AI automation agency vs in-house build: which should a UK SME choose?

The short answer

The deciding question is not skill or cost; it is how much continuous work there is. If software is central to what you sell and changes every week, build an in-house team — eventually. If you need a defined system built, connected and then kept running, an agency usually delivers it faster and with less risk, because it brings every skill the work needs without you hiring each one.

Most SMEs are in the second position more often than they think. And there is a third answer that neither an agency nor a recruiter will volunteer: sometimes the right move is to buy an off-the-shelf product and build nothing.

What a production system actually needs

A working business system is not just code. Over its life it needs:

• Analysis — understanding how the work is really done before anything is designed.
• Design — screens people can use without training manuals.
• Engineering — the application itself, front and back.
• Data and migration — moving what you have into the new system without losing or corrupting it — see migration.
• Testing — so changes do not break what already works.
• Security and access — who can see and change what — see access control.
• Hosting and support — keeping it running, backed up and patched — see hosting and support.

One person is rarely strong in all of these. An agency's main advantage is not cheaper hours; it is breadth — each part handled by someone who does that part well.

The key-person arithmetic

A business system runs all year: 52 × 168 = 8,736 hours. One full-time developer, after 5.6 weeks of statutory leave, works about 37.5 × 46.4 ≈ 1,740 contracted hours.

1,740 ÷ 8,736 ≈ 19.9%. A single in-house developer is available for about one fifth of the hours the system is running — and completely absent for 5.6 weeks a year.

Two developers help, but less than people expect. If their leave never overlaps, there are still 11.2 weeks a year — about 21.5% of weeks — when only one of them is there. And if one leaves, much of what they knew leaves too, unless it was written down. This is the same coverage arithmetic behind AI receptionist versus human receptionist, applied to the people who keep the systems running.

When in-house wins

• Software is the product, or a large part of what makes you different, and it changes continuously.
• There is enough work to keep a small team busy all year, not just during projects.
• You can hire at least two, so holidays and departures do not stop everything, and someone senior enough to set standards.
• You can manage engineers. Someone inside the business can judge the quality of the work, not just whether it looks finished.

When an agency wins

• The work comes in projects, with quiet periods between them that an in-house team would struggle to fill.
• You need the breadth — design, data, security, hosting — for a period, not permanently.
• Speed matters more than building internal capability right now.
• You would rather manage an outcome than manage engineers.

The risk with an agency is dependency. It is managed by ownership and documentation, set out below — not by avoiding agencies.

The hybrid most SMEs end up with

In practice the strongest arrangement is often a mix: an agency designs, builds and hosts the system, and a named person inside the business owns it — decides priorities, accepts changes, and knows how it works. That person does not need to write code. They need to understand the business process well enough to say whether the system is right.

Over time, if the workload grows, that owner is the natural first member of an in-house team, and the agency's documentation is what lets the team take over without starting again.

What to insist on, whichever you choose

1. You own the code and the data, and both live in accounts you control.
2. It is documented and tested, so someone new can pick it up.
3. You can export all of your data in a usable form, whenever you like.
4. Hosting is in your name, or can be moved to it without the builder's permission.
5. There is a named person, internally or at the agency, accountable for it working.
6. Every automated action is logged — see audit trail — especially where AI employees are involved.

And before either route, check whether you need to build at all. Many problems are solved by configuring or connecting what you already own, as we set out in bespoke CRM vs off-the-shelf and what integration actually involves. If a project does not need AI, a good builder should say so — see what the AI terms actually mean.