Share this post:
Artificial intelligence (AI), technically known as “machine intelligence,” is a buzzword in the tech industry today. Businesses are finding innovative ways of mining data to generate insights that help them understand their customers better. In an era in which marketing rules, business leaders see a strong need to personalize marketing content to make it more effective.
Step back 10 years, when AI wasn’t a buzzword because there wasn’t adequate data available to understand people, products, networks and so on. With the explosion in smartphone usage, the adoption of Internet of Things (IoT) devices and extensive social media usage, that has changed completely. Most business “know” their customers very well now—especially their personal preferences, usage of products and services, social networks and so on. Information like this can be a gold mine for businesses. The question is, how do you mine the data effectively to get relevant insights and use those insights?
The most common artificial intelligence questions for businesses
We had the privilege of running an AI workshop for a large client recently, and the experience taught us a few things. We’ve also been meeting with ecosystem partners, clients with early interest in AI and independent software vendors (ISVs) that are trying to make a mark in this space.
In all these discussions, two important questions that come up are:
- What are the AI models we should use for X use case?
- How do we mine useful data and transform it so that AI models can consume the information easily?
There are no easy answers to these questions; however, a methodical approach can help address them. In this two-part blog series, we’ll look at a possible approach to building an AI strategy and roadmap for your business.
Artificial intelligence: You need a strategy
Here are some starting points to consider for building an AI strategy:
- Begin with one or two AI use cases for your organization. Your criteria to shortlist them could emphasize the availability of the data needed to build the use case.
- Is there an expected outcome defined in the data? Does historical data show conclusive outcomes? This is useful for prediction use cases.
- Are you looking to classify your customers by behavior, preferences, usage or needs? An outcome may not be needed; however, it’s important to ensure that the current data has the required features to help classify the groups.
- The next and most important step is to have a data scientist conduct a study on the availability of the data, as well as the extraction and transformation methods that are practical and suitable to your existing environment. Do you have the right software for deep learning? The data scientist should be supported by domain specialists who can call out relevant versus irrelevant data. This is referred to as “feature engineering.”
- Feature engineering is in the data science profession for machine learning/deep learning (ML/DL) modeling and is considered a critical task that ensures ML/DL algorithms produce the expected results. This involves applying testing methods to understand the impact of a feature over the expected outcome. There are many statistical models that can help accomplish this, and a data scientist and ML/DL specialist can brainstorm together and agree on the best option. Your domain specialists play a very important role in validating the results of feature engineering.
This is just the beginning of forming an AI strategy. Learn more here. And read my second blog post here.