying is usually the better commercial decision.

What is the biggest hidden cost of buying off-the-shelf software?

Implementation and integration, followed by licences nobody uses. Industry analysis puts SaaS total cost of ownership at roughly two-and-a-half to four times the headline price once implementation, migration, administration and training are included, and finds that around 46% of licences go unused in a typical month.

What is the middle option between building and buying?

Buying a platform with a capable API and building only the thin layer that is specific to your business. You inherit the vendor’s security, compliance and upgrade path while spending your own engineering budget solely on the part that differentiates you. It is the most commonly overlooked option and frequently the right one.

Disclaimer

This article is general commercial and technical guidance, not legal, financial or procurement advice. Cost bands, day rates and market figures are indicative and drawn from industry aggregators and vendor-published research rather than official statistics; verify them against current quotes and your own circumstances before relying on them for budgeting. Thoughtgears provides software engineering and team augmentation services, so we have a commercial interest here, which is precisely why the framework above is built to tell you when not to build.

Need a dedicated development team?

Thoughtgears sources, vets and fully manages the engineers who work as your team – you simply direct the work. Book a free discovery call to scope what you need.

Book a discovery call

More from the blog

Book a call