Most AI projects fail before the tech rollout even begins

The AI projects I've seen fail, rarely fail for technical reasons.

They fail because of how they're implemented, and that shapes everything that comes after.

Right now, most conversations in this space focus on:

• Model choice

• Training data

• Token costs

• Self-serve configuration

  
  

But in practice, implementation matters more than the technology itself.

I'm starting a series on how to implement AI successfully. This is part one. I break down the 4 implementation models you need to know, and in later articles, I'll show you how to choose the right one for your organisation

The 4 AI Implementation Models: How Much Technical Support and Operational Capacity Do You Really Have?

1️⃣ In-House team

Hire AI engineers, product managers and governance leads. Build orchestration, monitoring and infrastructure internally.

2️⃣ Self-serve platform

Buy a platform and configure internally. Assign an owner. Minimal external involvement. E.g.Using a customer service tool like Intercom or Zendesk AI to deploy an FAQ bot

3️⃣ Tech-led AI partner

Work with major AI vendors or consultancies who bring deep technical expertise, but not deep domain knowledge of your sector. E.g. OpenAI or Palantir

4️⃣ Domain specialist AI Partner

Specialists who combine technical AI capability with deep understanding of your industry’s workflows, regulation, and operational realities. E.g. Snorkl (for regulated lending and collections use cases).

Why this matters more than most people realise

AI isn't a one-time rollout. These systems require ongoing monitoring, refinement, and adaptation, more like managing a team than deploying a SaaS tool.

Most organisations treat it like the latter. That's where things go wrong.

I watched one organisation start with a simple self-serve setup. Seemed sensible at the time. But as operational complexity grew, that single tool evolved into 160 separate workflows to manage, a massive overhead they had no internal capacity to sustain. Self-serve isn’t always the wrong idea. It was the wrong approach for their needs. And by the time that was obvious, they were already buried.

Choosing the wrong technology is costly.

Choosing the wrong implementation model can cost you years of ownership overhead & wasted pilot value.

The next posts in this series will go deeper on each model, when to use it, what it actually costs, and how to know which one fits where you are right now.

Follow Snorkl so you don't miss them, or reach out directly: [email protected]