Do you feel like your company could do more with the data that it has on hands? Or maybe you feel like a lot of time is getting wasted on unnecessary data wrangling and Excel spreadsheets? Or, on the contrary, do you feel like your data science efforts and investments haven’t really paid up and they could support the business in a more effective manner? You need to start asking if you have the right analytics organization in place.
There are multiple analytics organization structures available that you can choose from and the key is to firstly diagnose your current situation and evaluate which one works best for you. Weather a model is appropriate or not greatly depends on the nature of your business, its analytics capabilities and maturity and the business problems that need to be solved.
Integrated Data Scientists
- When to use it: In the very beginning of your analytics efforts
- Where you shouldn’t use it: When looking to scale up analytics and deploy pilot projects into production
This model involves having isolated data scientists in the teams that mostly need them. For this model to be effective, you need to make sure the data scientists are experienced enough to be able to investigate, formulate and come up on their own with the best and simplest solutions to the recurrent issues of your company.
This model works best especially if you are just starting out your analytics efforts and ensures that while the outcome will not be spectacular, the solutions and analyses will be deployed where they are most effective.
Most often than not you will only need a good data analyst that can start automating some of the reports where most of the time is getting wasted. The first step and most effective success is to be able to measure the performance of the business processes and to constantly monitor them on business intelligence dashboards that can give you simple and actionable insights. Brainstorming on dashboards can make everyone start asking the right questions that might require more advanced data science analyses down the line.
The drawbacks of this approach are that while you get to reap the low hanging fruits, once the data scientists need to tackle more difficult challenges they will miss a knowledge sharing environment or access to the right architecture and tools to be able to scale up their efforts.
They will also miss the influence to get unrestricted access to key company systems and so, they will need a data science manger that can negotiate and facilitate the access that they need more quickly. In addition, this approach could lead to higher costs if the acquisition and management of the data science tools and architecture is not centrally managed.
Hub and Spoke
Where to use it: In large mature companies looking to support business operations with analytics (Marketing, Sales, Manufacturing, Maintenance, HR, Finance, Operations, Supply Chain, etc.)
Where you shouldn’t use it: Product Development, Research
This setup enables leveraging your data science resources in the pressing projects of the business. In this setup, you would have a queue of projects (such as those that support your annual company objectives) which can be prioritized according to their potential impact. With this model, you can easily make sure that the right person is working on the right task according to their skills set.
Moreover, you can assemble and disassemble the teams according to the stage and difficulty of the projects to constantly adapt the allocated resources to changing specifications and new discoveries. Another advantage is that the data scientist will always have a backing team to go back to in case they need to gain knowledge or use more advanced software and tools.
However, being moved from one analytics project to another as the priorities change might lead to no results and increased frustration for your data scientists. As such, make sure that the business priorities are clear and that changes are not frequently made.
Stand-alone Data Science Teams
- Where to use it: Product Development and Product Redesign
- Where you shouldn’t use it: Operations
You should put in place a standalone data science team when you are planning to build a product that is mostly relying on data science algorithms. As such, the data scientists will be able to closely collaborate on research and coming up with the right solutions that can later on be included in the product. It is vital to carefully balance having your data scientists work in a research friendly environment with making sure they are closely working together with the product development teams.
The disadvantage of this approach is that the team will work in isolation and might not be connected to the relevant business needs. As such, this setup is not recommended in day to day business operations.
To sum up, there is no right or wrong setup. Every mentioned model has its pros and cons and the key to successfully ramping up your data science efforts is correctly understanding the current analytics maturity of your company.
As such, our recommendation is to firstly ask help for our help here and we will accurately diagnose your situation and help you get started. Then tackle small projects by deploying data science resources in key areas of your company.
As you scale up the efforts, centralize the resources into a Data Science Center of Excellence that would constantly and agilely support your large business initiatives and once you come up with a winning recipe make a product out of it and sell it to the others.