Automation through Machine Learning solutions is a journey, not a one-time plug and play

As Machine Learning continues to be the buzz word across most organizations, the question which is asked quite often by the CXOs is - “Are there ready-made machine learning products/ solutions out there in the market which they can procure, deploy and start getting ROI right away through automating or transforming the existing process?” The answer is a big NO - not because there is any dearth of vendors offering machine learning solutions but because of the way and the mindset with which most of the business executives expect these solutions to work. When it comes to technology solutions and the software application deployment, the businesses are used to the typical way of buying solutions which are mostly pre-configured, can be plugged into an existing technology architecture and are ready to go right away. However, this is not the way ML solutions work, no matter how much a tech vendor might advertise that they do. In my opinion, deploying a robust ML based solution to automate/ transform an existing business process is a journey which takes time rather than a one time ready-to-go implementation. Below are two key fundamental reasons which are inherent to the mathematical nature of machine learning because of which these solutions can only work wonders over time, no matter how much they have been pre-trained and pre-configured:

1.    Uniqueness of an organization’s data: A lot of technology vendors are today advertising about ready-made plug and play industry specific ML/AI based solutions which have been trained and modelled on a large set of data relevant to the problem being solved. However, this is just a marketing gimmick because once a business entity deploys such solutions, these still take a lot of time to actually start generating a desired output with some accuracy. This is because of the fundamental reason that no matter how much generic data was used by the tech vendor to train the ML algorithms, when it comes to actual deployment of the solution, what really matters is training the solution on the business’s own specific enterprise data. Every business has its own unique data which truly encapsulates the nuance and complexity of its specific business operating model. And hence any ML solution takes time to train and get customized to this business data irrespective of all the pre-configured training models. For example - If a B2B business is using an ERP software application for managing its sales and it decides to integrate it with a third party Machine learning solution for predicting the probability of lead conversion or predicting future sales, then this third party solution will take time to learn from the sales data being entered in the ERP of the company by sales team and will accordingly customize the models to generate a desired output over time. 

2.    Accuracy improvement through iteration: If you ever talk to any data scientist who codes machine learning mathematical algorithms, they will tell you that there is no “best algorithm” which gives the perfect accuracy results right away to solve a problem. It is always an iterative approach where the machine learning solution initially returns an output which is less accurate and then over time learns from more and more data which is fed into the system or through the log of errors which it makes over time. This is exactly what happens when an organization deploys a third party or an internally developed solution. Irrespective of the initial training data, once the solution hits production environment and start processing real-time data, it will be less accurate to begin with and then over time it learns from the mistakes it makes (which are pointed to the system by data scientists reviewing the output) or as more data is fed into it. The perfect example of this is the popular Netflix’s ML based movie recommendation system which has been a huge success. While it works with amazing accuracy today, but the journey of this AI system started back in 2006 when Netflix organized a crowdsourcing ML competition called “The Netflix Prize” with a reward money of 1 million dollars for the team which would make the company’s existing recommendation engine 10% more accurate using machine learning algorithms. Ever since, the data scientists at Netflix have worked through all sorts of complex state-of-the-art ML techniques over all these years to finally reach the success which we see today. 

In conclusion, if a company wants to leverage the full potential of Machine Learning to get transformative business outcomes, it has to commit resources towards a long term journey instead of expecting instant results. The sooner any organization gets on this journey, the more will be the barrier of entry for other competitors who start late on the journey and hence higher will be differentiation for your business. Perfect ML solution does not come pre-packaged, it has to be made perfect over time by the company.