Check out the on-demand sessions from the Low-Code/No-Code Summit to learn how to successfully innovate and achieve efficiency by upskilling and scaling citizen developers. Watch now.

Imagine the perfect personal assistant.

This partner would understand your needs — often before you’ve even expressed them — and know exactly how to deliver what you’re asking for. They would make helpful suggestions without becoming intrusive, and keep you from missing appointments and opportunities. Most importantly, this personal assistant would be someone you can trust implicitly.

Now, how do you embody those traits in an artificial intelligence-powered service? Our experience creating our travel assistant app, Mezi, illustrates key principles of AI regarding the ongoing role of human involvement and how to draw the dividing line between valued assistance and unwelcome intrusion. Here’s what we’ve learned recently.

Gaining trust

Our customers typically spend several hundred dollars with us on each transaction. To earn that kind of business, it’s essential for people to feel comfortable — to gain a sense of trust and empathy from the very beginning of our relationship. In that regard, trust is a prerequisite for everything to follow, and building trust must be the first goal for your AI-powered service.


Intelligent Security Summit

Learn the critical role of AI & ML in cybersecurity and industry specific case studies on December 8. Register for your free pass today.

Register Now

At that initial stage — or at any stage, really — the question customers are asking themselves isn’t “Is this a perfect AI implementation?” Instead, your customers are asking themselves, “Will this service always understand my requests and respond with meaningful recommendations?” That’s a crucial point to understand. Many AI services fail through a well-intentioned but misguided quest for purity rather than a focus on effectiveness. Remember, the goal isn’t for your AI to be perfect, it’s for your AI to be part of a perfect service.

As you’re building your AI technology, the goal should be that users will be completely unaware that they’re dealing with a human-assisted AI bot. It’s hard even for us to tell the difference between human and AI interactions in our service. At this point, some of our flight and hotel interactions are powered by AI and some require human interaction. The goal is for our AI machines to constantly learn from the human interaction.

Adding proactive to reactive

So far, we’ve been talking about reactive AI: responding to explicit user requests. That’s important, but it’s not the whole story. How many consumers have firsthand experience with the perfect personal assistant we described earlier and know the full range of possibilities available? Both to surface the full value proposition of your service, and to deepen that sense of trust, it’s important for your AI to anticipate a customer’s needs and offer useful help without having to be asked. Otherwise, you risk being seen — and used — as nothing more than a glorified search engine.

If you want to avoid falling into the search engine trap, you need to try to stay with the customer throughout their purchase journey. The sale or transaction should be the beginning of the interaction, not the end. If customers are using your AI to order groceries, be able to anticipate that they are buying the ingredients for a salad so you can ask them what type of salad dressing they would like to order. Or remind them they forgot to order the cucumbers they ordered last time. In other words, your AI needs to do more than just take an order. It needs to be smart enough that it adds to the customer experience as opposed to just completing a transaction.

The deepest trust comes when people know that their assistant always has their back when things get hectic. That’s what AI companies must strive to provide.

Knowing where to draw the ‘human’ line

With all the work you’re doing to create a humanlike persona for your AI, the last thing you want to do is break the spell with a lapse into botlike behavior — especially in the form of intrusive marketing. You might want to build your AI to offer tips and suggestions, but I would advise that you do so carefully and judiciously, with a specific intention, as a good personal assistant would. Your suggestion should never be “Buy NOW and save!” Instead, your suggestions should focus on adding to the experience, such as “Do you have plans this weekend that you need to shop for?” or “Did you find a gift for your son’s birthday, or want me to share a few more suggestions?” As a rule, any message that comes from your AI should be something that could plausibly have come from a human assistant — or a trusted friend.

As you design your service, always remember that what matters most is what happens on the front end — with the customer’s experience — not on the back end. As proud as you’ll be of your beautifully designed and well-trained AI, all your customers will care about is whether their personal assistant is coming through for them when they need it.

Snehal Shinde is the CTO and cofounder at Mezi, the travel concierge service.

Above: The Machine Intelligence Landscape. This article is part of our Artificial Intelligence series. You can download a high-resolution version of the landscape featuring 288 companies by clicking the image.

VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.