How to enable your AI-assisted Analysts with Microsoft Copilot

Generative AI has taken the world by storm and it was just a matter of time before it was introduced into data visualisation tools, giving analysts a new way to create and edit outputs.

There are tools released or in development by all the major analytics players, Microsoft, Tableau, Zoho and Qlik which represent early attempts to transform and enhance the workflow of data analysts and visualisers. This particular application of generative AI technology differs from previous iterations which focused on interpreting outputs (think Power BI QA and Tableau Pulse), but has received little attention, even as AI captures headlines across the board.

Off the back of a recent customer project where we've been exploring potential use cases for Copilot for Power BI, we’re taking a closer look at some of the features we believe will be standout for the AI-assisted Analyst.

Features that we love

Copilot can eliminate the initial report-building step by providing instant dashboard outputs. In this helpful video, several new ways to generate reports are demonstrated.

One of the standout features, 'Suggest Content', leverages available columns and data types to offer prompts that help manipulate the data. The more descriptive the columns and consistent the data, the better the suggested prompts. Alternatively, users can write their own prompts, similar to how they would interact with traditional LLMs.

The reports that Copilot generates act as an excellent starting point, giving the AI-assisted analyst a quick method of visualising a dataset in a concise, one-page summary. While the visuals aren’t the most varied or complex, the feature holds significant potential for increasing efficiency by allowing more time for enhancing the output and developing complex functionality.

Another impressive feature is Narrative Visual, which can be applied to a report page. This provides key insights highlighted and dynamically changing based on the user's selection, offering a dynamic, auto-updating summary as filters, slicers, and data refreshes are applied.

For analysts tasked with retrieving more complex data from their underlying models, Copilot simplifies DAX query creation. The AI-assisted analyst can describe what they need from the semantic model, and Copilot will generate the corresponding DAX query. It can also explain what a particular DAX query is doing and aid in understanding DAX concepts by explaining specific functions or suggesting the appropriate DAX function for a given task.

These features combine to make Copilot a powerful tool for analysts, providing a quick and efficient way to generate initial visualisations, gain insights through dynamic narratives, and effortlessly write complex queries with the help of AI.

Limitations to consider

Like any emerging technology, Copilot has its limitations that analysts should be aware of.

Firstly, the accuracy of AI-generated outputs is fundamentally tied to the quality of the data used. While the data we tend to use in proof-of-concepts or demonstrations often showcases Copilot’s capabilities, its effectiveness needs to be thoroughly tested with an organisation’s production data to ensure accurate interpretation and relevant insights.

Continuing on the theme of rubbish-in, rubbish-out, creating effective prompts is crucial for Copilot to deliver meaningful results. Vague or overly broad prompts can lead to less useful outputs, and some visuals may be simpler to design manually than to describe through precise prompts that Copilot understands.

Lastly on our list of limitations is one to consider as part of integration into existing workflows and ROI calculations; the ongoing need for human review. Despite Copilot's advanced capabilities, human oversight is still necessary to verify that the correct fields, calculations, and rules are applied. A routine review ensures the generated outputs are accurate and aligned with business requirements.

Helpfully, Microsoft understand and accept the limitations of their tools and encourage their safe and transparent use in this guidance.

How can organisations get started?

We are working with several large organisations to help them trial Copilot within their teams. Return on investment (ROI) is a critical question and often challenging to quantify with emerging technologies. However, there is growing evidence of the productivity benefits that AI assistants can deliver, making this our suggested starting point.

To get started, it's crucial to map out your organisation's existing analytical processes and identify high-volume, repetitive tasks that Copilot is well-suited to handle. By establishing a thorough understanding of current workflows, particularly focusing on high-repeat tasks, organisations can then quantify the time currently spent on these tasks to establish a baseline. Key metrics might include the time it takes to create standard visualisations, the frequency of human errors that Copilot could minimise, and the time required to generate useful insights using current workflows.

After implementing Copilot, it's important to monitor and measure the reduction in time and errors, as well as improvements in insight generation. By tracking time saved post-implementation, organisations can assess improvements in efficiency, whether through reduced time-to-create visualisations, fewer human errors, or a faster time-to-insight with the generated narrative.

However, challenges remain as these tools are still new and the landscape is shifting rapidly, with new features being released monthly. Even Microsoft is learning and evolving its tools, meaning organisations should remain agile and adaptable.

An early adopter's primary objective should be to understand their workforce and specific needs, enabling them to quantify potential benefits when integrating Copilot into their analytics function. Only after this understanding has been developed should organisations proceed to run repeatable trials to generate user insights and feedback. This iterative approach will help refine the implementation strategy.

Overall, our view is that AI assistants like Copilot hold immense potential. However, due diligence is essential to ensure their adoption is safe, ethical, and inclusive of the workforce. Taking a thoughtful, measured approach will maximise the benefits while ensuring a smooth transition for teams.

Are you interested in the future of the AI-assisted Analyst?

If you or your analytical community are interested in learning more about how to get started with Copilot in Power BI, on your journey to develop AI-assisted Analysts, get in touch with us and we’d love to hear your ideas!