How to Train Your Robot: 4 Ways Startups Can Apply AI
Despite nationwide venture funding hitting a multiyear low, venture capital deployed to artificial intelligence startups has reached a record high.
Source: Vince Lynch
Last year, VCs struck 658 deals with AI companies, nearly five times the number that signed on the dotted line four years before. To date, the market contains 2,045 AI startups and more than 17,000 market followers, with more joining by the day.
AI’s rapid rise has swept up startups and enterprises alike, including U.S. automaker Ford, which recently bought AI startup Argo for $1 billion. The acquisition cements experts’ suspicions of Ford’s coming foray into self-driving technology. Other startups -- so many, in fact, that entrepreneurs need a “best of” guide -- are betting heavily on bot platforms.
So while we’ve just glimpsed the tip of this innovation iceberg, it’s clear AI is no longer some nebulous technology of the future. Sixty-eight percent of marketing executives, according to media research firm Weber Shandwick, report using AI in their operations. For a technology that only went mainstream in 2016 and barely existed four years ago, that’s a remarkable adoption rate. How, regardless of the platform you choose, can you join forward-thinking entrepreneurs and build your business with AI? So far, four ways stand out:
1. Get to know your next customer.
A politician wouldn’t dream of delivering a small-town stump speech to her urban constituents. Why? Because you’ve got to know your audience. The same is true for entrepreneurs. Before you broadcast your message, you need to know who you’re trying to reach.
Node, an account-based intelligence startup, uses natural language processing -- a fancy term for teaching a computer to understand how we humans speak and write -- to develop customer profiles. Node is crunching vast swaths of data to connect the dots between marketers and the companies they’re trying to reach.
Once you have ample customer data -- Node uses data crawlers to scrape information from social media, news sites and more -- pair machine learning and natural language processing models to extract sentiments from unstructured data. Then, just as senators segment constituents into demographic groups, Node uses cluster analysis to sort clients’ customers into like cohorts.
2. See how people truly use your product.
If, heaven forbid, you forgot to tag your neighbor at last week’s house party, Facebook was no doubt there to remind you of your error. How does Facebook know which of your friends you left untagged? It has gone all-in on an AI technique called convolutional neural networks.
Convolutional neural networks, which loosely model how the brain’s visual cortex interacts with the eyes, work by separating an image into tiny portions before running each of those specks through a multilayered filter. It then “sees” where each speck overlaps with other parts of the image, and through automated iterations, it puts together a full image.
Many different ways exist to apply this technology, but retail businesses can start with image classification. Try using a convolutional neural network to break down photos of your products posted online. The model can identify customer segments that frequently use your product, where they’re using it and whether they commonly pair other products with yours. Essentially, this automated image analysis can show you how your products fit into customers’ lives, allowing you to tailor your marketing materials to fit.
3. Get inside the user’s head.
Success on social media requires careful listening and quick action. When a social campaign isn’t working, it’s best to put it out of its misery quickly. On the other hand, when one strikes a chord with customers, doubling down pays dividends.
But to do so, you need real-time insights about customers’ reactions to your content. Fortunately, AI can take the emotional temperature of thousands of customers at once. Dumbstruck, a video-testing and analytics startup that I advise, has added natural language processing to its emotional analytics stack. This allows it to provide moment-by-moment insights into viewers’ reactions to media. Dumbstruck’s model grows stronger with each reaction analyzed, producing a program that perceives human emotions even better than some people can.
4. Provide affordable, always-on support.
Customer service is -- or should be, according to consumers -- the department that never sleeps. More than half of people, 50.6 percent to be precise, believe a business should be available 24/7 to answer their every question and concern. When asked whether businesses should be available via a messaging app, the “yes” votes jump to nearly two in three.
Fortunately, bots don’t sleep, eat or go off-script. A well-built bot can offer cost-effective, constant customer service. Of course, grooming your bot to serve customers requires front-end data -- ideally hundreds of thousands of example conversations -- but you can get started with a human-chatbot hybrid. With this approach, the bot answers run-of-the-mill questions, while a human takes over for the more complex ones. Then, as the data builds and the model matures, you can phase in full automation.
Taco Bell, for example, knows taco cravings come at all times of night. Its TacoBot takes customer orders via Slack chat. Ticketmaster, a popular concert ticket provider, also offers an automated chat widget that informs users about new events and provides order details.
Artificial intelligence is not just the stuff of Netflix shows and sci-fi paperbacks. Today, it’s as real as the customers it serves, the marketers it informs and the companies it scales. The intelligence may be artificial, but the results are absolutely not.