Simplifying Machine Learning (ML) and Artificial Intelligence (AI) – Perspective on Enterprise Adoption

It is an established fact that the future impact of AI or ML will be phenomenal and unprecedented across every Industry with many use cases already tried and being envisaged. It is touted to be a key differentiation that organizations would deploy to thwart their competition in near to long term future.

While the classic distinction between the two (AI & ML) has been made around the web extensively and hopefully well understood, my attempt in this blog is to highlight some trends/success factors across Enterprise segment adoption. Feel free to comment or add per your practical experience.

ML/AI Adoption – Critical Success Factors for Enterprise

Listing down a few factors that are critical for success in terms of delivering a desired ROI for an AI initiative at most organizations.

1.)   Use case Identification – This is a major decision point for enterprises in order to maximize the investment return and have a demonstrable ROI in short or medium term. Mission critical processes/functions where higher degree of accuracy is required, AI might fall short in its current form given the prediction levels today and investment required in training. Many times, projects are shelved due to this reason and Enterprises think that AI is not ready yet. It is OK to fail but analyzing the use case carefully can reduce or eliminate the probability of failure.

Then there are other factors like dimensionality of the input parameters so that the learning model does not gets confused with loads of input data that has multiple dimensions and variability.

2.)   Clear benefits case definition – Enterprises are deploying AI/ML for various business benefits including predictive analytics, automation, decision making and deriving insights from data in turn driving efficiencies/revenue. It is very important to set the right expectation with the stakeholders for the success of the program and also keep in mind that if a particular problem can be solved thru traditional app/program or automation with proven technologies then why use cannon to kill a fly.

3.)   Data, data, data – Given the success of data-driven digital companies like Google/FB/Amazon et al, enterprises are investing heavily in data collection, storage & governance. This is another major area where an organization has to build a seamless data layer for providing good quality data (Structured & Unstructured) to train & optimize the models properly so that the predictions/decisions are relevant. Organizations with loads of quality data will have an edge here, case in the point are self-driven vehicles which are today tested across the length and breadth of various countries to generate tons of data/scenarios. This is where companies like Tesla will have a big advantage as they run a datacenter on wheels and collecting tons of data to train their AI engine for real time scenarios.

However, practically we know organizations today are not able to harness structured data fully leave alone the unstructured data. Unstructured data use cases have to be evaluated carefully for ROI depending on the Industry use case, few examples or potential areas are digitizing documents in a template less manner to extract information for sense making, scanning thru images to predict/identify data or analyzing voice call records in a call center to derive insights. Here again results vary as per the quality of original data (scanned docs/image data) and language/dialect variance across voice data.

4.)   Horizontal Vs Vertical AI – Just thought of classifying AI applications in two ways, would term programs like Chat Bots (natural language processing), voice recognition engines or image processing AI software et al as horizontal AI programs and Vertical AI programs would be more domain/Industry focused software products that are embedding built in AI engine within them e.g. in Financial services Industry, AML vendors or Fraud detection software are becoming ever intelligent. In the horizontal AI space, there is a diverse combination of established consulting players & startups. Startups are mostly building narrow AI applications (sometimes just Industry focused to solve a particular problem like many Fintechs) either using open source packages or other industrialized packages. Another important consideration esp. in horizontal AI is for organizations to replicate these capabilities across vision, voice or language spectrum solving multiple problems in that domain rather than just one narrow area, which in turn would mean training models on varied data.

What is helping the cause is large computational power becoming available on cloud for these open source or industrialized packages/libraries to be available as APIs thus shortening the development cycle for complex models or neural networks for Deep Learning. Also, some of these models are unsupervised (not 100% today) and have self learning capabilities as well.

In nutshell, I believe days of advanced AI with conscious are still far away (haven’t found any definitive timeline so far) but selective adoption across various areas in different Industries is happening as we speak and will mature with data and trained models. Would welcome more perspectives to enhance mutual learning.