Does your dashboard need an AI chat?

The bitter truth.

People rarely look at data to make a decision. More often, they look for confirmation of what they have already decided.

In psychology, this is called confirmation bias. If you have ever picked your favorite hypothesis first and then looked for a chart to support it, you know exactly what I mean.

And this is not some rare edge case. This is how many business meetings actually work.

That's where the main dashboard problem begins.

Teams build them, present them, launch them. And then almost nobody opens them.

And now, against this backdrop, AI agents are showing up. Most often - in the form of chat.

Yandex DataLens is integrating chat. Metabase now has an AI assistant even in the free version. In Google Sheets, you can already work with data through Gemini. YouTube is full of tutorials like: "upload your CSV to Claude and get insights in 5 minutes."

In short, "talking to data" is becoming normal.

Question: will it help?

Not by itself.

If the data doesn't answer real business questions and doesn't lead to decisions, chat just adds one more interface to uselessness.

AI will confidently explain garbage, generate "great hypotheses," and admit mistakes after the third follow-up.

So does that mean chat is useless?

No. Actually, it's needed. More precisely, you need some kind of agentic interface.

It could be:

• chat inside a dashboard;
• a Telegram bot;
• a Slack bot;
• an MCP interface;
• a standalone AI assistant.

The format doesn't matter that much.

What matters is that a person can ask questions in plain language. Without looking for the right tab. Without remembering the metric name. Without manually building filters.

Just ask: "what changed?", "where is the problem?", "what should we check?"

This scenario is becoming more and more expected. And at this point, taking it away from users feels strange. It's almost table stakes.

But just adding chat is not enough.

It needs context around it. Just like any decent dashboard.

First, the user should understand why they are going to the chat in the first place.

That could be:

• suggested questions;
• prompts inside the interface;
• short documentation;
• ready-made analysis scenarios;
• examples of good questions.

Not "open the chat and sit there wondering what to ask the table."

But proper navigation: what questions can be explored here, where the agent is useful, and what decisions it can help with.

Second, something should happen after the answer.

For example:

• pausing a campaign;
• creating a task;
• sending a message to a contractor;
• starting monitoring;
• sending an alert to Slack or Telegram.

Not "chat with the bot and go have a smoke."

But quickly starting the feedback loop: question → answer → decision → action → result check.

Right now, as an external Head of Marketing Analytics, I help teams set up these agents for marketing analytics.

And this is definitely not just "chatting with data."

Together with teams, we design:

• what questions the agent should help ask;
• what answers it should give;
• what actions can be triggered after the answer;
• where it must say "I don't know."

Because the value is not in talking to data.

The value is in something changing after that conversation.

And here we come back to the beginning.

If a team only uses data to confirm its favorite hypothesis, an AI agent will simply speed that up. It will help them find a beautiful explanation for what everyone already wanted to hear.

So yes, AI is needed.

But not as "let's just add chat to the dashboard."

Same old story: the interface is not the point. The system is. A system that helps people ask better questions and turn answers into decisions.

An AI agent is just a new interface for this old job.

If you want to add AI to marketing analytics without circus and hallucinations - DM me.