I’m assuming you’re reading this article as someone who’s working in an (at least) mildly complex organization. That’s basically anything more complex than just doing analytics for yourself.
Every organization that I’ve worked with had a bottleneck in analytics. There could always be more “data people” to generate more “insights”. There are dashboards (many of them rarely looked at), and there are data analysts working on the special questions that demand more than just a few pie charts.
Then AI hit the scene and suddenly you could upload a spreadsheet to ChatGPT and just ask things, or connect your cloud data warehouse to it with 2 clicks.
The early version of this really sucked, but modern AI systems have become extremely capable.
But they’re still no silver bullet. (And probably never will be.)
I believe all three forms – dashboards, AI analysis, and professional data analysts – still have their place. But how much you need each of them depends on your organization.
Your job as a business leader is to map the right task to the right analytical tool.
Here’s how I see them:
Dashboards
Dashboards are great for questions we already know we’ll ask, multiple times. Such as:
How much revenue did we make?
How many new subscribers did we get?
What were the Top 10 most clicked articles?
etc.
That’s why classical BI is and will stay relevant. It is basically standard reporting.
Whenever your metric is well-defined, you want easy access, you need to scan multiple values at the same time, and you’re looking at it consistently, go ahead and build a dashboard.
I think the best comparison is a “real” dashboard in the physical world. When you’re driving a car, you want to see things like your speed and fuel at a glance, and you’ll look at them frequently.
Classic Porsche 911 dashboard considered one of the best of all time
Imagine asking your co-driver every time how fast you’re going. It doesn’t make any sense.
As long as you’re in the driver’s seat, you’ll want a dashboard.
And there’s little benefit in recreating it every day.
Dashboards are excellent at distributing known answers.
AI
AI will increasingly be used for the long tail of follow-up questions.
Just because it makes so much sense.
Imagine your dashboard shows something interesting:
“Huh, that value for sure does not look normal!”
The next logical question is: What happened? Why do I see this value?
The classical 5-Why Analysis was built on exactly that principle.
To get to the “Why” you have two options:
A) You look at the underlying data yourself
B) You ask someone else to look at the data for you
In many organizations with a low level of data literacy, B happens to be the default choice. That’s one reason why data teams are so overworked and the number of “insights” always lags behind the number of questions.
“We’re drowning in data, but starving for knowledge.”
Contrary to common belief, the people making the decisions never really had the time (or motivation?) to build their own reports in Power BI or create pivot tables in Excel. At least that’s what I’ve seen in many organizations. Which is why Self-Service BI often never really delivered on its promise despite massive efforts on upskilling.
AI to the rescue.
Me, using my AI Data Analyst
When done right, modern AI is totally capable of taking a business question and then building the deep-dive for you. It never gets tired of looking at another pivot table, which is why you don’t have to stop after 5 Whys. You can easily ask 10 more.
The trick is to make this process reliable and build AI systems where “I can’t answer that” is a built-in answer.
Which brings me to the last point.
Humans
You still need humans (“data professionals”) for questions where governed datasets are missing or where the analysis itself is uncertain.
Sometimes the real work is deciding:
what data should be used to derive a metric
which method is appropriate
what defines “good” data quality
who’s allowed to see what
Off the top of my head, I could think of the following examples that would be bad cases for both dashboards and AI Data Analysts:
What does the market suggest for our next product launch?
Why did this new campaign fail?
How can we track CSAT as a new metric?
Will the new price lead to more sales?
AI can support all of these analyses. But it shouldn’t be in the lead for them.
Because if the underlying meaning or method is still being figured out, you have to do this job first.
You can delegate thinking, but you can’t delegate understanding.
AI is strong when the analytical lane is known. Humans are needed when the lane itself has yet to be defined.
A simple decision framework
This has absolutely no ambition to be academically correct, but if I had to break it down into a short table, I’d recommend mapping analytical tasks like this:
Situation | Best fit |
|---|---|
Same question, repeatedly | Dashboard |
Different questions, but same data and methods | AI Data Analyst |
New question that requires new data and/or new methods | Human analyst (+ AI) |
The more stable the question, the more it belongs to traditional BI.
The more variable the question becomes, the more interesting AI gets – until the question moves outside of what’s available in the governed data and methodology space.
That’s the consequential hand-off to a professional (human).
So, can you dump your dashboards?
Absolutely not. But you may need fewer of them.
There’s absolutely no point anymore in bothering your analytics team to create endless variations and pivots of the same data.
I’d even argue that in many cases, you don’t need another set of drop-downs and switches in your dashboard either.
AI is perfectly capable of building many of these variations for you on the fly.
AI may not kill BI.
But it could kill a lot of dashboards we never really needed.
See you next Saturday,
Tobias