The Dashboard is Dead

If you haven’t arrived at this conclusion, you will.  If you’ve started transitioning some of your analytics to agents, you’ll know you’ll be here soon.  Reporting is fundamentally different, and 100x better in the AI-era.  

I don't say that to be provocative. I say it because of what we've changed internally at Dremio over the last several quarters.  We are drinking our own champagne here, and with the combination of Claude Code (or CoWork) and Dremio, all of our analytics are agentic, hence the hot market term, “agentic analytics”

Since my early marketing days at Amazon, I’ve always been a proponent of self-service.  AI is making that possible in many realms, even analytics. Today, it is officially true, we have no “marketing analyst” or “BI engineer” on the team. And, we don’t use google sheets or excel.  Everyone uses AI and everyone is an analyst.  The results are amazing, the insights are coming faster and better than ever.  MQL conversion rates have doubled, our database is growing faster than ever, we know the patterns, and Claude knows what we should do to pour gas on what works. 

The BI Era Was a Big Improvement, but Very Painful

Dashboards have become the dominant reporting format because they solve a very specific problem of a very specific era: most business users couldn't query data themselves, so someone had to bake the data knowledge, semantics, and the queries into a report. 

The process was a business owner talking to a BI Engineer or analyst, explaining what they wanted, waiting 4-6 weeks, getting back not what they wanted, and then entering a cycle again. Not fun. It brings back poor memories just writing it.  Once the dashboard was working, business definitions may have changed, priorities may have shifted, and so the dashboards were of limited value, even after all of that work.  The purpose of the dashboard was to lead to taking action that improved the operation of the business (efficiency, velocity, etc ...).  

Very little action was taken off of dashboards because dashboards told you the “what,” but never the “why”, or “what to do.” It would take a month for an analyst to put together separate projects, all managed by different business owners, to develop any actionable insights. So unless the problem was really bad, which meant teams would put in this work for a deep dive, you went into weekly meetings, looked and numbers, and said things like, “we’re off 20% this week, what’s going on” and teams would say “we’re not sure,” and then it would roll into the next week without insight or action, because of prioritization.    

Dremio + Claude Code Changed the Game

A few quarters ago, we made a deliberate decision to rethink how our go-to-market team runs its weekly reporting and QBRs. We stopped organizing reviews around a fixed set of dashboard tiles and started organizing them around determining the actions we needed to take, even if those actions were different from what was in the existing plan.

The shift sounds simple. It is not.

It starts with business level ownership.  You own your function, you own your numbers, you own your reporting.  There is no “I’m waiting on the analyst,” or "I'm not sure.”  There are no excuses.

Next, your reports need to cover the “what”, but also the “why,” and the “what to do.”  Our ops reviews also contains current priority projects and status, so we can make adjustments based on new learnings each week.  As you’ll see, reports are now just .pdfs or docs based on Claude SKILLS. It takes minutes to get to “what to do” vs. the 8-10 week cycle above. If our new process is 10 minutes, that is literally 8,000x faster than the old model.  

In addition, there might be deep dive questions that I’ll have, or that others will have during the reviews.  We answer these questions live, to determine if we should make a change or not. That was never possible before.

In this model we have lowered costs and massively improved time-to-action. Below are examples (not real data) of what our reports look like for every function.

Database Marketing - yes email still matters

How do you get there?

The Two Non-Negotiables: All Data and a Semantic Layer

First: you need to bring your data together.  If you’re in marketing, you’re probably well aware of the 15ish data sources most of us deal with, if not, here’s a sample of the list: Marketo or Hubspot, SFDC, web analytics like Google Analytics, LinkedIn Ads, 6Sense, product data, etc …

Two technologies are your friends here, Fivetran and Dremio. If the data is already in a data lake, you're set; and if it’s in databases, a warehouse and a lake you're also set. For the most part, Dremio can connect to the data source, and part one is done.  For some sources, such as Linkedin ads and Google Analytics, you’ll need to drop those into Dremio and Fivetran makes that simple as a single click.  

Second: you need to provide agents with the context they need to give you accurate answers.

In marketing, we are highly reliant on a single view of a customer (often across 12-15 IDs), related to accounts, related to opportunities, and on top of that we have a large set of business metrics, from lead to opportunity to revenue. There is a lot we know about our business and our domain, and AI needs all of this information to do a great job.  

All of this information lives in a semantic layer; within Dremio you have one semantic layer across all of the data, which is different from other data platforms, and critical for AI.  We also make it simple to use AI to help build the data model that is going to give you consistent answers and new insights.

Pull up Claude Code and use the Dremio CLI (don’t worry, Claude Code figures this all out), then start describing what you need. Another short cut can be taking your existing reports, and feeding them to Claude Code and it starts to figure out your key metrics and how to calculate them.  When you’re done it will be easy for AI to know your fiscal calendar, what an MQL means and how to calculate it, how to attribute value to marketing, and so on.

With your data pulled together, and context delivered by Agents for Agents, you’re ready to rock and roll.  This takes a few hours to do well.  If you’re interested in a workshop on how to do this, I’ll lead one, just go to Dremio and Contact Us.  This isn’t a marketing program, but well, maybe it should be.  

SKILL.md If you don’t yet know how to create and use SKILL.md, you need to learn, fast. Skills are repeatable items that make it simple for an agent to create a desired output over and over again. As you work with a skill, such as "social-media-report," when you're done, you might work with the agent to improve the output. When you're finished, you simply say "improve the skill," and it will, so your AI can get better and better as you work with your agent on that task.

In this case, work with AI, tell it the format you want, tell it you want “X KPIs” and you want to track weekly progress.  With the metrics set, you want commentary on what happened for the week. Then it gets really interesting, it will tell me why it happened and what to do about it. These are high reasoning, complex tasks, and I recommend using high context models like Opus 4.6. When you have the report where you want it, just say, “Create a skill for generating this report.”

The output is really your choice, you can go simple with .pdfs or Google Docs. You can create interactive reports which are html, or streamlit apps, and these just run on your local machine. If you want a hosted, sharable app, you can throw it in Vercel, but watch permissions. It isn’t something I do, but you could. I like simple and put it in a Google Doc and share with whoever needs it. All of the information is there, and everyone gets the information. 

Schedule it with Claude Code.  Just tell it, data closes at midnight on Sunday so early Monday, create the report. Done. You want it daily? No problem. Most business reporting isn’t realtime or hourly, that is more system monitoring, so we’re not solving for that in this article. 

Aiiiiiiight 

It’s time to ditch the dashboards and all the muck and pain associated and move to agentic analytics.  It’ll take a few hours and change management on your team, but it is a surprisingly simple lift.  Typically there are a few bumps with a metric definition that is hidden in a report and no one knows how it was calculated, but those get worked through in a few weeks.

Good luck, you won’t need it.

Read Maloney is the CMO of Dremio, the Agentic Lakehouse — the only Iceberg-native data platform built for agents and managed by agents. Before Dremio, Read led marketing at AWS, Oracle, and H2O.ai, and served as a Captain in the United States Marine Corps.