Order Management & Machine Learning

We are in a world which is being redefined by machine learning advances. Whole product ecosystems are being developed privately or open source which are ushering us in next era. What that era will be and how it looks like is anyone's guess right now (see below picture courtesy waitbutwhy.com). But AI or Machine learning definitely has its place in this future.

Why we can not see AI taking over?

We have some AI.. rule based or algorithmic

But what does that mean for traditional enterprise applications which exist now? Most of these applications can be said to be static or rule based AI. Rule based AI or algorithmic AI has been in place for decades. Typical examples of this will be optimized sourcing or scheduling with objective of minimum waiting time for customer and lower total cost. Typical modern days applications will use myriad of configurations and settings like lead times, calendars, handling cost, inventory cost, transit costs etc. 

But is it self correcting?

But the fact is this. These applications do not reorganize themselves after executing some transactions and apply that learning. In a typical order management use case once a order has been accepted and fulfilled by the company it contributes to various macro and micro facts or statistics. If the warehouses have modern warehouse management systems they will be capturing various costs and times at all the points of a particular order. In some of the systems it will be even possible to attribute macro costs to particular order.

All of these find their place in various operational reports and financial statements. After that fact some decisions will be taken and it should then be possible tune the applications with some new figures.

But it is too late..

By the time the machinery which is responsible for analysis and define operational parameters 6 months may be 1 year will be gone, as typically that is the time frame of planning and budgeting etc in organizations. The corrections which should have been done months ago will now be done, and situation now might have changed. Order patterns, Supply disruptions, transport prices etc which are really probabilistic and need to be considered right now will have delayed response by organization resulting in high costs and low customer satisfactions. 

Direction forward?

Enterprise application will need to leave rule based AI and move toward soft AI based architecture. The function which needs to learn and correct should never be static and it should get feedback to train itself. Not leaning towards any particular AI theory, it can be anything from neural networks to predictive models, but we have the right ingredients, transactions, and feedback which is how any learning mechanism learns about environment.

Final word

Product managers now need to strategize about how their years of work in making those applications can be saved and still be relevant in future. It is really difficult to think about so much uncertainty about how it is going to turn out but we can see some patterns emerging which are interesting.

1. Open Platforms Google TensorFlow, Torch.. to name a few
2. API Economy
3. On premise platforms Apache Spark etc.

In my opinion it is not a typical build vs buy decision anymore. Build vs Buy will not impact the ultimate outcome. It is more about visualizing your product with AI and decide based on what makes more sense for your product choosing right architecture. what ever you choose you will have both options, always :)