The conversation around voice and AI seems to have moved away from “if” brands should be investing in these technologies and towards the “how” of actually implementing them.

In Canada, AI-powered voice is a bit under-developed compared to the U.S., thanks largely to the fact that Google Home and Amazon Echo only recently launched here. But given how we’ve seen the voice assistant space grow, the time to play wait-and-see is long gone.

“Voice is the most competitive field ever, because it completely levels the field among every brand active there,” said Pat Higbie, CEO of developer XAPPmedia. “This space is moving faster than you think, so it’s important to get in the race now because this isn’t a market where you can build it and they will come. You need to drive users to your voice experiences. With web search, you get a page full of results. With voice, there is only one, so if you’re any lower than that, you’re invisible.”

Illustrating how important discoverability is, Higbie played a clip of a user asking to find an SUV, which brought up a vehicle finder voice app from Toyota. While the app brought up questions that hit on Toyota’s brand values (one prompt asked the user if they wanted something that was “rugged” or “elegant”), the user would have never experienced them if it had landed on an app from another automaker.

“Search is important, and users are asking a lot of questions we haven’t even thought of,” Higbie said. “That’s why it’s important to measure and reassess and adjust constantly.”

Higbie said voice needs to be integrated with the rest of a brand’s marketing and communications strategy for it to be effective, with Doug Rozen, chief media officer of 360i, saying they should be promoted like social handles, call numbers or website URLs if people are actually going to use them.

For marketers who can’t shake images of AI systems going rogue and potentially doing something that negatively impacts a brand, they don’t need to be experts to think proactively. But they need to have a basic knowledge of how it works so they can clearly communicate with agencies and developers the needs and eventual use cases of the AI system. That way it can be trained and tested properly.

“We get clients that have trust concerns,” said Rob Wilk, head of North American search sales for Microsoft. “So how do they know if it’s real, fake, bad or if it’s going to work. And we just say to test it, the way you would anything else. Due diligence works and not doing it doesn’t.”

“One of the concepts of machine-learning is that you can’t assume it’s going to work,” added Evan Simeone, SVP of product management at PubMatic. “You take some of your data and set it aside as a test case. It’s a shift from an engineering perspective to a scientific and experimental process. You gotta try it out, and use your data and problem set. You need to think about the data you’re using from the beginning.”

Wilk brought up the example of an automatic soap dispenser that was developed by a team of white engineers and would not work when non-white hands passed in front of the sensor. He said that showed how building something on incomplete data and not thinking through all the use cases can lead to a system that doesn’t work, and can actually be harmful for a product or company.