Generative AI (GenAI) is promising to reshape how we work and innovate. Yet, many organisations hesitate, wondering, “Is our data perfect enough?” However, perfection often isn’t the prerequisite you think it is.
What is it?
The Data-GenAI connection – think of Generative AI models as incredibly smart apprentices: their skills are shaped by the data they learn from or the data you provide it in a prompt. This data, be it text, images, or code, forms GenAI’s understanding and ability to generate fresh, original outputs.
So, yes, the quality, relevance, and context of your data matter. But here’s the nuance: “quality” isn’t a rigid, one-size-fits-all benchmark. The necessary data standard flexes with the specific job you’re giving your GenAI.
What does it mean from a business perspective?
Match data quality to the task – there’s a common misconception that enterprise data must be flawless before you can tap into GenAI. In my experience, the reality is more practical: the quality of data should match the nature and risk of the task – match data quality to risk level.
- High-Stakes Decisions Demand High-Quality Data: If GenAI is informing critical decisions with significant consequences – think medical diagnoses, financial lending approvals, or safety-critical engineering – then absolutely, your data must be impeccably accurate, complete, and verifiable. The margin for error is minimal.
- “Good Enough” for Exploration & Creativity: Conversely, if you’re using GenAI for brainstorming campaign ideas, drafting initial content, summarising documents for a quick understanding, exploring new concepts, or looking to reduce ambiguity in a complex decision, the data often just needs to be “good enough” to inform, not decide. It means the data is fit for that specific purpose, sparking ideas even if not pristine.
- Context Can Trump Perfection (Sometimes!): GenAI can be surprisingly adept with less-than-perfect data if the context is rich. For example, slightly messy CRM data might still unearth useful slogans for marketing. However, for something like an AI-assisted loan application, the data must be exact. GenAI thrives on context, and sometimes a “disorganised SharePoint folder” has enough of it to get you started on lower-risk tasks (I have such a SharePoint folder and an Agent that helps me with research and product development – it works).
Tailor Data Needs to GenAI’s Role – Risk-Based Framing:
- As a Co-pilot (Content Generation): For drafting emails or reports, moderately curated data works well, with humans validating.
- As a Sparring Partner (Exploration/Ambiguity Reduction): To sift through research or brainstorm, GenAI can leverage varied, even unstructured, data sources to help you refine your understanding.
- As an Automated Agent (Decision Making): Here, data precision and completeness are non-negotiable. The more autonomy the AI has, the higher the data bar.
What do I do with it?
Your pragmatic path to GenAI use is to move past the “perfect data” hurdle. Here are some concrete steps:
- Start with Your ‘Why’: Pinpoint the business problem or opportunity. Your specific use case helps you determine your data quality needs.
- Assess Your Data As-Is: Look at what you have and where it lives. For initial exploratory tasks, your current data (even if in “messy SharePoint folders”) might be good enough to get started.
- Define ‘Good Enough’ for This Task: For your pilot projects, what level of data quality will deliver value? Is it absolute accuracy, or is it about capturing broad themes and relevant language to inform, not decide?
- Pilot, Learn, Iterate: Kick off with lower-risk, exploratory projects. This is your chance to test GenAI with existing data, learn, and then improve data processes as you eye more critical applications.
- Value Context, Not Just Structure: Don’t assume you need immaculate relational databases for every GenAI venture. The rich contextual information in documents, emails, and internal wikis can be incredibly revealing.
Getting started with GenAI doesn’t always demand a perfectly polished data landscape from day one. A pragmatic approach, matching data quality to the specific task and its associated risks, is often the smartest way to begin. This mindset lowers the adoption barrier, allowing you to start tapping into GenAI’s incredible potential much sooner.