7+1 Recommendations to Become a Data-Driven Company
It’s not enough to call yourself a data-driven company, you must adapt your practices to truly become one. Of course, the shift is never easy and requires some painful changes in technologies, company culture, and your perception of success, including the ways you communicate with each other. Although the following steps can’t guarantee your company success, they have greatly helped TourRadar in making smarter, quicker business decisions that have led to truly impressive growth.
Save as much data as you can
It may sound obvious now, but it’s one of the best pieces of advice we wish we had gotten several years ago when our company was first formed.
It can happen that the vast majority of your data is never used, but the problem is – you never know what information will help to answer that business-critical question in a year or two when you’re suffering from a conversion rate drop. Nowadays, if we build a new service or feature, we make sure that every user action and intent, every request and response, every change of state of any object is logged in some way.
To leverage the costs, we use different types of architecture – for high-demand data it’s an RDS instance on Amazon or the pricier option of Redshift. For supplementary, but not yet essential data, we prefer much cheaper options like S3.
Trust your data
Why would you collect such incredible amounts of data, if you don’t trust it, right? Every report or a data pipeline should include a validation step – for example, if we calculate a conversion rate (one of our business-level KPIs) both the number of purchases and the number of visits are controlled from several independent data sources – from a main relational database, from our internal tracking system (PostgreSQL + Redshift), from Google Analytics and even from the logs of our Customer Support team.
If we observe any significant discrepancies, we don’t continue until we eliminate them. Once we build a completely reliable pipeline, it is the data, not the opinions of individual people (no matter how influential they are), that become the only source of truth in decision-making.
Use simple and clear reports
Your data analysts can be very smart people, but one of the most important skills for them to possess is the ability to explain complex structures and relationships in a simple manner. The Data Viz Project contains more than 150 types of data visualizations, from easy-to-read bar graphs or line charts to exotic partition layer chart icicle diagrams, and deciding which one to use in a report should involve a simple question – who are we talking to? If it’s kind of BI-to-BI communication, and we’re sure our collocutor understands what we’re talking about, the report can be as complex as we wish, but for a wider audience with different levels of experience and involvement we’re usually guided by the “weakest” participant and not ashamed by using the simplest types of visualization – even as simple as a scaled-up number.
Educate colleagues
One of the benefits of being a startup is that we’re surrounded by clever and effective people, and setting an open environment among others means sharing the knowledge and experience between each other – and at TourRadar it’s one of our company-level OKRs. It may sound peculiar, but in a data-driven company literally everyone, from CEO to content manager, from customer support agent to marketer, from frontend developer to office manager, should know how to deal with data and be able to set and achieve quantifiable performance metrics.
If a department in your company doesn’t require data to work, you can be guaranteed that it’s not efficient enough; from the other side if someone creates a ticket for the BI department every time they need to import data from CSV format to Google Spreadsheets, it’s an efficiency problem as well. To find the golden mean, just make sure that everyone in a company knows the basics of sheets, understands the difference between dimensions and metrics, knows how to build simple pivot tables or calculate a mean (or a median) of a given measure in a dataset. It can become even a gamified competition between people and departments with a clear learning path – from numerical methods to statistics, from statistics to spreadsheets, from spreadsheets to charts and pivot tables, then basics of regular expressions or SQL, and so on.
Ask the right questions
It requires a bit of patience and time, but it’s worth repeating again and again – it makes the corporate life and communication between people easier if everyone knows how to ask the right questions. Every analysis should begin with a question, and every move towards the goal should be an attempt to answer it, at least to some extent. To achieve this, it must be clear for every participant, that either the question itself should be measurable and single-valued, or the problem should be broken into parts until each part becomes simple enough to be unambiguously defined and understood by everyone.
“What’s the purchase path of young TourRadar customers?” is good information to know, but it’s completely wrong from a data analysis perspective. Whereas “What are 10 the most popular landing pages for people in Australia aged between 18 and 25?” is already a good enough question to start working with, despite not yet being in its ideal form.
Run more A/B tests
Post hoc ergo propter hoc or after this, therefore because of this, is a popular logical fallacy and for good reason. Every project manager is familiar with the problem of a late Friday release turning into 10% conversion rate drop on Saturday, and the emergency team spends the entire weekend trying to understand if it’s caused by the changes on a website or just a normal symptom of seasonality.
A/B experiments can eliminate these uncertainties and clearly show if people like the changes on your website. Obviously, it requires some investment in education and a bit of infrastructure, but the confidence in the product is worth the effort. Simple and free tools like Google Optimize can be a good start (even though they don’t care about some important complications like the multiple comparison problem and also don’t allow for the easy exporting of raw data), but once you get more experience, move towards more sophisticated tools or even own implementation – it’s not as difficult as it seems at first sight.
Build a culture
Data, data everywhere. Maybe it sounds boring and geeky, but SMART goals are easy to understand and track progress effectively, they also help you detect problems earlier in the production process. Every department (and with that every member of it) should have measurable time-related goals, and sometimes they’re not so easy to come up with, but take it as an exercise in creativity. It may even involve a different way of presenting the company results – instead of “we’re doing well, but not perfect” it can be “we’re 25% better in metric X than last year, but we’re supposed to be 30% better”.
Bonus: Don’t be scared
It may cast down, especially if you’re only at the beginning of the journey, but switching to a data-driven approach doesn’t necessarily require dramatic changes – so no need to “start a new life” next Monday. Begin with something measurable, achievable and simple enough that you’ll be able to start thinking differently straight away. The changes can be as simple as hosting two workshops and running one new report and a/b experiment in your first month. Simply accelerate this pace month after month and before you know it, you’ll have tangible and actionable results at your disposal to help achieve your company’s unique goals.
