In 2017, it’s time for artificial intelligence (AI) to be part of your company’s strategy. More than 10 million Amazon Echo devices have been sold in the U.S. alone, and 1 billion people use Facebook Messenger each month. Consumers are slowly starting to embrace the tens of thousands of third-party skills, chatbots, and mobile AI interfaces.

Brands have the opportunity to add their services to the growing list of popular platforms. However, many underestimate the complex journey that awaits them when developing AI — especially considering how unique each brand’s bots, audience, and optimal platforms can be.

It’s important to determine a smart approach and figure out how to develop skills for bots that translate into something useful for consumers. Each journey is different. As you consider building a product there are a few things to keep in mind along the way:

1. Choose the right platform for your assistant

Deciding which platform will work best is the first decision when building any product. While every interaction essentially comes down to a natural language conversation, different platform capabilities and usage scenarios dictate which platform makes the most sense for your brand.

There are many options out there today, and you may wonder if you need to be on every platform.

Ask yourself these questions as you consider where to build your assistant:

  •  Audience: What demographics and geographies apply to the intended users?
  •  Context: Is the product intended for individuals or multiple user conversations?
  •  Capabilities: What does the product need to do?
  •  Type of task: How much input is needed to successfully complete a task, how long will that take, and can it be accomplished hands-free?

If you are working on a productivity tool, for example, you may pursue Slack. For hands-free tasks, perhaps use Alexa or another voice platform.

2. Screens and speakers are two different things

Before building, it’s important to understand the nuances between voice and screen assistants. There are distinct requirements for each, and it’s important to put the right pieces into play.

Voice assistants try to mimic an actual human that provides support to a user in a hands-free scenario. They must be truly conversational. Screen-based assistants, on the other hand, can also be built as a set of menus. Voice generally works best for quick interactions, such as tracking a flight. Information that’s spoken aloud can be overheard, and every additional piece of information can quickly add too much cognitive load on voice interactions with a bot.

Conversely, screens allow for rich information and media to be put in front of a user. Because of this, complex tasks — such as booking a trip with multiple flights — work well on a screen-based bot.

Screens can also be used to augment voice assistants. However, it’s important that every voice assistant task is solvable with voice alone. Otherwise, it interrupts the hand-free scenario, which, in general, is the major reason for using voice.

3. Don’t be afraid of natural language processing

Incorporating natural language processing (NLP) into your assistant depends on its use case. Not every chatbot needs to understand NLP and many could offer something as simple as menus and webviews a user can tap. Not typing is sometimes even advantageous. For example, we saw strong upticks in usage of Kayak’s Messenger bot when we allowed people to answer questions with simple touch replies.

However, if you do want to incorporate NLP, there are cloud platforms available — like or Amazon Lex — that are solely driven by menus, buttons, or webviews. You can also build your own solution. Building an in-house solution can provide strong benefits, like allowing for different languages, tighter integration, and domain expertise.

Depending on the industry and product, building up NLP capabilities can be a strong asset, but only if you’re trying to achieve state-of-the-art performance.

In the end, be useful

Ultimately, AI agents need to prove that they help users get things done easier. Not every task matches this criteria. But to be useful, you have to help the user navigate and be successful in using your assistant.

Remember, we are still in the early days of AI, so many users aren’t aware of how to use these technologies, and they tend to quickly lose interest. You can’t put a web page in a bot, for example, so you want to have a real product that solves problems like a website would solve them. It’s important to test the product prior to launch, invest in solid on-boarding, and quickly determine where users first experience issues.

As AI becomes ingrained in daily life and continues transforming verticals, brands are understandably feeling the pressure to jump on the bandwagon. While it is wise to adapt early to keep ahead of the curve, there is no one-size-fits-all model for AI integration. The brands that take the time to determine the smartest approach — and invest in the right skills and platforms — will be rewarded for providing real value to consumers.

Matthias Keller is Chief Scientist at Kayak and is a key driver of the company’s AI efforts.

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.