by Nagendra Shishodia
NEW YORK – Organizations have shown a lot of excitement for AI and its ability to transform business functions. This was shown by a survey of over 800 executives around the world EXL carried out in partnership with Harvard Business Review Analytics Services, with results indicating that nearly one-third of organizations anticipated deploying artificial intelligence by the end of 2018.
However, many corporations struggle with developing and implementing AI solutions successfully. They falter at critical steps, and fail to generate expected ROI. Organizations need to raise their AI IQ to avoid common pitfalls while working on AI initiatives.
Identify the right AI use cases
A key task for businesses is to first find appropriate AI use cases. There are two important factors that should be considered when making this decision. First, what is the potential business value that can be generated by using AI? Second, can the organization in its current state develop the needed AI solution?
Business owners, data scientists and AI practitioners need to work together during this process and identify high impact use cases where AI is likely to create value. They should also align on success criteria for the select use cases.
Development of an AI solution
There are a few components which are absolutely critical for the development of AI solution: data, strong computing infrastructure, and expertise in data science. The initiative is likely to fail if any of these components is missing.
Data Science Talent
Organizations need to ensure that they have access to the best data scientists who can work on the relevant data and infrastructure. The challenge here is that the data science skillset is scarce and in high demand, often making staffing of data scientists a bottleneck in the process.
Some firms attempt to circumvent this challenge by using vendors or service providers. A troubling issue here is that despite many vendors claiming that they have AI expertise, only a few of them have actually developed AI solutions at scale.
Evaluating vendors on their capabilities to develop AI is critical before deciding to partner with them. Organizations that decide to engage in a proof of concept (POC) to evaluate vendor capabilities should make sure that the PoC is structured in a way that AI capabilities can be evaluated in an objective manner. Accomplishing this means giving access to a large amount of data with enough variability to the vendors, and then testing the accuracy of their solution on a wide variety of inputs. Highly successful PoCs on a very narrow scope may not be successful when the solution is scaled. Having an experienced data scientist involved during the PoC and evaluation process will help judge vendor capabilities much better.
Data and Infrastructure
Irrespective of what use case is selected, training AI models requires access to lots of training data. This data has to be created and labeled if it is not readily available at the start of the AI project, which can be a time consuming and expensive exercise. Use cases for which there is no easy way to get access to relevant data should not be prioritized for AI initiatives.
Data scientists will develop and evaluate different type of models on the data in an iterative manner. For certain type of AI models, powerful compute infrastructure enabled by GPU servers or cloud computing may be required. Businesses which are serious about leveraging AI in their business need to think strategically about their infrastructure, which will be the backbone of development and deployment of AI solutions.
Deploying an AI solution
Once the solution is developed and tested, deploying it in production will require expertise of machine learning engineers and big data engineers. Software engineers without experience in deploying machine learning models may not make optimal choices when creating data pipelines. This can cause production issues and will hinder adoption of a solution within the organization.
Need for collaboration
For successful AI initiatives, domain experts, data scientists, and machine learning engineers need to collaborate. This can prove challenging if they are working within different teams across the organization, but the success or failure of the project will be determined based on how effectively a business can have these teams work together.
An AI assisted workforce can help organizations improve performance, increase revenue and generate deeper insights. By orchestrating AI and human resources, businesses can see results far beyond pure technology upgrades or staffing increases can create on their own.
To reach that point, companies need to make several decisions around which use case to work on, how to assemble the team, which vendors to work with, how to create relevant data, what infrastructure to work on, and how to integrate the solution in the work flow – or in summary they need to raise their AI IQ.
Nagendra leads the development of AI and ML solutions at EXL. His focus has been on developing solutions that enable better decision making and efficiency gains through the use of Machine Learning, NLP, Big Data and more recently Deep Learning technologies. Nagendra has over 17 years of experience developing such solutions for global firms in healthcare, insurance, banking, retail and travel verticals.