LAST YEAR, SWEDBANK in Sweden “hired” Nina, a dynamo of a company rep. Nina is adept at interacting with customers on the bank’s website, answering their questions and routing them to the right human representative. She typically engages in some 45,000 conversations a month and successfully answers eight out of 10 inquiries.
But aside from her stellar performance, there’s something else notable about Nina: She’s a bot.
Not only can Nina understand and mimic human speech, she can also use data to make connections and recommendations. “Nina’s brain for Swedbank is educated on all things Swedbank,” says Robert Weideman, EVP and GM of the enterprise division at Nuance Communications, which created Nina. “You can search for the closest branch, you can ask about wire transfers. Instead of clicking through the website, you just ask a question.”
This AI implementation isn’t just a neat trick. Weideman notes that 65 percent of phone calls to a contact center come after a consumer has tried to get the information on a website or through a mobile app. “So if you can answer the question on a mobile app, the customer’s happy and they get their question answered,” Weideman says. Best of all, Nina is available 24/7.
Such is the promise of AI for business. In addition to providing tools to empower human workers to be more efficient, AI can also help save customers time and leave them with a better feeling about the brand. “AI will become the new digital spokesperson for leading companies,” says Nicola Morini, managing director of artificial intelligence at Accenture. “Customers will experience a company’s brand through personalized, 100 percent consistent and natural interactions with AI service agents—and even engage with the brand through other companies’ AI interfaces.”
Benefits on the Back End
Front-facing interfaces are just one aspect of corporate AI. Another equally important application is on the back end. Consumer-facing bots like Nina transcribe their conversations into text, which an AI-based system can mine for data. “They can do it on a scale no individual can,” says Mike Gualtieri, an analyst with Forrester, referring to AI-based systems. Not many are using such systems; the technology is so new that few companies have yet employed it. “Part of it is because analyzing unstructured data like text is just messy and hard to do,” he says. Deep learning is catching up, but companies still need a large number of such transcriptions to find significant patterns.
When companies analyze these troves of data, the benefits are obvious. InterActiveTel in Houston, for instance, works with car dealerships to record conversations with customers and then transcribe that speech to text. Using open-sourced APIs, the company is now converting those conversations to text in real time and analyzing it live to alert managers about selling opportunities.
The consumer-facing aspect of AI is just one component. Robotic process automation (RPA) uses software-based bots to perform repetitive, routine business functions. Using RPA, one insurer was able to handle 500 premium advice notes in 30 minutes. The process used to take two days. Such efficiencies can be multiplied across an organization.
Bumps on the Road to AI
Despite such clear benefits, a Forrester study showed that just 12% of enterprises say they are using AI. While some of that figure is based on semantics (machine learning is widespread but often not known to be a component of AI), most companies are still trying to figure out how AI can be applied to their businesses.
Generally there’s no single person within an organization who spearheads an AI push. Instead, a combination of business and IT executives tend to be the AI advocates. “It’s usually driven by the C-suite asking, ‘What are we doing about AI?’” says Gualtieri. “But it also comes from data science teams, and on the IT side it’s coming from big data.”
Implementation isn’t as easy as signing up for a SaaS product, however. Though there are open source tools on the market that incorporate AI capabilities, deep learning benefits from very specialized, expensive hardware and expertise. Setting up effective models is also time-consuming. “A text analytics job using traditional methods could take five minutes to process,” Gaultieri says. “That same job with deep learning could take nine hours. Deep learning takes longer to train a model, but it is worth it because the model can be more accurate.”
Given the commitment involved in creating deep learning, having an on-site data center might be more cost-effective in the long run than using the cloud. Given the potential benefits of AI, though, investing in that hardware may just be worth it.
This story was produced by the WIRED Brand Lab for Accenture. For more information, check out Accenture’s Tech Vision 2017.