Decoding Data Products

Big Data is, indisputably, the fundamental building block of the digital transformation era, ranging from a static websites, single vendor e-commerce sites, multi-vendor SaaS platforms, Web3 applications or Applications of Artificial Intelligence and Machine Learning, data acts either as the means to an end or core facilitator of the products/platforms.

In context of some of the widely consumed digital products in our everyday lives, data now has gained the business viability to exist in the state of a product or as-a-product. In this article we lay emphasis on the understanding of data products using google as a case study.

What is Data Product?

As defined by Simon O'Regan, Designing Data Products, data product can be defined as products whose primary objective is to use data to facilitate an end goal. So can we say Wikipedia is a data product? Nope, with wiki we assimilate data and publish it for learning or information purposes, data is neither the objective not the means to the end, it only facilitates the platform. However, the wiki search tool is a data product we enables web-mining and data retrieval within the Wikipedia domain.

Type of Data Products

To further understand how data products can be identified, lets understand their functional categories, here we go:

?Raw Data: Products that enable Data Collection, Storage and Retrieval without pre-processing of any form. Example Google Forms or Survey Monkey Forms that enable gathering of raw data that is captured in a structural manner for further analytics.

?Derived Data: Product that facilitates Data Manipulation and Augmentation without the use of Machine Learning. Example: API's that enable data labelling, Data Cleaning Python Libraries such as NLTK, RE are used to pre-process raw textual data for advanced natural language processing purposes.

?Algorithms: Products that use feature extraction, modelling and deployment of Machine Learning/Deep Learning or Neural Networks/Transfer Learning to analyze data. Example: Pre-trained Transfer Learning models that can be reused for Semantic Classification or Sentiment Analysis.

?Decision Support: Product that provide Data Visualization/ Dashboarding/ Analytics/ Business Intelligence for end-user decision making. Example: Power BI , Tableau, Google Analytics, Python Library Matplotlib, Seaborne.

?Automated Decision Making: Recommendation systems to decide on behalf of the end-user. Example: Netflix watchlist and Spotify playlist recommendations based on your recent activities on the platform.

Based on these 5 categories we can now correlate few examples of Google's offerings to discussion how data products actually work.

Google as a Case Study to understand data products

Lets consider Gmail, does it qualify as a data product? Gmail does in fact possess some of the most important information of an individual of our generation: our bank statements, our pay-checks, our tax redemption documents, or a sleazy-scandalous email exchange. However, the answer is no Gmail is not a data product by itself, basically it is an email service for asynchronous communication between two parties but it doesn't account to be a data product. However, the sorting feature that sorts important and not-important emails can be accounted for as a data product, because this leverages natural language processing for sorting.

What about GDrive? Does it qualify to be a data product? Well, Google Drive is a cloud storage platform which is primarily used to store and retrieve data. This could be an example of Software-as-a-Service, rather than a data product. So, the inference is, though Gdrive deals with data as its core facilitator, it would not be a classic example of a data product.

Google Photos is represents another category of data products with Algorithm-as-a-service data product. Do you remember waking up on your birthday to a collage of all your previous birthday photos made into a timeseries video by google photos for you to consume? Well, essentially the features of each image with your facial features is being extracted and classification is being done based on features extracted, pixel quality, time frame and may be some additional features, then matched to similar features to create a montage of images.

Google Image Search is another example of Algorithm-as-a-Service is to retrieve relevant images for a given search query. For instance, you spot a ravishing coat on the window of a Burberry store, you are too prudent to go inside and look the store assistant in the eye and ask for the price. So you just click and image search for it, most probably google will show the product listings on uk.burberry.com or harrods.uk, with its price and if you are indeed lucky a discount on it as well!

Google Analytics is certainly a data product, it essential is a cloud platform-as-a-service that is available for data analytics and visualization. The important thing to remember here is as follows: while we have taken design-decisions in data collection, derivation of new data, in choosing what data to display and how to display it, the user is still tasked with interpreting the data themselves. They are in control of the decision to act (or not act) on that data.

Let's consider Google Search as a data product. I think google search engine is the most sophisticated examples of a data product, it is one of top search engines, known for contextual information retrieval. Suppose we use pun intended words(Polysemic) or use words that have opposite means in different contexts(Polarity Divergent), the google search engine algorithm is wired to contextual evaluate the relevant search outputs based on user behavior, user persona and user search history using natural language processing. It also contributes to automatic decision making on behalf of the users by providing recommendations no sooner than the user has started typing the search queries and retrieved results are benchmarked based standard Search Engine Optimization metrics, such as number of views, quality of content and richness in backlinks.

In this article we have explore the functional categorization of data products and their application in the our everyday scenarios. We can trace the lifecycle of data from its raw form, to derived form upon feature extraction or text augmentation, further analysis using with machine learning, deep learning or neural network algorithms, visualization and decision making for end user consumption. This gives scope to further discussions on how different data-products are to data-as a-product.

Reference: