Unfortunately, that’s to be expected. I see it every day at our agency, which specializes in conversational AI, when chatting with prospective clients: There’s a huge gap between the expectations and the actionable, concrete promises of AI when it comes to its real-world business applications.
Conversational AI Is Hard
Successfully developing your own AI strategy and future-proofing your business involves a careful recalibration of your expectations, and it starts with looking at the realities of the market and AI’s current technological capabilities.
According to Gartner research (subscription required), 70% of customer interactions will involve emerging technology, such as chatbots or machine-learning applications, by 2022. This is up from 15% in 2018. While these stats indicate a bright future, it’s important to understand that AI is not magic. It takes time. It takes money. And most of all, it takes high-quality data -- loads of it.
To set yourself up for success, I’ve created a quick list of the most common misconceptions about AI, so you know exactly what to expect, and what not to expect, as you begin drafting your AI strategy.
Myth No. 1: AI understands everything and can talk about anything.
AI assistants or chatbots are not generalists. They’re specialists. Even Siri and Alexa have a hard time chatting about any given topic at any given time, even when millions (if not billions) of dollars have been poured into them.
It’s unreasonable to expect your AI assistant to know everything about your brand, the weather, what Kim Kardashian wore to the Met Gala, and who the Prime Minister of Malaysia is.
Reality: AI can’t freestyle language on your behalf, and if your brand is dear to your heart, you shouldn’t let it. An AI assistant is an extension of your brand and your customer support team, not a replacement. As such, it should learn from as much of your existing data as possible (i.e., product databases, CRM, FAQs, etc.) and from ongoing interactions with your customer service team.
The takeaway? A good AI assistant is not a jack of all trades, and it shouldn’t be. It’s a brand specialist and a sidekick for your customer support team.
Myth No. 2: A chatbot can be built in 10 minutes.
No matter what those sensational headlines from chatbot development companies suggest, don’t drink the Kool-Aid. Tools that promise to deliver a chatbot in 10 minutes are fine if you’re looking for a base-level starting point or prototyping tools. However, these are not sophisticated AI-powered chatbots by any means. They have no brains, and at best, they are automated workflows.
Reality: If you want something that will stand the test of time, that is intelligent and malleable enough to change with seasons, campaigns and customer needs, it takes much more time to craft a great AI assistant. Ultimately, you get what you pay for. A proper conversational AI assistant needs far more than a third of your lunch break and requires extensive training, lots of time and lots of iterations. Depending on available data on the brand side and speed of collaboration with the internal team, it can realistically take two to six months to have a solid first version of this type of tech.
Myth No. 3: It can learn by itself and figure it all out.
Machine learning opens up a world of possibilities, but it’s not magic. It’s called machine learning for a reason, and it’s only as good as the data it's fed. So if that data isn’t pristine, you might end up with a PR catastrophe on your hands (let’s not forget about Tay from Microsoft, which turned surprisingly racist in a quick 48 hours).
Reality: You must keep things in check and have a team supervising your AI training, which is ideally composed of nontechnical people and developers alike. A healthy approach to machine learning incorporates both rule-based tools as well as neural net training. Because the reality is that we’re just scratching the surface of AI, and the more tools and perspectives, the better.
Myth No. 4: An AI application is built, and then it’s done.
Launch Day might be the most exciting day for your AI assistant, but it’s actually the worst for its performance. It’s not like a mobile app or website, where the heavy lifting is more or less over once you’ve shipped it. It’s not one and done. You’ve taught your AI as much as possible, but once it starts interacting with people for the first time, it’s bound to trip a few times before it starts walking.
Reality: It takes time and data to improve AI. When an AI assistant is shipped and deployed, that’s when the project starts, and the hard work begins. It’s important to continuously monitor your conversation logs to see your chatbot's strengths and weaknesses, and flag opportunities for improvement. And this is one of the greatest benefits of a chatbot: its malleability. Nothing is set in stone, and you can always make adjustments on the fly based on your customers' interactions, preferences, seasonality, marketing campaigns and more.
Three Steps To A Viable Path To Conversational AI
So, now that we’ve cleared the air and you know what really goes on with building and launching a conversational assistant, how do you get started?
• Open up a 1:1 communication channel (i.e., live chat, Messenger, etc.).
• Gather and analyze data (from customer conversations).
• Continuously iterate and improve your AI assistant.
In the end, the idea is not to simply survive the AI disruption, but to thrive and embrace it, and nurture a sustainable competitive advantage. If Gartner’s previously mentioned forecast is right, you have less than three years to get your AI strategy right and future-proof your company. Better start now.
