Here at OptimalBI we have recently trialled a slightly different approach to our data modelling, with great success.
As a consulting business our customers hire us because we’re good and we deliver, usually in an area they don’t have the resource available for in-house. Sometimes this can come with a pretty tight deadline.
In the past, in regards to analytical modelling contracts, they’ve always been a bit of a “one man band” kind of a job. One person lives at the customer’s site for a few months and immerses themselves in the data, not really coming up for a breath until it’s all over. This works, but it could work better.
There are a few problems with this kind of approach:
One person holds all the IP. Yes, that one person documents everything but there is still a vast difference in knowledge between the person that reads the handover documents (if they actually do) and the person that created the model.
Assumptions will be made. This is a given and most of the time they will be absolutely correct. As an Analyst, your mind is constantly asking questions – some you’re quite capable of answering and some you make judgment calls on, based on your experience. There is not going to be someone available from the business to ask every single question the model builder has, at the time they think of it. This can lead to starting and stopping parts of the model until the answer can be tracked down.
Rabbit holes will be travelled. Anyone that has done any kind of data analysis will have been down one of these. That beautiful path you are so proud of and spent hours creating, only to find out that it is completely unrelated or unnecessary for the project you are working on.
It will take time. There is an age-old argument about the time it takes to build a model – expect to receive an answer something along the lines of “how long is a piece of string”. As a general rule of thumb, for a relatively simple model, allow a few months for data preparation and a month for model building (in an ideal raw data world!).
The solution? Two people. One Model. Here’s why it works:
The IP is shared. The age-old DR scenario, “if you got hit by a bus…”, I’m sorry to say but that wouldn’t be a problem. The peer modelling approach is not limited to two people from the same consultancy company. Imagine if you could up skill one of your own analysts and at the same time create an analytical model where the IP is kept within the company.
Questions will be answered. It is always nice to have someone to use as a sounding board when designing and building a model. Someone that can take a slightly different view on things and, between the two of you, prevent the model containing too many rabbit holes.
Skill sets will be increased. In a world where everyone is after the elusive “Data Scientist” (who, from definitions these days, is so diversely skilled I don’t believe they actually exist), it is still possible for a model to be built by people with combined Data Scientist skills. I find the peer modelling approach a great learning experience. If you get the opportunity to work with someone that complements your skill set, and enables you to learn from theirs, it can be a great project for all involved.
It will take less time. Just how much less time depends on the model being built and how well you’re able to work as a team. One of the arguments against peer modelling is that it’s more expensive because there are two people involved. Yes, paying two people is more expensive but what if it could take these two people half the time to build you a model within your deadline?
It’s fun. Sometimes it’s just nice to know that you’re not out there battling away by yourself. Stress is always lessened when shared and late nights are much more entertaining while watching someone you know dance around a cleaner.
Keep it colourful.