Between the two of us, we've spent more than 15 years working across data foundations, analytics, visualization, and AI/ML. We've built the dashboards. We’ve watched dashboards multiply across teams, functions, and leadership meetings. And we have learned one unavoidable truth: dashboards always lead to more dashboards.
It always starts similarly. Build an executive dashboard with the three to four most important metrics. Then someone asks to drill deeper into one of them. Then another team wants its own version. Then you add secondary metrics, filters, custom views, and user groups. Before long, you have a library of beautiful dashboards that nobody can navigate. If people can't find what they need, it might as well not exist.
Well, you better build another dashboard…
There is a better way though, and it's already here: agentic analytics.
Yes, this is another blog about AI. Before you rage-click away, hear us out. We're not going to tell you about a trend to watch. We're going to talk about something already changing how organizations work with data, what it takes to be ready for it, and ask you honestly whether you're ready or falling behind.
# The Dashboard Model Is Already Changing
After attending the recent Tableau Conference, it was hard to miss how quickly the dashboard model is changing.
Major enterprise platforms are all moving toward the same experience. Databricks calls it Genie. Microsoft calls it Copilot in Fabric. Salesforce calls it Einstein or Tableau Pulse. The names are different, but the direction is similar: a business user types a question in plain English, and an agent finds the answer and returns a value, a table, a chart, or an explanation without requiring a dashboard, BI ticket, or days of waiting.
In one live demo at the conference, a Slackbot connected directly to a Tableau MCP server. The user typed a question and an agent returned a chart in seconds. No portal. No navigation menu. Just an answer, delivered where the person was already working.
This is not a prototype. Organizations are already investing in the data infrastructure required to power tools like Databricks Genie, Tableau Pulse, and other agentic analytics capabilities. Some are also restructuring analytics and visualization teams and, in some cases, cutting roles as they reevaluate where BI analyst time creates the most value.
Ultimately, BI analysts aren’t going away, but the version of the job built around dashboard production is getting harder to defend.
For organizations that spent the last few years building clean, solid data foundations, this shift creates a new opportunity. They are better positioned to turn on agentic tools and get real results right away. Organizations that focused on building more dashboards may have more foundational work to do first.
# The Cost of Dashboard Sprawl
Let us be clear about something: we are not telling you to delete every dashboard you have. That would be wasteful.
Dashboards are cheap to run. Build one, and thousands of people can use it for almost nothing. For reporting that happens on a fixed schedule, like your weekly numbers, board deck, and Monday morning review, a dashboard still makes sense. The cost math works in its favor.
But fixed reporting is only a small slice of what your organization needs from data. Everything else is the long tail: one-off questions, ad hoc requests, and the familiar "can you just pull this for me real quick?" That is where dashboards get expensive. Not in computing cost, but in people cost. An analyst takes the request, figures out what the person actually wants, pulls the data, builds something, sends it back, and often ends up building yet another dashboard. When that cycle happens every day across every team, it adds up fast, and it does not get more efficient over time.
Agents change the cost equation for the long tail. Yes, there is a cost to running AI agents against your data, and your organization should budget for it. But that cost needs to be compared against the analyst hours currently spent answering ad hoc requests, rebuilding similar reports, and translating business questions into data pulls.
When the data foundation is ready, agents can answer those questions in seconds instead of days. The value is not just the lower effort, but faster access to the information people need to make decisions.
A better framework is not zero dashboards. It is intentional dashboards: a small, deliberate set for recurring reporting and highest-frequency needs, with an agentic layer handling everything else.
# Are You Ready? What Agentic Analytics Requires From Your Data Environment
Knowing where things are headed isn’t the same as knowing what to do about it. So, let's get into what your organization actually needs in place for agentic analytics to work.
Most of this is not new. These are established practices that organizations running agentic tools successfully have often had in place for years. If you don't have them, that's fine. Gaps can be fixed, but you need to know where yours are to prioritize the work.
There are five areas to assess. For each one, don't ask "Do we have this?" Ask: “Is what we have good enough to be trusted by a system that won't ask follow-up questions?”
# 1. Your Data Foundation
Agentic analytics only works if the data underneath it is complete, consistent, and structured around how the business actually makes decisions. Everything else sits on this. If the foundation is shaky, nothing on top of it will work, no matter how good the tool.
A solid foundation means your data is complete, consistently formatted, and reliably loaded. Your history is intact. Your data model reflects how the business actually thinks, not how someone structured it years ago for convenience.
The warning sign: if your analysts must remember which version of a table to use, or filter out records everyone knows are bad, then your foundation has problems. An agent won't know how to do either. It will query what is there and return an answer. That answer is only as good as the data underneath it.
# 2. Your Data Pipeline Infrastructure
Your foundation is about what data you have. Your pipelines are about how reliably that data gets updated and how quickly you know when something breaks.
That reliability matters even more when business users are getting answers directly from an agent. Agents need fresh data. When someone asks, "What were yesterday's sales?" they mean yesterday. That means your pipelines need to run on schedule, flag failures immediately, and check data quality at the point of ingestion, not three days later when a manager spots something weird in a meeting.
This is the least exciting part of the readiness conversation. Nobody wants to talk about pipeline monitoring when the AI demo is right there. But an agent running on stale or broken data doesn't just give bad answers; it gives bad answers confidently. That is worse than no answer at all, and it is the fastest way to lose trust in a new tool.
Simple test: if your most important pipeline failed silently tonight, how long before anyone noticed? If the honest answer is "a few days," you have real work to do here before you turn on an agent.
# 3. Your Semantic and Integration Layers
Most organizations skip the semantic layer or let it grow without structure. It's also the one that matters most for agentic analytics to work correctly.
The semantic layer is where you define what your data means. What counts as a customer? How is margin calculated? Which tables connect to which? In a traditional BI setup, analysts carry this knowledge in their heads or bake it into individual reports. When the analyst leaves, the knowledge goes with them.
Agents cannot work that way. They need a single, centralized place where business logic is written down, something any tool can read and any person can audit. Microsoft's Unified Data Model, dbt Semantic Layer, LookML, and similar tools have existed for years. The organizations successfully running agents today built this layer before they needed it.
The integration question is related: can your agents reach your data across your warehouse, CRM, ERP, and other core systems? Are they connected in a consistent way, or does each one require its own custom setup? Platforms like Tableau and Salesforce are releasing MCP servers that simplify this, but your architecture still needs to be ready to connect.
If your business logic lives in hundreds of scattered report measures, dozens of custom calculated fields, and one analyst's spreadsheet, that's not a semantic layer. That is debt that will slow you down when the business starts expecting faster answers.
# 4. Your Data Governance Framework
Agents don't paper over data governance gaps the way humans do. They expose them.
When someone asks an agent, "What was our Q1 revenue?" the agent needs to know which definition of revenue to use, which source is authoritative, and whether that person is allowed to see the answer. A human analyst can use context and judgment to navigate those decisions. An agent works only from what has been explicitly defined.
Good governance in this context means three things.
- Access Controls: Who can see what, enforced at the row and column level, and carried through to every agent response.
- Data Ownership: Every domain has a person responsible for its accuracy.
- Lineage: A record of where data came from and how it was transformed, so when something looks wrong, you can investigate instead of guessing.
None of this is new, but agentic analytics raises the stakes because governance gaps show up directly in the answers people use to make decisions.
Without governance, you don't get just wrong dashboards. You get wrong answers delivered fast and with confidence to the people making decisions. That's a worse outcome, and it's preventable.
# 5. KPI Alignment and Business Metric Definitions
This is the one that organizations underestimate the most, and it can derail an agentic rollout fast.
Ask Sales, Finance, and Marketing to define revenue. You'll get three answers, maybe five. In a dashboard world, that disagreement can stay hidden because an analyst picked a definition at build time, and everyone sees the same number. Nobody has to agree because nobody has to discuss it.
In an agentic world, two people ask the same question slightly differently and get different answers. Within days, trust in the tool is gone. Once people stop trusting a data tool, it's very hard to win them back.
The fix is a metrics catalog where every KPI has one definition, one owner, one agreed calculation, and a sign-off from the business stakeholders who use it. This cannot live only with the data team. The CFO, VP of Sales, and other business leaders need to agree on what revenue means before an agent surfaces the disagreement in front of your CEO.
Most organizations will be stronger in some of these areas than others. That's a normal place to be. The goal is not to have everything perfect before you start, but to understand which gaps create the most risk and which foundations need attention first.
# How the BI Analyst Role Changes
BI analysts aren't going away. Their role is shifting toward work that is more valuable.
Analysts will spend less time formatting charts, rebuilding similar dashboards, and fielding "can you just pull this?" requests. They will spend more time defining metrics, documenting data models, building the semantic layer, and giving AI tools the context they need to return accurate, consistent answers.
The skill is moving from making data visible to making data trustworthy and machine-ready.
As more business questions move into conversational, agent-supported experiences, organizations will need cleaner data, clearer definitions, stronger governance, and analysts who can connect all of it.
Not sure where your organization stands? Propeller helps teams assess the data, governance, metrics, and operating foundations needed for agentic analytics, then sequence the work into a practical roadmap.
