Advanced analysis: multi-step analytical workflows

1. Advanced analysis: multi-step analytical workflows

Until now we've stayed inside Excel's normal workflows. We've just made them faster and a lot less tedious.

2. Beyond formulas

But what if we want to push these capabilities to understand how spend, revenue, units, and region all correlate? This sort of analysis takes multiple steps, with calculations, visualizations, and careful validation of each step. There's often a lot of manual work here, and many people switch to coding tools to avoid this. Or imagine we have hundreds of free-text survey responses where we need to identify the top themes—no Excel formula can do that. These scenarios need multi-step exploration, and they're where Copilot goes past what a formula can do.

3. How Copilot performs multi-step analysis

If we gave Copilot a prompt like: "Investigate the correlation between the variables in the workbook." Copilot reasons through the data available, plans the series of steps it will take to conduct the analysis. Then, as long as the mode is set to Allow editing, it will start editing your workbook to create any supporting tables and visualizations to help you make a conclusion. Because everything is done step-by-step, you can inspect every step of the analysis, leaving nothing to chance. And remember, if you don't want Copilot to edit your workbook directly, switch the mode to Chat only. Let's test this out!

4. Example: Investigating correlation

Our workbook is an employee pulse survey. Fifty responses, and for each one a region, a role, a salary, a tenure, an engagement score and a written comment. We open Copilot and ask how salary relates to role and region.

5. Example: Investigating correlation

Copilot plans the analysis before it starts - here it reasoned in nine steps - and you can open that plan to see each one. Look at where the output went. It built a separate sheet called Salary Analysis, so the survey data itself is untouched.

6. Example: Investigating correlation

Then read the last row. Role and region jointly explain only four point four percent of salary variation, and neither association is statistically significant in this small, uneven sample. Copilot is telling you its own finding is weak. Read that line before you read the averages above it.

7. Example: Summarizing free-text into themes

Now something formulas cannot do. That same survey has a Feedback column of written comments, so we ask Copilot for the top improvement themes, with a count for each.

8. Example: Summarizing free-text into themes

Copilot sorts all fifty comments into a handful of themes and writes out what each one covers, so you can see which comments it would have put in which bucket. That is where your judgement comes in, because the buckets decide the ranking. Copilot has split flexibility and remote work out as a theme of its own, separate from workload. Group those two together and the order changes. Nothing here is wrong, but the ranking you report is a consequence of a choice Copilot made on your behalf, so read the definitions against the original comments before you rely on it.

9. Let's practice!

Time to try out performing multi-step analysis through natural language with Copilot!

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.