Machine learning nuggets to draw users
Machine Learning (ML) has crept into our lives in small, but increasing ways that we don’t realise and perhaps even in likeable ways…maybe even take them for granted.
Machine Learning doesn't always need to be about super complicated/heavy duty algorithms. (Not to say that any of the examples I have below are easy).
In my observation, in the initial phase, ML based results, that have too much at stake, tend to shove away users as they prefer to follow their own gut than an algorithmic output they don't trust. We risk loosing the opportunity to create new means for users to interact and work with their systems/solutions.
Instead, if we introduced ML based features slowly and in small bits, it could contribute in drawing the users to the system to precisely leverage these features to ease their work.
Below are a few examples from the (consumer) software world that I particularly like:
Predictive Keyboard
While in special cases this can be annoying but for 80% of the time, it makes it quite easy to type using this feature.
Predictive replies
LinkedIn has this really easy way to use their messaging feature. It makes it really fast to respond to people.
Predictive Sizes
This is one of the latest I stumbled upon. While ordering shoes using Zalando.de they recommend the size I should order.
I feel, this might be a bit more complex than the previous two examples. However, if this is based on my purchasing history and returns due to sizes, how the shoe is designed (e.g. narrow/wide etc.) I could probably take their recommendation. (This shoe is currently in my cart — maybe I should test the algorithm :-P )
The base of all of this is ofcourse the quality of the historical data. Plus the models etc.
The world of Enterprise software is in a different stage. Different industries are at different phases of adopting and productively using ML based results.
In industries that are just testing waters, I really believe ML should be introduced in small ways — with no critical (business) consequences incase of inaccurate results/wrong decisions based on algorithmic results. E.g. Suggesting/predicting logical (based on historical data/frequency etc.) next tasks; pulling and presenting disparate but contextually related information etc. ML needs to creep in almost in un-noticable ways.
Its important to address the fear of users to base their decisions on ML based results. Especially when these users are almost always experts in what they do.
Food for thought!