A white elephant is something that costs more to keep than it is worth, and is hard to get rid of once you have it.
I watched a team spend more than a year building a dashboard that put project, compliance and sales information in one place for executive leadership. People needed training before they could use it.
It still has not reached the people it was built for.
I built something better in a week. It had no screen to learn, and it sent each person a short message about their own work.
Why most dashboards go unused
Most teams build the same way. Gather the data and put it on a screen. Train people to go find what they need, then add an AI chatbot at the end.
The first step is right, because the collection of information has to be automated. It is also where most projects stop.
A dashboard asks a busy person to come to it, work out what matters, and then go somewhere else to act. Each step costs time, so most people skip the whole thing.
Gartner and BARC figures, reported by TechTarget, show business intelligence tools reach 25 to 35 percent of the workforce. As of 2023, that share had not grown.
Putting AI on top has not fixed it. Nielsen Norman Group’s Kate Moran wrote that companies “rushed into implementing useless AI features and chatbots.” In one of its 2026 studies, a participant said they would never have noticed the site’s AI chatbot without being told: “I just thought that was just a graphic.”
A dashboard shows what the system knows. The person wants to know what to do today.
Start with intent
AI work starts from a different place. It starts with intent.
Engineers are trained to describe every step of a task so it can become code. That is a real skill, and it is the wrong starting point here. With AI you describe the outcome you want and the boundaries the system must stay inside. Jakob Nielsen put it simply: with AI, “the user tells the computer what outcome they want.”
That shift favors a different kind of thinker. Call them the idea people. When one of them asks how to do a ten hour job in eight hours, they are already writing an AI requirement.
To that person, automation is only the first step. The rest of the requirement sounds like this.
- Show me only what I need to act on, when I need it, where I already work.
- Tell me what to do and why.
- When I answer, go do it, and learn from my answers.
Before anyone designs a screen, I work out who needs to act on this data and what each of them needs to see this week. Then I work out where they already work and what should happen when they answer.
Only after that do I ask whether anyone needs a dashboard. Dashboards are interesting, but mostly useless. Focused intelligence, delivered to the person who needs it, is the present.
One message, twice a week
I have seen this work.
In one services organization, a single system produced dozens of reports. Almost none of them changed what a project lead did that week.
So I replaced them with one message, twice a week. Each lead received only the items they needed to handle, in order, with the reason attached. Nothing else.
The hard part was deciding what to leave out. Every item had to pass one test: this lead would do something different this week because of it. Anything else stayed off the message.
It started with a few dozen leads. Today it reaches about 100, nearly every project lead in the organization. Many of them report to other managers. That has not mattered. A short message about your own work, with the reason attached, needs no line authority.
Leads reply to the message, and the replies go into a running record. Resolved items drop off, so the next message carries only what is still open. Every reply is on record.
The objection: alert fatigue
Alert fatigue is real. A 2006 review in the Journal of the American Medical Informatics Association found clinicians override drug safety alerts in 49 to 96 percent of cases. Microsoft found that the 20 percent of workers who receive the most pings get one about every two minutes during the workday, 275 in a day.
I concede that people ignore most alerts. I still insist the fix is fewer and better messages, delivered where people already work.
An alert gets ignored when it is noise, or when acting on it means opening three other systems. Imagine the opposite. When a message arrives, you know it matters, because it is about your work and nothing else. And you can act on it right there, inside the message.
That is the intent. Replace the constant, mostly ignored alerts with a few that fit the person and can be acted on in place. Leadership should get a clearer picture of the business than it has had before.
One condition
There is one condition. What you push has to be right. A 2024 study in Radiology found that physicians’ accuracy fell to about a quarter when the AI’s advice was wrong. People trust what arrives, so the check on what gets sent matters as much as the model.
That dashboard took a year of work and a training program, and it still has not reached its users.
The next time someone proposes a dashboard, ask one question before the first screen is designed.
Why a dashboard, and not intelligence in the flow of work?