Share this post:
Every company houses an iceberg of data.
Above the surface floats the information that is used. Under the surface lies a massive collection of unused data that is practically invisible to employees. For many companies, the value of this data is almost incalculable. Unlocking it could be a game changer, and artificial intelligence (AI) could be the key.
To find out how, IBM Cloud Podcast hosts Ian Lynch and Steve Choquette got together with Nico Frantzen, a senior technical architect in the AI practice for IBM business partner Perficient, a large US IT consulting firm of 3,000 professionals. Among them is a crack team of 30 people dedicated to building AI solutions.
A sea of unstructured data
Thanks to a rich history of business intelligence systems, companies are adept at handling structured data: the row-and column-based stuff in spreadsheets and databases. What they struggle with is the sea of unstructured data that has sat dormant for years in word processor documents, videos, slide decks and audio recordings. AI can help unlock that buried data, explains Frantzen.
The low-hanging fruit is audio and video transcription. Call center recordings contain thousands of clues about customers’ likes, dislikes and concerns. An audio call can reveal a lot about how a customer views a company and its product, but for a human analyst to listen to those calls and extract that information, it takes a lot of time and effort. AI, however, can handle tens of thousands of files in just a few hours.
Companies can use this data to help create AI-powered virtual agents that can have natural conversations with customers about basic topics and react appropriately. Frantzen highlights Watson AI Assistant as his favorite tool for this purpose because it leaves the more complex interactions for human employees, so they can spend more time doing what they do best.
AI enhances employees and makes them more productive, says Frantzen. He recalls a system that Perficient built to identify potential cancer patients for clinical trials. It aids medical staff by mining thousands of data points and surfacing useful information for those professionals, helping them make informed, data-driven decisions.
AI machine-learning algorithms can quickly create predictive analytics models from structured and unstructured data. These can expose insights that previously might have taken months of analysis and reveal patterns that employees didn’t know about.
Ears to the ground
AI tools are also making it easier for companies to find new value in their existing data. After transcribing content, they can extract meaning from transcripts thanks to pretrained machine-learning models that are available as online APIs. Frantzen highlights Watson Discovery, which ships with Watson Natural Language Understanding built in, as his go-to tool.
He also cites social media as an example of how AI can help corporate leaders get ahead of negative stories or customer issues by monitoring and analyzing feeds, categorizing posts into different topics and detecting trends to be addressed. A machine learning algorithm can highlight these trends in a report that can be delivered to a PC or a smartphone, giving users the chance to put out a potential fire with a thoughtful, timely response before it becomes a problem.
You don’t have to wait weeks for a team of analysts to spot a conflagration on social media long after it’s happened.
Frantzen acknowledges that most companies are just starting out with AI. If an organization adopts cognitive technologies without taking the proper advice, it can quickly run into challenges.
- Foregoing an evolved cloud strategy. AI projects thrive on high data volumes, which require lots of computing power and storage. This is what cloud platforms are good at, thanks to their ability to scale on demand. If customers want to master AI, they must first master the cloud.
- Expecting too much without the necessary investment. AI isn’t just a turnkey solution, Frantzen points out. While basic functions such as speech processing are available as online services, the real value comes in binding those services together and layering extra AI training on top of them. This enables AI systems to become more valuable by learning about concepts relevant to a company’s specific application and subject matter. Underestimating the investment required often creates the third hurdle to an effective AI program.
- Lacking C-level support. Decision makers must drive these projects to provide not just funding but other initiatives, such as cross-departmental data sharing, that make AI more effective.
These early investments pay off later as AI applications scale up to handle high transaction volumes and provide services across multiple channels. An effectively trained virtual agent can power an omnichannel strategy that covers everything from phone systems to Facebook Messenger.
Don’t try to do everything at once, Frantzen recommends. Instead, follow a long-term strategy and take things one step at a time rather than frantically try to adopt AI across the board. Controlled experimentation is important, and the occasional failure isn’t just acceptable, it’s mandatory. AI implementation teams can learn from those failures even as they build on their successes.
Learn more about Frantzen’s take on using AI to deliver value from unstructured data in the latest episode of the IBM Cloud Podcast and sharpen your skills at Think 2019.