Advances in Artificial Intelligence (AI) in Language Processing (e.g. chatbots, smart speakers) and Image Processing (e.g. vision in autonomous vehicles) have been astounding. Fortunately, the magic does not stop there. Deep Learning models have also been very effective in other domains and have produced wonders when it comes to reducing company spend and boosting customer experience. Concepts such as Robotic Process Automation (RPA) and Product Recommenders are mere examples and have led to tremendous productivity gains in companies large and small.
While the promises are abound, the number of failures in AI projects are not. Adherence to “Best Practices” has never (in my career) meant more than the present. The challenge of adopting AI is building an end-to-end pipeline that is well suited to the needs and is bounded by limitations of the host. The challenge starts with data ingestion and only end when the suitable API calls are made (sort of). Aside from performance; factors such as security, cost-effectiveness, lack of bias, explainability, and compliance are just a few entries in a long list of requirements, and none can be ignored. Tools are plentiful, options are extensive, and dimensionality of the problem is immense. The saying that “start small and gradually add to the top” is truly valid in this context. To address complexity and risk, launching a major program with a pilot phase is the way to go. One very effective method to do this is to utilize prebuilt Deep Learning pipelines made out of a solid and proven core bolted to highly flexible periphery adaptable to the environment.
We have made immense progress in building such customizable platforms and have proven their effectiveness. Our initial focus has been in the following use cases:
- Forecasting – Forecasting key metrics such as ‘Product Demand’, ‘Sales Volume’, ‘Expected Consumption’, and ‘Projected Traffic’ is an absolute necessity for any business and that should explain the popularity of ‘Time Series Forecasting’ techniques. The nuances in this science have ‘AI’ written all over it. An ensemble of Neural Networks working in conjunction with older traditional models have recently produced remarkable and unprecedented results
- Customer Churn Prediction – The cost of acquiring a new customer is far greater than the cost of preserving an ‘unhappy’ one. Machine Learning models have been astonishingly effective in detecting early indications of dissatisfaction and can prompt vendors to intervene quickly to save customers
- Knowledge Base Search Engine – A sizable chunk of most knowledge workers’ time is spent on searching for fresh, relevant company-specific information. Helping employees find the most relevant and timely data can save a tremendous amount of time. Natural Language Processing techniques have been extremely effective in sifting through, prioritizing, and indexing huge knowledge bases that are stored non-homogeneously in multiple silos. This should explain why they have been so effective in responding to search queries
Fortunately, there are many resources that can ease the AI adoption journey. All three major cloud computing vendors (AWS, Google Cloud, and Microsoft Azure) offer tremendous tools and resources that can dramatically ease the burden. Of course, this comes with a price.
Finally, it is time to draw some conclusions. Presenting philosophical views is my way of signaling the end of an article and this time is not an exception. I can very confidently stay behind the following statement. If you are in possession of historical data about your operation, customers, finances, production, and competitors; AI can find and extract highly valuable and actionable insights. It has been done many times by many companies and you can do it too. We can certainly help.