Making AI Success

 

Making AI Success

Share this:

Contents

Why AI projects fail?3

Unrealistic expectations4

Wrong problem statement4

Wrong understanding of business4

Wrong Data4

Data size4

Limited EDA (Exploratory Data Analysis)4

Wrong tools, model and people4

Wrong success metrics5

Conclusion:5

Making AI Success

In the last few years there has been a surge in the interest towards AI, which can be quantified based on the relevant Google searches over the past 5 years.

Enterprises are investing in AI and implementing AI projects. Based on a PwC research around 80% of companies have implemented AI in some capacity:

  • 20% of enterprises plan to deploy AI Enterprise wise
  • 21% have already implemented AI in multiple areas
  • 15% pPlan to deploy AI in multiple areas
  • 16% have implemented pilot projects within discrete areas
  • 22% are investigating use of AI

Source: https://www.pwc.com/us/en/services/consulting/library/artificial-intelligence-predictions-2019.html

The AI market is growing and set for exponential growth in the near future. The global business value derived from artificial intelligence (AI) is projected to reach $3.9 trillion in 2022, as predicted by Gartner Inc.

But Gartner also predicts that 85% of AI projects will fail. Which leads us to the next question on:

Why do AI projects fail?

There are multiple reasons due to which an AI project fails. Avoiding these can help you to make your AI/ML project a measurable success

Unrealistic expectations – with so much buzz going around AI and companies (service providers, ISVs, etc.) making AI sound very easy to achieve, businesses perceive AI as a magical wand that provides a simple solution to their challenges with minimal efforts.

Wrong problem statement – defining the problem statement is very important for the successful implementation of AI within the organization. Problem statements need to focus on the end goal of the AI project. This will lead to framing the actual outcome from the AI project in terms of increasing efficiency or reducing cost, more accurate. This is one of the key focus areas and can help make or break the AI initiatives.

Wrong understanding of business – understanding business processes is very important for building accurate machine learning models to solve business problems. Without understanding business processes, AI/ML projects face a much higher risk of failure.

Wrong Data – having wrong/ faulty data can lead to a “Garbage in, Garbage out” Scenario. This leads to wrong models & inaccurate predictions

Data size - other issues for AI/ML project failures is due to the availability of too little or too much input datasets for model building. Not enough data due to security reasons or unavailability due to other reasons can lead a limited number of features for model building and possible scenario of not having enough input data points for model building. Having too much data can have potential to fail the AI/ML project as it requires more processing power, time, removing feature, etc.

Limited EDA (Exploratory Data Analysis) – EDA is the most important step of any AI/ML project as it helps in building the understanding of the data, do statistical modeling for removing outliers, identifying significant features, feature engineering, etc. But at the same time, it’s the most ignored task of an AI/ML project. Putting insufficient resources for EDA leads to increased time & cost in the later stages of the AI/ML project.

Wrong tools, model and people – tools, models and people make the AI/ML project a success. Selecting the wrong tool for development can produce suboptimal analytics pipelines. Model performance depends on the domain and the underlying training dataset. It is important to use different models and multiple iterations to find the best model to solve the given problem while trading over accuracy, complexity & time.

Having the right team involving skills of Data engineers, Data science engineers, Data scientists, Architecture can make the AI/ML project a resounding success.

Wrong success metrics and not defining Failure metrics – setting up too rigid a success criteria & not defining failure metrics often leads to failure of the AI project. Identifying current heuristics is also very important which will tell you what will happen if AI is not implemented.

Enterprises would do well to keep the above possible pitfalls while planning their next AI/ML project and making it an overwhelming success.

To know more about our capabilities, you can reach out to us at info-Data&Analytics@sonata- software.com