Introducing Machine Learning or building intelligence into the product(s) has caught every enterprise’s imagination. If they haven’t, then they possibly are in trouble sooner than later of losing out to their competitors who have already taken steps to build this capability in their existing and future systems. So, with this, we can comfortably say one thing, that the adoption of machine learning in the enterprise space is no longer a luxury but a necessity for survival and more so in the post-pandemic economy of 2020.
Now since we have taken the argument of whether enterprises are going for adopting machine learning or not out of our way, we need to look into the aspect that, what would it take to build machine learning products from a Product Manager’s perspective. In this blog I intend to present the building blocks of a Machine Learning based product from a Product Manager’s standpoint.
Build intelligence for a solution and not for inducing magic:
Machine Learning has come a long way and has got commoditized to a larger extent. But still, the community is kind of divided into three camps. First are of those who expect magic out of machine learning, the second camp encompasses those who skeptically believe machine learning is here but it will take longer to realize any real value out of it today and then there is a third camp which understands machine learning and has a realistic expectation to solve a problem. As a Product Manager, you should be in the third camp or your move should be in that direction, else it is better to move out of this space.
Note: The second camp is also not an option, as that would be a path to become the dinosaur of the future.
The question now is, what you should and should not be doing if you belong to the third camp?
As a Product Manager, you should identify the problems and see machine learning as a solution. Even if the organizational direction is to “Make Products Intelligent”, you as a Product Manager should still be looking for the problems that can be better solved with the help of machine learning or how the user experience can be improved with the help of the machine learning. Below are a few, absolute no as an approach:
- We should not start on a cool looking machine learning project first and then try to find and retrofit a problem to the solution that we think we have already designed, because it works out very costly to fill the gaps later.
- Do not start by finding some business data and then asking your data science team to deliver some magic out of it.
- Never ever try implementing machine learning just to align with the leadership vision, because customers relate to their problems and not to your leadership’s vision.
We should have a framework-based approach to deliver value for our product with the help of machine learning. The user problem should be the starting point and also the focal point of the solution. It is very important to define the problem, identify the use cases, and understand your users. And based on this you need to identify the scenarios where machine intelligence can complement human intelligence and, in the process, define the boundary of the problem where this needs the intelligence need to be built before even you get set to engineer it. Make it an iterative process till the point you do not have the acceptable solution in place. Below is a process map for the thinking framework:
If you love your machine learning model, then think about data:
This mode of vehicle called Machine Learning runs on the fuel called data. And like any machine to work uninterrupted and well performing, the quality and the quantity of the fuel have to meet the desired criteria.
In the case of the machine learning systems, as a Product Manager, you will have to collaborate closely with the Product Architect and Engineering team to ensure that:
- The data storage required to store data for machine learning model building is performant, reliable, and cost-effective.
- Define the benchmarks for the data availability, data quality, and data retention time frame.
- Define the user experience for the data collection.
Above mentioned three points are critical success criteria for the machine learning systems. And as a Product Manager, you will have to be aware, involve, and define the requirements around it.
Data Collection: As a Product Manager you will have to ensure that the user experience for setting up and kick-starting the data collection is seamless. This is the first and very critical step and if this is tedious, exhaustive, or ineffective, it would directly impact your product adoption. A bad user experience at this stage will ensure that the customers get frustrated and never ever try your actual product and see its value unpacking for them.
Data Storage: Based on the scale of the problem that you are solving, if you do not get the performance and reliability of your data storage right then you will never be able to deliver a good user experience, while non-cost optimal storage can impact your product pricing which in turn can either render the product uncompetitive or unsustainable in the market, based on the compromise you make.
Data: Carefully define the requirements for the data availability latency tolerance, from source to right into your data lake. Define the outcomes and scenarios for the machine learning model well to ensure that the engineering factors in the checks and balances for the data quality and retention.
Have a road map for accounting user feedback:
Let us accept the fact that, we may not have the machine learning model ground running on the first try. It will have to be tuned and perfected based on the feedback from the user.
There is no denying that talking to customers/users can give a wealth of information for us Product Managers from the use case and user scenario standpoint. But all of it to translate into a perfect or near-perfect machine learning model calibrated to perform as per the user requirement and data you will need to factor in functionality in the product to capture the user feedback data and then tune the model based on it. Avoiding this will, lead to having longer periods for the system to stabilize and the users being able to unlock the value in it. So instead of testing your user’s patience, plan to build feedback capturing right from day one. And gradually evolve it to enable auto-tuning of the models as well to deliver value continuously improve upon it in faster iterations.
Conclusion:
Taking these as building blocks for conceptualizing, executing, and delivering machine learning-based intelligence into the products, we will be able to assess the value proposition of building intelligence into solution, the cost to benefit ratio of building it, and set the right expectation for the engineering, sales/marketing, and the end-users.
Cheers!