“Enterprise users don’t know the difference between algorithms and logarithms,“ jokes Marco Casalaina, VP of Product Management at Salesforce Einstein, during an enterprise AI event we recently hosted for executives and entrepreneurs in San Francisco.
Sadly, he’s right. Despite non-stop media hype about “artificial intelligence” and “deep learning”, very few people even know what those terms mean, much less how the technologies work. Casalaina’s had stints as a product and technology leader at the biggest names in enterprise – Oracle, SAP, now Salesforce – and years of experience designing and selling complex business solutions. His biggest pet peeve? When people don’t clearly convey product benefits and resort to meaningless marketing speak, like “actionable insights” or “predictive analytics”.
“Entrepreneurs talk about ‘democratizing data science’, but then in a pitch the first words out of their mouth are ‘we have this new auto-feature-engineering tool’,” complains Casalaina. “There are 20,000 people in the world who understand what feature engineering is. Once you say that, you’ve lost your audience.”
Casalaina’s keynote was followed by a panel of enterprise A.I. experts, ranging from entrepreneurs and investors, to designers, marketers, and engineers. I asked them a few tough questions to spark some lively debate.
[cp_modal display=”inline” id=”cp_id_fdd73″][/cp_modal]
Where is the highest potential ROI for AI in enterprise?
Brand engagement, says Eva Steele-Saccio, Writer and Conversational Designer at PullString. She shares an incredible statistic: only 1% of people click through in email ad campaigns, but she sees 90% click through rates on PullString’s chatbot messages on Facebook.
Nikhil Balaraman, Director of Product Marketing at Infer, stands by his statement that “lead scoring is maybe the highest ROI and the ROI is measurable,” directly disagreeing with Casalaina’s view that the space is overcrowded. Ayush Agarwal, Head of Enterprise Products at Facebook, agrees on the potential of ROI in lead scoring, but emphasizes that we need to look “at the whole business process and build a clean system,” rather than just apply AI to a narrow task.
What’s your biggest challenge in selling AI to customers that don’t understand AI?
Steele-Saccio emphasizes that many companies assume chatbots can chat about anything, but creating a truly great experience entails focusing on the right use cases for the brand and their users. “Keeping the focus tight can be a challenge,” she warns.
Nathan Ross, Co-Founder and COO of Radbots, says he’s been most successful by giving customers a small piece of AI they can use immediately, then demonstrate how to scale the value. According to Ross, “if we can show them a way to save 5% of their time for one task, that opens the floodgates.”
How do you approach building AI?
Agarwal suggests that prior to writing code for chatbots, you should act out the interactions and conversations with other humans. “We get a person to pretend to be the bot, so we can test, iterate, and evolve instantly – all without damaging our customer relations during concept development.”
John Forrester, CMO of Inbenta, agrees. “You can’t go too fast on the technical side. You need to make sure that the systems, the data and the integrations are ready. The groundwork really determines the success of the pilot.”
Do you have any funny stories about customers misusing bots?
Turns out bots that have avatars inspire far more…interesting conversations. Forrester reveals that “if you have an avatar, 30% of the chats aren’t about the product or company and are not politically correct at all.” Steele-Saccio adds that “you really have to prepare your chatbot to respond appropriately to profanity.”
Customers don’t always abuse bots. They sometimes fall in love with them. Ross shares how one customer’s chatbot was based unofficially on the Baby Groot character on Guardians of the Galaxy. Even though the chatbot only said “I am Groot”, people kept engaging and built a relationship with it, just like they would a talking stuffed animal.
What are the most common mistakes in enterprise AI?
“It’s critical to educate your customer,” Agarwal. “Especially when they have the misconceptions that your AI will send revenues through the roof. Customer retention will be a problem if you set expectations too high, but I see this happen a lot. Lower expectations and give them concrete value.”
Forrester adds that companies are often ingrained in Silicon Valley culture without realizing they’re in a bubble. “We must understand Middle America and different countries and design experiences for them.”
In five years, what will be possible with AI that isn’t possible now?
“We all hate the amount of email we get. Booking a simple meeting can take five messages,” Agarwal points out. “My prediction is: in five years, we will have cracked email, calendaring, and team communications.”
Sara Ahmadian, CEO of Seamless Planet, believes chatbots will be ubiquitous. Ross agrees and is actively building a product called Network which connects chatbots to all your IoT, smart car, and computing devices. The vision is to enable intelligent agents to track and manage all aspects of your life.
What shouldn’t be delegated to AI?
Ross believes everything should be delegated to AI, so we can focus on the next step of human evolution. Agarwal isn’t so sure, suggesting he might not want a judge in a legal system to be an AI robot.
Balaraman points out that previous software used for sentencing turned out to be incredibly biased, largely because the machine learning was based on historical human judgements which were flawed.
Casalaina concludes: “For as much as we want to automate everything, we can’t. We still need people.”
Want more details on the challenges of enterprise A.I.? Watch the full video of the panel discussion below: