TL;DR
- If your work is taking place in a Notebook, begin with a scheduled hosted drag and drop Notebook Dashboard like Deepnote, and then if you need distributed computing levels of scale move to Python + Streamlit or Python + Plotly Dash with a distribution framework like Dask.
- If the types of data collected changes often, and you have Data Engineers backing up your Analytics and management teams, use Looker.
- If your dataset schemas are mature and you want to maximise the power of analysts use Tableau + a Front-end to SQL. Tableau is the most fully-featured in terms of types of plots and styling, at the cost of a steeper learning curve. SQL front-ends allow for rapid iteration of views into your data. If you have problems with setting up Redash yourself, there are some immediate alternatives you might want to consider.
- If you don’t have dedicated data analysts and are already paying a Microsoft subscription, Power BI is the most user-friendly. However, Power BI achieves this by forgoing certain features and doesn’t quite match the capabilities of Looker or Tableau.
Estimated Reading Time: 15 minutes 🕒
Data analytics is all about finding insights and causing business strategy and operations decisions. While the most challenging part is finding new insights, the most important part is packaging these insights into something that is consumable, reproducible, and update-able and that tracks the effects of change.
Enter Dashboards, a (usually) interactive collection of visualizations that have a live connection to the data, and update over time. With the explosion in Data Science/Analytics as a field, so has there been an explosion in Dashboard products and tools, including ones that wouldn’t traditionally be thought of as dashboards.
This article explores some of these tools, and ultimately gives recommendations for which tool you should use given your use-case. The tools looked at here will be
- Business Intelligence (No/Low Code) Tools – Microsoft Power BI, Tableau, Looker
- Hosted Notebook Dashboard Apps – Deepnote Dashboard, Hex Data Apps
- Front-end to SQL with automated plotting – Redash, Deepnote, Hex, Databricks etc.
- Python web apps – Plotly Dash, Streamlit

Enterprise Business Intelligence – The major players
Business intelligence tools like Tableau, Power BI and Looker are designed for corporations with mature data stored in nice relational tables. They provide easy ways to pull data out of almost any database management system, enable no-code plots, and can be scheduled to automatically refresh and ping the people who need to see the dashboards. They can all be hosted in the cloud, or deployed on-prem, with Looker having the most involved installation process on-prem.
For data that’s been cleaned, and needs to be presented to higher levels of management, these are the tools to use. They create dashboards that look amazing, have intuitive user interfaces, and enable drilling-down into the data at arbitrary resolution.
The major differentiator between these tools is their target audience, and that is reflected mainly in their usability.
Tableau is designed for specialist data analysts. It is extremely feature-rich and encourages extreme levels of customization for each plot. Because of this, it has quite a steep learning curve. To get the most out of Tableau, you need your analysts to be able to invest time into learning its features.
Tableau is the default option for many major enterprises, and for good reason. If you have analysts who can ‘live’ in Tableau, they can be incredibly productive, and their work can be almost immediately shipped as presentation or report ready. If the plots and insights you’re producing could be done, albeit with extreme effort, in Excel, Tableau is probably the right tool for your workflow.

Looker tries to democratise the insight production part of data analytics to enable all members of a business to build dashboards. It sets up Looks into the data as a product that technically strong data analysts/software engineers provide to the whole business. This means that understanding the Looker experience needs to be broken into two components; firstly for the analyst/software engineer setting up the Look, and secondly for the business users.
Looker’s internal language LookML is really syntactic sugar to SQL that is much easier to manage at scale. Since this is made with software engineers in mind, it has great version control, DevOps practises (dev to staging to prod) and is convenient. However, since Looker places the data analyst closer to a software engineering role producing software to use internally, these analysts will need to be more technically capable and their statistical knowledge will be relatively less important. This might mean that the analysts in your business may not be well-suited to this role, and may need retraining. This also means that there is a large upfront setup cost in using Looker.

The business user side of things does well at being simple, but certainly leans on the side of being fully-featured at the cost of universal accessibility. Looker’s UI is more directly intuitive than Tableau’s, but the business user will want to be technically savvy, and will benefit greatly from having done even a small amount of SQL before. If you’re a medium-sized or large tech business, with a technical workforce, Looker is probably the best tool for you.
Microsoft’s Power BI is the simplest of the enterprise tools, targeting people with no data analytics/technical background. Despite this, it is still extremely capable, and is the first thing I’ll personally use if I’m looking at a subset of my data on my local machine. It has a snappier UI than Tableau for small datasets, and comes with many of the same integrations and features.
In short, Power BI has totally replaced Excel for investigating and plotting my data once the statistics have been calculated. If you’re trying to make quick plots for a school report, or currently live in Excel, I’d strongly recommend trying Power BI and seeing if it can replace your Excel workflows.
Hosted Notebooks with Integrated Dashboards
With the rise of Notebooks in Data Science, with built-in SQL, Python, R, and Julia (to name a few) interpreters, we’ve seen product offerings like Deepnote and Hex Technologies add a feature that allows dashboard/data app functionality. I think this is an awesome way to get started, and rapidly prototype what data and interactions are useful to downstream users.
The approach is simple but powerful; Notebook outputs can be added to a dashboard, and then can be tiled however you like via resizing and dragging and dropping. Then, input is done via built in Input Cell
types.

This allows for arbitrary complexity in the dashboard, since the logical backend is just Python. However, this also means that all logic has to be explicitly written by the Dashboard creator, instead of being handled by a backend built into a service like Tableau.
In my business, we’re often sending Notebooks back and forth all the time, and re-running notebooks to see updated visualisations. For this workflow where all the analysis work has already been done, converting to a Dashboard like this takes almost no time investment, and allows for a much less intimidating data and more information dense product to be sent around the business.

This is a great place for analysts to produce dashboards that they can send around to another analysts as they try and iterate the best views into the data. That being said, I don’t see these as replacement to fully featured dashboarding software for dashboards that need to be presented to executive, or that will be forked and modified frequently.
Front-End to SQL – Redash
Redash is an awesome tool that allows automatic visualizations directly from SQL queries, and then placing them in whichever grid you would like as in the hosted Notebook solution.
Redash was recently bought by Databricks and have decided to stop running their cloud server. This relegates its usage to businesses which have dedicated IT infrastructure to manage updates and set up the on-prem server.
Whilst Redash is great, their purchase by Databricks has meant that Databricks is swallowing most of the functionality into Databricks SQL Dashboards, which are competing directly with the hosted Notebook solutions as above. If you’re already using Databricks and not Redash, you’ve got a cool feature coming soon! Otherwise, its features are rapidly being added to many competing notebook solutions.
If you have problems with setting up Redash yourself, there are some immediate alternatives you might want to consider.
Python Libraries – Streamlit + Dash
The final dashboarding technology are Python libraries like Streamlit and Plotly Dash. These give you full access (more or less) to a web app framework. For example, Streamlit can allow the user to upload arbitrary files that can be processed, and Dash gives you full access to HTML to position elements. Whilst these tools are extremely powerful since they can use execute arbitrary Python, I would say that they make it difficult for users to generate their own insights that you haven’t seen, as you have to design each view into the data.
Personally, I often turn a personal data story project into a Streamlit web app and host it on Heroku for friends to play with. Or, at the end of a quarter I will collect all my findings and write them up in a Streamlit web app that collects and runs on new data so that if I come back and see what I was working on the data is up-to-date. I find this much more useful than static pages hosted on an internal wiki (i.e. Confluence), and updating internal business pages pages programatically is significantly more hassle than building a Streamlit app.
Here’s an example app built in Streamlit

These also have by far the highest technical bar, with Dash forcing you to explicitly understand how HTML positions elements and forcing you to do it yourself. Then again, Streamlit/Dash produce extremely interesting Dashboards with features that do not appear in any other tool.
For a detailed comparison between the two, JP Hwang’s comparison is a great read with the summary being Streamlit is easier at the cost of customization (focusses on rapid iteration) whereas Dash allows for enterprise level Dashboards (focusses on customization).
Overview
Building good dashboards is a crucial part of every data analysts workflow and are becoming increasingly important as data generates more and more value. An entire industry of products has risen to meet those needs, with each product specialising to a certain type of analyst/business workflow. This has meant that people are having to choose where to focus their learning, and businesses are having to choose where to invest their time and money. Here, we looked at four different categories of dashboard – and explored the differences between some of the leaders in the space.
I hope that this has taught you something about the world of dashboards in modern data analytics, and it’d be great to hear your thoughts in the comments!