The Real Costs of AI Projects: Beyond the Proof of Concept

After the AI Proof of Concept is working, the majority of the code would be written on 'infrastructure integration':

• Data and model pipelines
• API integrations
• Security layers
• Cron jobs
• Quality automatic checks
• etc.

Most people are surprised when we say that.

The "Glue Code" Phenomenon

To be funny, we can call it glue code. In practice, the breakdown costs of an AI development project has only 5% of 'fancy' Deep learning software, 95% of all the rest.

Ongoing Costs

Once the solution is live:

• 10-15% of the total development costs are going to be in maintenance to keep the solution running:
  • Retraining
  • Changes/improvements of the model
  • Data additions
  • etc.
• 4-8% in AI/data governance costs

Calculating ROI

If you are calculating your Return on Investment (ROI) of an AI project, keep these numbers in mind:

• Initial development: 100%
• Ongoing maintenance: 10-15% of initial development cost per year
• Governance: 4-8% of initial development cost per year

Remember: The success of an AI project often depends more on the robustness of its infrastructure than on the sophistication of its algorithms.