rais.io
AI-driven Customer Data Management and Utilisation platform

Blog

Helping commerce businesses succeed with their customers.

Ecommerce Data Planning - getting tactical to prepare for analytics

In my previous article on ecommerce data planning I covered the three questions you need to ask yourself when you embark on a customer insight project or when building your customer database. Following on from that there are three additional ecommerce data planning related questions that help you get into the detail of what's required for good ecommerce data planning.

1. How easy is it to get the data we need (i.e. is the data ours, someone else’s or a mixture?)

So by now you should have a good idea about the most important data you need to achieve your goals.

Now you need to figure out where the data sits. Is it inside your organisation? Or do you have to go hunting elsewhere?

This is where data dictionaries come in really handy, as they’re a quick way of telling you what data you have and haven’t got. They also define the meaning behind of variables, helping to avoid any confusion. So they're a great data planning tool. Without data dictionaries, it can take time to know not only what data exists but also what it means. It can be a bit like walking into a library without good categorisation. It would take a long time to find the books you’re after!

What if you don’t have the data you need? You need to figure out whether its more opportune to generate it or procure it. Companies like Acxiom and Experian sell data, which when appended to proprietary data sources can potentially add a lot of value. Other companies allow you to append data via open APIs like FullContact. Perhaps you need to get hold of some of your trade customers' data if you’re an ecommerce business that sells predominantly wholesale? That can obviously be a challenge given the value and sensitivity of data and their unwillingness to share it. It may be more cost-effective to create your own data via consumer surveys. Really it all depends on what you’re trying to achieve and the extent to which you need 100% accurate data to draw viable conclusions from. Clearly, procuring data from elsewhere can come at a cost. So ask yourself how much you need it and if you can still achieve your goals without it.

2. How should the data be structured for analysis?

Before getting your hands on the data, you need to know how data are structured currently. Certainly your internal proprietary data is likely to be stored in a large database of some sorts, perhaps in different data stores. For example there may be a data store containing transactional data. There may be another containing customer records. The ability to link data from different data stores is dependent on a variety of common data variables. If these exist they can often update or be missing for given order records, limiting the ability to create an accurate profile of each customer, which could jeopardise targeted marketing efforts.

So at this point you'll learn to what extent the database has been strategically designed for insight in mind!

Clear data architecture documentation will hopefully exist to help you understand what data exists where and the links between data stores. It will also tell you whether an environment already exists that is suitable for analysis.

You shouldn’t perform analysis directly on your main database, because of its size and the potential for data to be compromised. There should be a single untouched version of data at all times. So it’s better to extract just the right amount and type of data needed to perform analysis. This can be established as a regular feed that extracts data and loads it into a separate environment for data analysis. We use an ElasticSearch database to do this on our software platform

3. How much data do we need?

Sounds like a simple one, but it’s important to be clear on the implications of the amount of data you think you need. Extracting and manipulating a lot of data can take time and space, so you will need to consider: time periods and sample size.

Do you need two time periods’ worth of data to perform comparative analysis? How long do those two time periods need to be? Would extracting a sample of data impact give you robust enough results for what you’re trying to achieve?

 

From ecommerce data planning to data analysis

Now you've worked through these questions, it's time to look towards the analysis of the data you've prepared. There are different approaches and implications associated with data analysis so i'll cover those in a future article.