Going from data science algorithms to delivering business products
People buy finished products and not its components. An iPhone has lot of great components such as taptic engine, sensor, screen module etc.. All these components are an engineering marvel. There is lot of technology which has gone into these individual components. However we always prefer buying the iPhone and as end users we do not really care about components which go in it
The similar question in data science world is what do business users prefer — Bunch of Algorithms or Business products which give business outcomes — surely the answer is the later.
The subject of this article is to show the why, how, what of going from algorithms to business products. I have borrowed the why, how, what framework from Simon Sinek book Start With Why.
Top 4 Most Popular Ai Articles:
1. Five Free AI Tools You Can Use to be More Productive at Work Today
2. Face recognition: realtime masks development
3. Understanding and building Generative Adversarial Networks(GANs)
4. Six AI Subscriptions to keep you Informed
WHY to make data science enabled business products?
Before jumping into the subject, important question to ask is why to make such business products. There are various reasons for this, but I mention some prominent ones.
Firstly, data science driven business products represent a multi-billion dollar opportunity. For example, an equipment manufacturer can develop predictive maintenance software product and sell it as a service to its customers. Medical clinics who collect patients data can develop products which gives patient health index paid service to its patients as well as insurers. There are various such examples which can justify the business value. It can also bring a completely new strategic direction to companies
Secondly, a business product brings the business focus to data science. How many times have we seen that data science work just stays in R&D or development phase. It never justifies the money spent on it because it does not give any business value. However if the objective is to make a business product, all data science efforts are channeled towards a right cause.
Third point is that developing such a product brings people from various disciplines together. Creating such a product is not only data science work, but you will need skills such as designers, software engineers, product managers etc.. Enterprise success relies on how multi-skill people can work together. Building products helps people of different background come together to build something great
HOW should such a product look like?
Steve Jobs has very famously said: “You’ve got to start with the customer experience and work back the technology — and not the other way around”. This quote is true also for data science based products. It should consider the customer experience first and not the algorithms. One of the most effective ways to consider customer experience is through the methodology called Design Thinking.
Shown here is an example of a data science driven app. On the left you can see the app without any design thinking applied. It was a complicated app to use and was not very successful. On the right image is the revamped version of the app with design thinking principles. It is much more intuitive
There are various sources which explain on how to use design thinking. They key is to design the user experience of the business product and then work back the necessary algorithms
WHAT should it contain?
Once you have defined the WHY and HOW, the next step is to define what should the business product contain from technical perspective. One of the successful ways of defining the technical architecture of a data science powered business product is to use a full-stack analytical architecture.
The front end is the user interface layer. This is designed based on design thinking principles. When the user interacts with the product, it initiates a REST API call which will use the data science models. These models are generally scoring models. The models have to be stored in some kind of model repository. The data science model would then, in many case, interact with a persistent data store
There are various technologies to implement the full-stack architecture. The exact choice of technology would depend on the use-case
We have seen till now the WHY, HOW and WHAT. Now let us see some other points to consider.
First is the team organization. When you are developing business products, the team roles are very different from that of data science use-case project. You will of-course need data scientists. But as the end result is a finished business product, you will need roles which are required during packaged software development such UI/UX resources for design thinking and front-end , QA for quality assurance, release managers to manage the different versions of software, product owner and other software development related roles.
The second and very important point to consider is the subject in this article represents a way in which data scientist can evolve from algorithm developer to contributing to business products. This is a very enriching feeling, as you are directly contributing to customer experience as well as revenue generation. I am sure that working on data science enabled business products will represent a huge career opportunity and success for all data scientist.