In an interview that took place this week during VentureBeat’s Transform 2020 summit, former Salesforce chief scientist Richard Socher spoke about AI and machine learning uptake and deployment in the enterprise. It’s his assertion that text and natural language are in many ways the ideal modalities for the market, because, as he notes, every company talks to its customers in some form or another in natural language — whether that’s voice or written text.
“To me, AI captures one of the things that makes us the most unique species on Earth — intelligence — and language is the most interesting manifestation of intelligence. It connects all the other areas of AI,” Socher said. “Companies realize that if they can answer 10% of requests about mundane things like password recovery using AI, that’s a huge savings for them.”
Socher gave reply recommendations as a concrete example of natural language solving a business use case. Google’s and Microsoft’s email products serve up AI-generated reply suggestions, as does Salesforce’s Sales Cloud service. Within the agent console, a feature called Einstein Reply Recommendations taps machine learning to proffer messages most likely to elicit a response.
When it comes to marketing, Socher says he’s seeing AI-driven solutions like opportunity scoring and attribution gain traction. Opportunity scoring identifies sales prospects most likely to be won, while attribution lets sales reps know which marketing efforts are yielding the best results. “Companies don’t want to spend time calling or emailing folks who don’t want to buy their products,” Socher said.
Chatbots are another technology whose adoption is on the rise among enterprises, according to Socher. Gartner agrees — it predicts chatbots will power 85% of customer service interactions by the year 2020. While the earliest chatbots struggled to field natural language questions, sophisticated variations like Amazon’s AI service agents can lighten the burden on human teams.
“A lot of companies aren’t able to keep up with the influx of traffic … and chatbots are a really great way to scale,” Socher said. “In the beginning, chatbots were mostly seen as a text interface, but in the future, there will be phone, web, and email chatbots to which you can deploy answers to questions … Today, we see a wide variety of [implementations] across our customer base, with some folks who are just getting started integrating chat capabilities in their websites in the first place.”
Some firms are at an advantage where AI deployment is concerned, Socher says, because their workflows generate data and thus eliminate the need to manually create it. For example, if a company uses a customer relationship management platform from which its salespeople answer chat requests, this workflow can be repurposed to collect training data that can teach a system how to provide answers automatically.
“If you don’t have these workflows at all, it’s a huge effort,” Socher said. “If you want to try to understand something like sentiment in a marketing campaign on social media, you need to ideally label all the posts that mention your company or products so that AI can inference how people are responding to the campaign.”
To ensure their AI and machine learning-powered product solutions launch without a hitch, Socher believes, companies need to keep in mind three core operational and managerial considerations. The first is deciding whether an AI project can be outsourced or developed in house. Equally important is determining how the AI might impact customers — in other words, avoiding harmful bias and accuracy issues. As for the third, it’s about establishing a process where if the AI makes a mistake, it can be flagged and escalated to a person.
“For a lot of AI applications, 80% of the work isn’t AI-related. It’s engineering work, changing your workflows, and things like that. It’s also seeing how many of the processes can you transfer from one use case to another,” Socher said. “Maybe there are different regulations in another country that you have to adhere to, for instance.”