What I've Learned from Setting Up Analytics at a Start-up

When I joined a start-up recently, I was tasked with setting up an analytics function from scratch. As things go with start-ups, you are often facing tasks which you might not be 100% ready for. While I am a big fan of anything to do with data, I am not super-technical and haven’t had to run an analytics function before. Below are the few things I've learned which will hopefully be useful if you are in a similar position.

Step 1: Read Around

There is a lot of advice out there, some good & some not so good. The article which I found particularly useful is The Startup Founder’s Guide to Analytics by Tristan Handy, Founder and CEO of Fishtown Analytics. In the article, Tristan lays out very nicely what you need to do (and not do) at each stage of growth. There are a lot of great points which I took on board, but won’t repeat them here, so read that article too.

Step 2: Understand the Goals

The goals will evolve as you understand what’s possible & sensible and what isn’t, but it is good to start shaping up an idea of what you’d like to achieve. For me, the key goals that I had are:

1. Create automated reports for basic MI (e.g. users, conversion, revenue)
2. Enable business users to answer their own data questions
3. Create a single source for all data & achieve a single view of a customer
4. Find ways to make life easy for future data scientists that join that team
5. Minimise tech development time & achieve all of this at a reasonable cost for a start-up

I would say that numbers 3 & 4 were at the back of my mind the most, as I have seen a lot of companies which cannot get basic insights because data is not there or isn't ready to be used. Or they end up wasting their data science talent as these highly skilled individuals are spending their time cleansing & assembling the data.

Step 3: Research the Tools

It is worth spending a bit of time doing extra research on what tools are out there. You’ll find that this space shifts a lot and even experienced people in this space will be surprised by a new tool that’s popped-out. One example of this is Panoply.io, a beautiful tool which helps you get data from various sources (ETL) and has a smart data warehouse capability (it is built on Amazon Redshift) which reduces your data admin time by a lot.

You can see the setup we have looked at below. The key bit in my mind is collating all data in a single place (Panoply.io), get that even if you don’t want to do a lot of fancy stuff, it is only from $150 a month - peanuts compare to what it can do for you.

If you don’t want to spend a lot of money, you can use open source BI tools (e.g. Redash) or just use SQL/Python to get the data out to Google Sheets / Excel and create quick reports from there and leave your data scientist to do more advanced stuff in Python as that's what they are probably used to anyway.

Step 4: Engage with the Business

Hopefully an obvious point, but collecting good business requirements, prioritising & tracking their delivery should be a key ongoing process to make sure that what you are doing is going to deliver real value to the business & early.

We have collated 200+ requirements and you probably would too, my suggestion to dealing with that is to treat analytics as any other product development: write 'user stories’, prioritise them, estimate difficulty and work through the backlog in a structured manner, testing & improving the outputs along the way. That way everyone knows what’s being worked on and you won't be overwhelmed by huge number of requests.

Step 5: Learn SQL & Python

At the end of the day, it will be very hard to run an analytics function without knowing SQL (for basic manipulation of data) and Python (for more advanced analysis). There are plenty of courses out there, edX and Coursera are probably good starting points.

Hopefully you've found this useful and I'm sure I'll learn more as we progress through the journey.