The Death of Dashboards
The analytics interface is changing faster than most teams realize.
For years, the dashboard was the centerpiece of data work. We built them, maintained them, presented them in quarterly reviews, and defended the numbers when different teams saw different figures. Entire analyst workflows were organized around keeping them current.
The tools we’re building now read data, explain it, and answer questions directly. The interface is changing, and the question is how far that change goes.
Why Dashboards Became Popular
Before self-serve Business Intelligence (BI) tools existed, getting an answer from data meant opening a ticket with the analytics team or writing SQL yourself. Most business users did neither. They waited for someone else to pull the numbers, package them into a report, and send back an answer.
Dashboards changed that. Products like Tableau and Power BI let business users explore data without analyst involvement. A marketing manager could filter by campaign, a product lead could compare cohorts, a sales VP could check numbers before a board meeting. No ticket required.
The adoption was fast because the problem was real: most business users needed data to make decisions, but they had no practical way to get answers themselves. Between 2010 and 2020, the global BI market grew from roughly $10 billion to over $23 billion. Companies that had relied on monthly Excel reports were suddenly running on live, visual data.
The dashboard became the default answer to the question: how do we make data accessible?
The Problem With Dashboards Today
Dashboards have accumulated faster than most organizations can manage.
At many mid-to-large companies, the count runs into the hundreds or even thousands, creating governance, duplication, and adoption problems. Gartner described the issue in 2025 as a cycle of “dashboard proliferation, limited adoption, and underwhelming business impact.”
The structural problem is that dashboards answer the questions analysts anticipated, not always the questions users actually need to ask at the moment.
Consider a regional sales director reviewing a revenue dashboard. It shows the drop in Northeast numbers. It doesn’t explain it. So the director sends a Slack message to the analyst, who pulls the data manually. The dashboard was there and accessible, but it wasn’t enough to close the loop.
There’s also a trust problem. When multiple dashboards show different revenue figures for the same period, because of different join logic, different date filters, or different metric definitions, users stop trusting any of them. The dashboard creates uncertainty rather than removing it.
Maintenance adds to the cost. Every new business question becomes a new dashboard or a new filter. Analysts spend a significant share of their time on dashboard requests rather than on actual analysis.
The Rise of Conversational Analytics
The shift started when LLM technology became capable enough to translate plain English into SQL reliably.
Natural language query (NLQ) tools existed before 2022, but they were fragile. You had to phrase questions in a specific way. They misread ambiguous terms. They required significant training on your specific data model to produce correct results.
GPT-4 changed the baseline. Text-to-SQL started working at a quality level that made production use viable. Combine that with a semantic layer that tells the model what “revenue” means in your schema, what “active customer” means, and which tables are authoritative, and you have something that can answer ad-hoc questions reliably.
The concept: a user types “What drove the increase in average order value last month?” and the system generates SQL, runs it, interprets the result, and writes a plain-English explanation. No dashboard required.
This is conversational analytics. The user asks. The system answers.
Real Technologies Driving This Shift
Several converging pieces are making this practical rather than theoretical.
LLMs with text-to-SQL capability
OpenAI’s GPT-4, Anthropic’s Claude, and Google’s Gemini can generate SQL from natural language with reasonable accuracy when given schema context. Most conversational analytics products built since 2023 use one of these models as the core query engine.
Semantic layers
Tools like dbt Semantic Layer and Cube create a business-friendly abstraction over raw data models. They define metrics, dimensions, and relationships in a way that LLMs can use for accurate query generation. Without this layer, LLMs generate SQL that sounds plausible but produces wrong numbers.
AI features built into BI platforms
Tableau Pulse sends automated insight digests to users without requiring them to open a dashboard. Power BI Copilot generates visualizations from natural language inputs. ThoughtSpot Sage adds an LLM layer on top of its search-based analytics engine. Databricks Genie lets teams configure Genie Spaces, curated environments where business users query their data in natural language.
Vector databases for context retrieval
Tools like Pinecone and Weaviate let analytics systems retrieve relevant prior queries, metric definitions, and documentation to ground LLM responses. This reduces incorrect outputs and improves consistency across queries.
Get Nathan Rosidi’s stories in your inbox
Join Medium for free to get updates from this writer.
None of these is mature yet. Text-to-SQL still fails on complex joins and ambiguous business logic. Semantic layers require significant setup investment. The trajectory is clear, though.
What Changes for Users
The biggest change is what the user has to do.
With a dashboard, a user arrives on a page, chooses filters, reads multiple charts, makes comparisons, and forms a conclusion. That cognitive work happens entirely on the human side. The dashboard is passive.
With conversational analytics, a user types a question and reads an answer. The cognitive work shifts to the system.
That’s a meaningful change in who the interface is designed for. Dashboards lowered the barrier to accessing data, but they still required genuine data literacy to use well. A user who didn’t know how to read a funnel chart, or who didn’t understand what “cohort” meant in a filter, couldn’t get value from most dashboards.
Conversational interfaces lower that bar further. A supply chain manager doesn’t need to know which table holds purchase order data. They ask.
The personalization aspect is underappreciated. A static dashboard shows the same view to every user. A conversational interface adapts to whoever is asking. The VP gets a summary. The analyst gets the methodology. The regional lead gets their geography without having to navigate global filters.
What Will Replace Dashboards?
Dashboards won’t disappear.
For monitoring, they’re still the right tool. An engineer on call wants to see service latency on a chart, not type a question. A finance team running a weekly close needs consistent, fixed views of specific metrics. Operational dashboards that display the current system state will stay.
What’s changing is the use case for ad-hoc questions and trend monitoring.
Ad hoc analysis will move to conversational interfaces. Instead of filing a dashboard request for a one-time question, teams will route those questions to an AI query tool and get a direct answer.
Trend monitoring will shift from users pulling data to systems sending insight. A business user won’t check a dashboard daily to notice a change. The system will push a message: “Churn in the Enterprise segment increased 8 percentage points week-over-week. The increase is concentrated in customers who haven’t used the new export feature.”
Automated narrative reports will supplement periodic PDF summaries. A weekly business review generated by a system that reads the actual data and writes a plain-English summary, tailored to the specific audience.
The dashboard category will narrow rather than disappear. It becomes one tool in a larger set, no longer the default answer to every analytics need.
The New Role of Data Analysts
The work changes significantly.
When dashboards were the primary output, analysts spent most of their time on them: requirements gathering, building, maintaining, debugging stale filters. A large share of that work was repetitive.
With conversational analytics, the bottleneck moves upstream. The system needs accurate semantic models, well-defined metrics, and clean trusted data. The analyst’s job shifts from building views to making the data environment reliable enough for AI to query correctly.
The work that remains is harder. Ambiguous metric definitions, messy source data, undocumented business rules, governance gaps: these become the analyst’s primary focus rather than background noise.
New technical work appears too. Semantic layer design, prompt engineering for analytics tools, evaluation frameworks for LLM-generated SQL, data quality monitoring for AI query accuracy. Most of this didn’t exist in job descriptions five years ago.
The floor for basic data access rises. Anyone with access to a conversational tool can now get numbers without analyst involvement. That frees up analysts for deeper work, and it also means the bar for what counts as analyst-level contribution goes up. The job gets more technical, not less.
What Comes Next
If you’re in data, the shift is already in progress. Most major BI vendors have shipped or announced conversational analytics features. LLM-based query tools are running in pilots at large companies across every industry.
A few things are worth starting now.
Audit your active dashboard portfolio and cut what isn’t actively used. A smaller, trusted core is more useful than a large, questionable set.
Invest in your semantic layer. A well-defined metric in dbt or Cube is reusable across dashboards, conversational tools, and automated reports. It’s the most durable investment you can make, given where the space is heading.
Run a small pilot with one conversational analytics tool. Tableau Pulse, Power BI Copilot, and ThoughtSpot Sage all have accessible entry points. Understanding where the gaps are now is better than learning them under pressure later.
The interface to analytics is changing. The teams that adapt early get a lot of their time back.
