Dashboarding in early-stage startup: learnings from 3 years of experimentation.

Almost two years ago, I applied to join a small and ambitious startup that envisioned disrupting the whole goods transportation industry.

The position was entitled “data analysis and transformation project” and I remember feeling ashamed in my first interview not to be fluent in Python or SQL, as the “data analyst” part of the job could have suggested.

The thing is — luckily for me at this time — pre-series A startup rarely needs “data analyst” to grow. They need problem-solvers, that can quickly adapt to new tools, learn technologies and hacks, get their hands dirty and efficiently implement the things the startup will need to grow.
They don’t need analysts capable of applying advanced statistics and complex reports to solve problems: they need straight-to-the-point people that can deliver smart and scalable solutions and focus on a handful of strategic KPIs.

Taking this into consideration, there is still some part of the data analyst job that comes down to an absolute requirement for any startup at any level: Dashboarding 📊.

Whatever you call it — whatever the delivery format, you need to have your collaborators align around those KPIs. As many other startups, we truly believe that data needed to be accessible, and consumable by any member of the team.

That’s no easy job. In two years working on growth and data at Everoad*, we learned a lot — failing quite a few times — on how to properly deliver KPIs to a wide range of counterparts.

Here are some of the key principles we have learned from developing data visualisation at Everoad. We won’t tackle specific design principles, nor stack or tool preferences, just a handful of general guidelines to avoid investing time on building something that people won’t use.

The 3 conditions rule.

Capturing the real efficiency of a dashboard can seem abstract: like everything that involves design and a variety of people to judge it, it’s easy to get lost on whether you did a good job or not. The key here is to remain factual, and for that, one can measure a data visualisation tool by its ability to meet the following 3 conditions:

1 • A dashboard should be flawless

This means that the data is accurate and cannot lead to misunderstandings. This requirement is to be taken as a boolean: it’s flawless or it’s not. And that’s it.

2 • It’s accurately delivered

This means that the dashboard is accessible and usable without condition, anytime users need it. This also entails that the tool is powerful enough to support computations and visualisations relentlessly.

3 • It’s used

This is in the end the most important one. Dashboard are to be measured as products: if they are good, they are used. If they are bad, they are not and that’s it. Period.
At Everoad, we make sure that every data initiative is measured with a target number of recurring users, every time we conduct such initiatives.
For that purpose, we use Google Analytics plugged in to our visualisation tools to clearly know whether the tool is used or not.

Dashboards answer a select set of questions.

This is somehow a way to tell you: ✋ beware of the ubiquitous/all-in-one dashboard.
When listening to end users, plenty of them will explain how they want one dashboard to display absolutely everything they need. That’s exactly where you do not want to go: a dashboard that displays everything, but that doesn’t properly answer any questions.
Focusing on a finite and well defined number of questions is an absolute necessity to build products that will remain.

Something else you want to avoid: building dashboards when the questions are still being determined. In other words, an exploratory analysis — performed with proper data analysis tools — should be conducted before the moment when you will think about building a dashboard.

You absolutely don’t want to start a dashboard at the discovery phase of what needs to be displayed. Only once the data is understood, the processes are defined and the KPIs are well thought out that you can think about the last step: Dashboarding.

About design: how to remain efficient.

Dashboarding is about having data easily accessible and understandable for everyone.
A very good way to do this is to make your data visual: it helps people to “tell stories” efficiently and esthetically.
That said, and for obvious reasons, design will be at stake.

The point is, good design might be hard to achieve. It can even require specific skills.
Unfortunately, people good with figures or with a sharp business acumen are not always good at designing things. And as a matter of facts, computing figures, and engineering data is a completely different job than making this same data visual well- designed.
To make it even more complicated, people tend to interpret the colors and shapes of a visual faster than the meaning of the display itself .

Fortunately this time, there are ways to be design-efficient, and build things that engage for users without too much struggle.

First, choose a color pallette that makes sense and stick to it. The fewer colors, the better. Start using your company colors— the dominant one on your company’s website — and decline it on coolor.co. If you have a more corporate color panel, perfect, use it.

Second, the key to design for data visualisation is efficiency: the fewer elements should enable you to learn as much as you can. The more complex, the more you will use use end users, the less efficient it will be.
Last thing, and we will keep this short: for 90% of you dataviz, you will only need scorecard, tables and barcharts — and that’s it.

In other terms, the more simple it is, the higher the chances you will have to propose a more than decent design.

Conclusion

Making data visual is an essential in modern businesses and is a key ingredient of a company’s data culture. Unfortunately, this challenge often comes undervalued or misunderstood.
Anyone can throw a histogram from an excel spreadsheet, but dashboards and data visualisation are more complicated than that. Good data visualisation stands at the cross-roads between data analysis, design, and product management.

Yet, this remains an essential brick in making data a differentiator, enabling young businesses like us to reshape entire industries — like trucking 😉

Thanks for reading! 🙌 🚚 ❤️

Give some claps 👏 if you loved this article, why not up to 50 claps 🎉 🍾 This will encourage us to share more knowledge, methods, and ideas from Everoad!

— — — — — -

*Everoad is a Paris-based business that brings data to the trucking industry in Europe through an online marketplace, connecting shippers and carriers together. And we are hiring 🚀 !

*Our stack is as follows:

• for main Dashboards with easy visualisation, we use Google Data Studio: it’s easy to use, beautiful, and very easily interfaced with all of our data warehouse (Big Query)
• For visualisation with advanced tables — in terms of complexity and formats, we use Google Sheets.
• For more complex visualisation, exploratory analysis, and advanced statistical analysis, we use Jupyter Notebooks.