0%
AI & Automation1 min read

Do we need our own custom-trained AI model, or can we use an off-the-shelf one?

Almost never, and building a custom model is usually the expensive way to solve a problem a well-prompted existing model already handles well. We reach for a genuinely custom-trained model only in the small minority of cases where off-the-shelf models demonstrably cannot do the job — highly specialised domain language, for instance. If someone is proposing a custom model as the first option rather than the last, it is worth asking why.

Short answer

Almost never, and building a custom model is usually the expensive way to solve a problem a well-prompted existing model already handles well. We reach for a genuinely custom-trained model only in the small minority of cases where off-the-shelf models demonstrably cannot do the job — highly specialised domain language, for instance. If someone is proposing a custom model as the first option rather than the last, it is worth asking why.

Almost never, and building a custom model is usually the expensive way to solve a problem a well-prompted existing model already handles well. We reach for a genuinely custom-trained model only in the small minority of cases where off-the-shelf models demonstrably cannot do the job — highly specialised domain language, for instance. If someone is proposing a custom model as the first option rather than the last, it is worth asking why.

Keep exploring

Related faqs

View all →
AI & AutomationAug 12, 2026

Do you build genuine AI features, or just bolt on a chatbot?

Where AI genuinely helps — summarising long documents, classifying support tickets, powering a smarter search, drafting a first version of routine copy — yes, and we build it properly. Where it is being requested because AI is simply expected to appear somewhere on the roadmap, we push back and say so directly. An AI feature that is confidently wrong ten percent of the time can cost more customer trust than it saves in staff time, so it has to earn its place like any other feature.
AI & AutomationAug 12, 2026

What does an AI or automation project typically cost?

Most automation work is priced closer to a focused software feature than a full platform, because it usually plugs into systems you already have rather than replacing them. A single well-defined automation — routing enquiries, generating a report, summarising submissions — is a modest, contained project. Something that touches multiple systems and needs ongoing model tuning costs more and needs a proper scoping conversation, which is exactly where we would start rather than quoting blind.
From article to action

Have a digital problem worth solving?

Bring us the context. We will help you find the clearest next move.

Start a conversation→