Ask what API integration costs and the answer is that it is the largest variable in any software quote, which is why one row of our published price table can run from £15,000 to £60,000 for what sounds like the same system. The features on your list are the cheap part. Connecting the new system to accounting, payments, identity, your CRM and whatever the office has run on since 2010 is where the budget and the timeline go.
And it keeps going after launch, because the other end of every integration belongs to someone else, and they change it.
Why do integrations cost more than features?
Because a feature is entirely under your supplier's control and an integration is half under someone else's. A booking screen is a known quantity. A connection to your accounting package depends on that package's documentation, its call limits, how its data is shaped, and what it does when something goes wrong at 2am.
MuleSoft's 2026 Connectivity Benchmark (February 2026, a survey of 1,050 enterprise IT leaders, so the app counts are far above a UK SME) found the average organisation runs 957 applications with 27 per cent of them connected, IT teams spend 36 per cent of their time building and testing integrations, and 26 per cent of projects were late in the last year. Postman's 2025 State of the API report (5,700 respondents) found 69 per cent of developers spend more than ten hours a week on API work. Over a third of technical effort, in large organisations that have been at this for decades, goes on plumbing.
Our cost guide says the same thing in one sentence: connecting to a modern accounting package with a documented API is a small piece of work, and connecting to a sixteen-year-old on-premise system with no API is a project in its own right.
What decides the API integration cost?
Six things, and each one can move a quote by more than any feature on the list.
Driver | Cheap end | Expensive end |
|---|---|---|
Documentation | A public API with a sandbox and examples | A system with no API, where the work is finding out what it will accept |
Authentication | Standard OAuth, one account | Per-customer credentials, certificates, an approval process |
Data shape | Their records map to yours | Their "customer" is three of your things, or half of one |
Rate limits | Generous or none | Xero allows 5,000 calls a day per organisation and 60 a minute, and answers with an error when you exceed them, so a busy sync has to be designed around the ceiling |
Compliance | None beyond a password | |
Number of systems | One | Seven or more, all of which have to agree with each other |
That last row is what a real platform looks like. PatientGo, the clinical trial platform we have built and run for more than five years, has more than seven external systems plumbed into one place and runs in 36 countries. The patient-facing features are the part people see. The part that took the years is everything the features have to agree with.
Is API integration a one-off cost or an ongoing one?
Ongoing, and the last eighteen months have been a lesson in why. On 2 March 2026 Xero retired its revenue-share model for connected apps and replaced it with five tiers carrying monthly fees, metered data egress, a certification requirement and an annual security assessment at the top tiers. Its changelog shows an API decommissioned in May 2026 and a permissions model replaced in August. Intuit launched a tiered partner programme with variable API fees for QuickBooks in July 2025. Google Maps removed its $200 monthly credit and moved three of its most-used APIs to legacy status on 1 March 2025. Twitter shut its old API tiers in 2023 with 30 days' notice and a $100-a-month floor.
None of those changes were requested by the businesses depending on them, and every one created work. Postman's 2025 report found only 26 per cent of API teams use semantic versioning, the convention that tells you whether a change will break you, so most of the changes arrive without a warning label.
Our published support figure is 15 to 20 per cent of the build cost a year, and integrations are the part of it that moves. A screen you built in 2024 still works. A connection to a platform that re-priced in 2026 needs someone to read the notice, work out what changed and fix it before the invoices stop syncing.
Which integrations are worth paying for?
The ones with a legal or a money reason behind them, in the first phase, and almost nothing else. HMRC's Making Tax Digital for Income Tax applies to sole traders and landlords over £50,000 from 6 April 2026, £30,000 from 2027 and £20,000 from 2028, and any software touching that data has to pass HMRC's approval process and send legally required headers on every call. Payments are the other: build it so card details never touch your servers and you stay in the lightest PCI self-assessment; take them yourself and Stripe's own guide says you may face more than 300 security controls and an external audit.
The rest can usually wait. In a recent scope for a property business we kept the maintenance workflow standing on its own in phase one: no money moves through it, so there is no accounting integration on day one, and the property and accounting systems the office already runs stay exactly where they are. That one decision took the accounting integration, and its risk, out of the first phase. It is the kind of call that gets made in discovery, by listing every system the new one has to talk to and asking, for each, what breaks if it does not on day one.
Every brief starts with a feature list. The price comes from the systems list, so write that one first. For how the rest of the number breaks down, see what moves an app quote up or down and, on the timing side, why the build is not what slows you down.