How to commission a custom AI system

What custom actually buys you

Off-the-shelf AI gives you the average of everyone’s needs. A custom system is shaped to your data, your workflow, and the specific decision you are trying to improve. The thing you are buying is fit, and fit is usually what generic tools cannot give you.

An MIT report on enterprise AI in 2025 found that around 95 per cent of generative AI pilots showed no measurable return, most often because the tool was bolted on rather than built into how the business actually works. Custom is not about having more AI. It is about having AI that fits.

Start from the outcome, not the model

The model is the last decision, not the first. A custom build that works starts from the result and works backwards.

• Name the decision or task you want to improve, and how you will know it worked.
• Bring your data and constraints early; they shape what is actually possible.
• Decide where a human must stay in the loop before you automate anything.

Keep a human where the decision matters

A custom system should make your people better at their jobs, not quietly take over their judgement. Decide up front which calls the system can make on its own and which it only assists, and design for that line rather than discovering it after launch. This is engineering discipline, not paperwork, and it is far cheaper to get right at the start.

What you should own at the end

The point of commissioning a custom system is that it is yours. A good engagement leaves you holding all of it.

• The running system in production, not a prototype.
• The code, the documentation, and the data behind it.
• A team that can operate and extend it without us.