Skip to main content
Skip to footer

Nagarajan Karuppiah

Head of ML and AI CoE (Retail)

Lured by success stories, enterprises are willing to adopt AI. According to a recent report, adoption of artificial intelligence (AI) in retail is projected to leap from 40% to 80% in three years. However, Gartner Inc. predicted that 75% of AI projects will remain at the prototype level as AI experts and organizational functions cannot engage in a productive dialogue—a surprising forecast despite all the buzz and talk about AI technologies.

Unclear business cases, lack of quality data, and technical infrastructure challenges are some common barriers to successful AI implementation. Here are seven ways how enterprises can escape from getting stuck in the proof of concept stage and scale their AI projects with confidence:

1. Find the ‘Why’: Identify AI Business Cases

Unclear business cases are significant barriers to AI implementation than lack of technology capabilities per se. Enterprises need to identify the right problem statements to demonstrate quick business value. To build a strong business case, there should be clarity on the following: what is the objective of the project; what technology framework will you leverage; who will use the solution; and how will you deliver it.

2. Acquire the Right Data

While most organizations use their enterprise data to deliver intelligence to AI projects, those organizations leveraging external data can deliver high quality predictions and early warning triggers to business teams. External data can be acquired through external contracts with market intelligence companies such as Nielsen, social data aggregators, public data sets available through government organizations, and APIs that publish weather data, commodity prices, micro and macro-economic indicators.

3. Escape the Proof-of-Concept Purgatory

For successful AI programs, focusing on the granular aspects is key. Once the AI goal is identified, acquire clean data at the required granularity and apply the right technique for constructing the AI model. Iterations can minimize errors and help in accurate outcomes. This can be followed by business value amplification and the generalized models can be leveraged for subsequent reuse.

To keep the momentum going, you need to keep pace with the rapidly changing nature of the underlying technology, look at the right avenues for continuous innovation, and overcome the fear of failure and lack of trust within the team.

4. Track your Progress

Determining the value of an AI project requires close evaluation of on-the-ground scenarios and trade-off analysis to arrive at the base opportunity value (BOV) of the project. Broadly, BOV computation focusses on direct benefits such as campaign effectiveness, marketing ROI, revenue uplift, margin uplift, and productivity improvements and optimization.

Organizations can visualize whether they are making progress on their AI journey by analyzing the following:

  • How many new ideas did they try in the last few months?

  • How many of these ideas were successful from a business perspective?

  • And, how many ideas were deployed?

5. Mitigate Technical Debt

Enterprises incur a lot of technical debt when they succumb to the pressure to move fast. While it is almost impossible to eradicate technical debt altogether, here are a few guidelines to reduce or mitigate it:

  • Legacy-based enterprises can leverage semantic views or data virtualization options by keeping the data where it is and develop wrapper views on top of it for any AI projects. Performance considerations are vital in this approach and it works for lesser or moderate volume of data These enterprises should also explore edge computing, pushing most of the AI models execution very close to where they are needed.

  • Modernization path enterprises can adopt microservices-based data exposure at different applications for insights. Additionally, they must use cloud-native AI solutions and out-of-the-box (OOTB) algorithms.

  • Digital-ready enterprises can leverage their data pipelines for efficient data provisioning for AI projects. These enterprises should leverage the data and analytical fabric for end-to-end seamless AI projects development, feature engineering, model training, and deployments.

6. Embrace Hardware Innovations

Advancements in hardware is speeding up the run time of algorithms, making it possible for retailers to get real-time contextual insights quicker than ever before. Higher adoption of graphics processing units (GPUs) delivers greater performance results and shortens the learning time significantly. IBM’s and Intel’s neuro chips, Microsoft’s field-programmable gate arrays (FPGA), and Google’s tensor processing units (TPUs) and recent advancement in quantum computing will play a significant hardware role in the future of AI projects.

7. Fail Fast, Fail Early

Now-a-days, most enterprises across the globe adopt agile, expecting incremental results. Incubation period in AI projects are little longer and retailers must be ready to hit some blind alleys. They need to decompose the problem into smaller chunks for easy experiments, identify the exit criteria for each logical phase of experiments, leverage open source algorithms, extrapolate business outcomes, and be cost conscious. Even if the experiment fails, they need to document the results and reasons.

Conclusion

AI is fast making inroads in every business and will continue to disrupt the status quo over the next couple of years. However, picking the right spot to apply AI is a key determinant of program success. Now, every moment of an AI project is captured and made available for all programs centered around AI for good goals to solve pressing societal challenges. Unicorn startups are already leveraging this data for applications such as more accurate weather forecasts to guide farmers for their day-to-day operations. Data availability, hardware innovations, and newer set of deep learning algorithms will continue to dominate the marketplace and create avenues for exploring newer business opportunities and scale for higher adoption.

About the author

Nagarajan Karuppiah
Nagarajan Karuppiah heads the ML and AI CoE, Retail. He has been with TCS for over 20 years and has held various leadership positions across retail accounts. An active member of IEEE, he has published white papers on diverse topics from enterprise data warehouses to decision sciences. Nagarajan is currently focusing on building ML solutions, NLP for personalization and recommender systems for retailers. He can be reached at LinkedIn: https://www.linkedin.com/in/nagarajank/?originalSubdomain=in
Contact Contact