Let AI Coworkers get that Power BI report for you

Everyone wants reports but no one has the time to build it.

Despite the data being easily accessible, powerful tools like Power BI being available and the requester knowing exactly what they want to see (service requests by status, how change backlogs are moving, the ROI of a change), IT workloads don’t allow for the quiet time that is required for a powerful report to be built.

Inevitably, the request languishes in a queue till it’s escalated multiple times. Then, someone takes time out on a weekend or a holiday to build the report.

I’ve heard this many times from our prospects but the scale of this problem struck me only when I saw a thread on Reddit by a ServiceNow admin who was trying to get their data from ServiceNow into Power BI. On top of running IT more or less singlehandedly, this admin was also trying to learn DAX on top of everything else they were responsible for.

For days after this, I kept thinking about this because you can be completely good at your job and still feel underwater when a reporting request quietly adds “become a BI developer” to your list.

This is exactly the kind of problem that AI Coworkers can take over for an IT team. Let me show you how.

What I set up

Atomicwork analytics data can already flow into Power BI on a scheduled refresh, thanks to the integration. That part matters a lot because step one of being able to quickly prepare a report is: having the data. A healthy pipeline does nothing for the person who needs incidents broken down by assignee for a meeting tomorrow morning.

Once I ensured that the data was flowing as intended (all puns also intended), I configured a Power BI Coworker — an AI Coworker with instructions and tool access scoped to that existing data, and a job description that amounted to: take a request written in plain English and build the report.

This means that anyone could raise a request for a report through the Atomicwork portal. This moves reports out of the realm of just leadership and empowers everyone in the team to access data to make decisions. All they had to do was create a request through Atom.

I recorded some of the sample test runs of this AI Coworker: a service requests report and an incident report. The links went straight into Power BI — charts and breakdowns by status, assignee, request type and priority, already laid out. Nobody had to leave their to-dos to arrange these visuals by hand for this request; the AI Coworker took care of it!

Building custom reports

When I asked it to build me a non-standard report, a cloud infrastructure dashboard, it didn’t guess or produce something generic that would be utterly useless (as most AI Agents do). Just like a human analyst would have, the AI Coworker asked me a list of questions including which metrics and what data sources should it consider for the dashboard.

That is the right behaviour, and it is also the clearest illustration I have seen of what actually separates an AI Coworker from a workflow. A workflow fires, executes the step it was configured for, and exits. If the step does not fit the situation, you get a bad output or an error.

An AI Coworker has a goal so when the goal cannot be met with what is in front of it, the sensible move is the same one a human analyst would make — ask. If you handed that same five-part request to someone on your BI team, you would expect a conversation about what “a cloud infrastructure dashboard” actually means to you. You would be a little suspicious, in fact, if you did not get one.

In an earlier run, it also flagged that some metrics were not available and had been left out. This does not seem that useful until you’re being questioned about your metrics in a meeting.

What still needs you

I’d hate for you to think that all of this happened because of an AI magic wand.

There is real setup behind all of this. The data has to be connected and fresh (a refresh schedule is key for this AI Coworker). The Coworker needs good instructions so that it understands what matters to you and tools scoped to what it should be allowed to touch — which is the part you govern, not the part you hope works out. And someone on your team still has to look at the result and decide whether it answers the business question that was actually being asked, which is a different question from whether the chart renders correctly.

None of that goes away. What changes is how and where the request starts.

Instead of landing in a queue and waiting for somebody to find three uninterrupted hours, it starts as a description of what someone needs, produces a first draft in Power BI, and moves into a conversation about the follow-up questions. The work your team does shifts from assembling the report to judging it — which is the part that needed them in the first place. (This is the same shift happening across ITSM analytics, not just reporting.)

For the stretched IT admin, this is the difference between an AI Agent that produces reports and an AI Coworker with a job role that can offload the work for you entirely and gets better with every request and report.

If you have a reporting backlog that you have quietly stopped apologising for, that is the one I would point an AI Coworker at first. Here is how building and governing one actually works.

Sign up for a demo today →