Embarking on an AI journey within a regulated sector like banking requires a plan that prioritises safety and strategy over speed. The first 30 days of our 90-day roadmap are dedicated to establishing this foundation. This isn't about rapid deployment; it's about building a robust 'capability stack' that integrates tools like Microsoft Copilot into professional workflows safely and effectively. The initial goal is to lay the groundwork for transitioning from manual reporting and analysis to AI-assisted processes, starting with a clear understanding of what we will and will not do.

Before unlocking any AI capabilities, we must first erect the fences. In banking, the cost of a data breach or compliance failure is astronomical. Therefore, we define strict 'no-go' zones where AI application is initially forbidden or heavily restricted. This approach minimises risk and builds a culture of responsible AI use from day one.

The 'human-in-the-loop' principle is particularly crucial. It ensures that while AI can assist and accelerate analysis, the final judgement on matters of credit, compliance, or customer relations remains with a qualified professional. This maintains accountability and aligns with regulatory expectations for oversight.

To enforce these 'no-go' zones, we can't rely on policy alone. We need technical guardrails. This is where tools like Microsoft Purview become essential. Purview allows us to classify and label data at its source. For example, a document can be tagged as 'Internal - PII' or 'Confidential - Financial'.

Once data is labelled, we can configure specific rules. For instance, we can create a policy that prevents any document tagged 'PII' from being processed by an external-facing AI service. This creates a 'walled garden' where our proprietary banking data is kept secure and is never used to train foundation models outside our control. This technical enforcement is the backbone of our data integrity strategy.

This structured approach also helps us address the significant risk of dark data – the vast amount of unstructured and untagged information that organisations collect but rarely use. By beginning a systematic process of classifying data with Purview, we start to bring this dark data into the light, making it either usable for safe analysis or marking it for secure archival or deletion.

To prove the value of AI integration, we need a 'before' picture. During the first 30 days, we establish baseline productivity metrics. We don't just guess; we use tools like Microsoft's Viva Insights to gather anonymised data on how teams currently spend their time. How many hours are spent in meetings? How much time is dedicated to creating reports or responding to routine emails?

These metrics provide a concrete starting point. As we roll out AI pilots in the next phase (days 31-60), we can measure the change. If Copilot helps reduce time spent on drafting reports by 25%, we have a clear, quantifiable return on investment (ROI). This data-driven approach moves the conversation from 'AI is interesting' to 'AI is delivering measurable value', which is critical for securing future investment and scaling adoption.

This initial 30-day phase of planning and erecting guardrails isn't the most glamorous part of the AI journey, but it is the most important. By being deliberate and safety-conscious from the start, we build a foundation of trust that enables faster, more ambitious innovation in the months to come.