Almost 2 years into “AI this. AI that.”, what’s the biggest hurdle to actually using it in your business?

Most of the issues are related to ethical concerns and the potential data flow that may result. We are currently drafting a policy to regulate this new activity, as well as establishing governance and management rules.

Understanding whether it fits the purpose is crucial. Sometimes, requirements are presented as solutions rather than problem statements, making it essential to reframe them as problems to properly assess if AI can address the challenge.
While conducting a POC with mock data is relatively straightforward, real-world scenarios involve different and more complex data, often requiring a hybrid solution to manage both structured and unstructured data. The model must be fine-tuned carefully to avoid overfitting or underfitting, with clear guidelines for tuning.
It's important to recognize that AI isn't a magic solution that can effortlessly solve every challenge without human involvement.

Data quality - Structured and non structured data accuracy and currency
Strategy - Function specific v/s general purpose, buy v/s build
Governance - Avoiding duplication of solutions and ensuring data security & confidentiality

Prioritization - Quantifying business value and estimating costs 

In all AI projects, what I have learnt is that doing a POC or Pilot is easy and exciting but moving to Scale and value realization requires a lot of hard work and perseverance, especially in : 
1. Improving data quality 
2. Finetuning the models 
3. Maintaining domain taxonomy 
4. Creating an effective feedback loop for model output measurement & refinement      

I would say the possibility to scale the solutions from a POC to a full solution. It´s easy to complete POCs with data samples but do you have all necessary data (and data quality) to scale the solution.