Data Products commercialization Framework

With democratization of AI/ML it has been seemingly difficult to differentiate between a production ready data product Vs a research toy. Starting an data science experiment and converting the same into a real world, commercial ready product is quite a journey. Even though there is no full proof way to decide if your data product is production ready solution, having a maturity framework to guide in your journey always helps.

Quadrants of Maturity

You could use the above framework like a checklist to create a map for yourself. Data science is team sport and the items on the x and y axis signify resources either your team or organization as whole bring to the table in form of commitment to your efforts. It helps to plot where do you stand in your journey to take your experiments/ research work to commercialization.

X Axis/ Investment Axis

If you intend to create data products for commercial purpose it would be a good idea to start working with collaboration partner who are a good representative of the market where you would want to commercialize the product. That in turn would lead to designing a crisp problem statement and it is important to make it very precise and crisp. Which would lend itself well to create a business value or bang for buck. Even though your experiment might be cool, but if it is not solving a unmet need that would not have any stickiness for end user or low willingness to pay. One of the most important aspect of commercialization of a data product is the legal framework under which it has been created and monetized. Data is the new oil but comes with high risk if not managed well. All of this comes with a big price tag, right from acquiring right kind of data and even after acquisition be able to curate that data, use expensive resources like GPU etc to train the models and surface them in a right way. Without the knowledge of enough funding it would just end being a cool experiment and sometimes not even that. With funding comes the right directive to even use the data products at right place and within right workflows. Sometimes management directive is the only way you make data product adoption see the light of day. The last 2 items on the checklist would depend which industry are you creating your data products for, regulatory compliance is getting important and you would need to have someone with decision making capability to to support your work get to commercialization.

Y axis / Data Competency

With acquisition and accumulation of data to create data products comes the question of data competency. Just because you have developed traditional software in the past does not qualify you as data competent team. Getting right data sets would define the shelf life of the data product. Having only synthetic data sets can only get you so far. That brings in the capability of curating the data sets, labeling the data and ability to create meta data directory which can be used to bootstrap your experiments. As said before data science and data product development is a team sport, you need to possess all the skills yourself or have a partner who can be a SME in the domain you are trying to solve the problem. With the ever changing landscape of technology in the space, tech competency is a must. Team members with a natural ability to adapt and learn new practices and tools. If your team is truly thinking about commercialization of data products there needs to be a effort for standardization of practices and tools without which scaling goes out of window. Once the data product is ready as an experiment for it to succeed you would need a last mile delivery for promotion of your data product. Deeper integrations of models can facilitate a feedback mechanism to help the data product learn continuously. Designing a data exhaust or feedback for your data product is extremely crucial for the stickiness of your data product. It could so happen that the last mile of delivery is been built by a different team in that situation the discussion of having a feedback channel should be done upfront.

Outcomes

Depending on how many items you can check from list, the outcome of your experiment would fall in either quadrant and can be tagged in four categories which are quite self explanatory. Wish you all a #Grandslam outcome.

1. #ResearchToys
2. #LostOpportunity
3. #FalsePositive
4. #Grandslam