Should You Fix Data Before Launching AI Projects?

Evolving Data Quality for AI

• 💡 The discussion explores whether perfect data is a prerequisite for launching AI projects, especially with the rise of generative AI.
• 🎯 While traditional AI strictly adhered to the "garbage in, garbage out" principle, generative AI models demonstrate more flexibility with data formats.
• 🔑 The core idea is that generative AI can be less strict about data format, but the correctness of the data points remains crucial.

The Challenge of Perfect Data

• ⚠️ A common barrier to AI adoption is the belief that data must be perfectly clean, often leading to lengthy and delayed data cleaning initiatives.
• 🌊 Data is viewed as a living asset, and continuous cleaning without addressing underlying issues like data governance is likened to bailing water from a leaky boat.
• 🌱 It's advised not to wait for data perfection to launch AI projects, but rather to start, iterate, and improve quality over time.

Generative AI's Impact on Data

• 🧠 Generative AI models are more creative and can complete incomplete data, changing the paradigm of data quality requirements.
• 🔍 The definition of data quality for AI is broader, emphasizing context and whether data is "fit for purpose" rather than just traditional metrics.
• 🛠️ AI itself can be leveraged to clean and prepare data for other AI use cases, such as flagging and profiling data.

Foundations and Practical Steps

• 🚀 There are no shortcuts; building strong data foundations through audits and governance is essential for sustainable AI success.
• ✅ Stakeholder trust is paramount, and unreliable data or models are a primary reason for project failure, highlighting the need for correct and contextual data.
• 📈 To start, focus on a specific use case, audit the relevant data, define a preparation plan, and then start small and iterate, linking data efforts to business KPIs.