Commentary

How to Ensure AI Doesn't Make Your Customers Hate You

“South Park” viewers got a taste of AI gone bad when characters on the show deliberately messed with viewers’ Amazon Echo devices, adding weird items to viewers’ grocery lists, changing their clocks and getting devices to repeat obscenities.  

While your marketing plan probably doesn’t aim to create chaos for consumers, you must be cautious about how your AI-powered outreach actually operates in the wild.

AI is a big category of technology that may or may not help you address your marketing needs, and it’s not all chatbots and games. AI can improve audience segmentation and target ad messages, (including new offerings from Nielsen,) create product-pricing models and direct customers through your website. It can even write content (like Curata and WordAI.)

Some companies have had positive momentum with very thoughtful applications of AI, but lots of AI is still pretty dumb. Whatever the case, your experiments with AI — which are quite necessary if you want to keep up with other marketers, since 86% plan to use it soon, according to Retail Dive — need to be examined to make sure they don’t hurt your relationship with your customers as you rush to embrace this potentially powerful tool.

advertisement

advertisement

Evaluate the (Real) Customer Experience

It’s enticing to imagine how cool it would be to have your customers order something with a chatbot like Domino’s or to deliver a seamless mobile shopping experience. But what you imagine may not align with what you can do. Be realistic about what resources your company can dedicate to making the project successful. Your organization may have limited data to feed the AI algorithm, or may be unable to develop or integrate around the technology. Your pre-launch tests need to account for these pitfalls.

After understanding how much work it takes on your end to load whatever AI tool you’re planning to use with data and insights, you have to examine the experience your customers will have as a result of a self-directed algorithm making decisions for you. Companies like Tesla can use their customers as guinea pigs; Tesla enthusiasts will tolerate a lot of bugs. But brands with less enthusiastic customers will suffer the consequences when algorithms make bad choices like racially discriminating against consumers.
 .
Ask your vendor for a demo that doesn’t just show the dashboard. Try to determine what the consumer actually experiences. You need to walk through all of the steps just as a customer would.  If customers have to do a lot more work than they would with a real customer service representative, your experiment might not be ready for prime time. Consumers prefer a human touch, according to a study from GetApp.  

Anticipate the Shortfall

Chatbots run out of answers and require customer service calls. AI writes content, but tends to miss important nuances that might force readers to have to search for more information. Beautifully targeted ads like those that Cosabella created, with assets pulled together by an algorithm, might have nothing to do with your online website, creating a jarring experience.

The more seamless the transition between AI-driven interactions and other interactions, the better. 1-800-Flowers has integrated several AI-driven ordering functions into its website and social media presence seamlessly, but it took a lot of planning to make sure the new experience was convenient enough to replace dialing “1-800-Flowers.”

Plan for Failure

To ensure your shoppers aren’t accidentally being racially profiled or stalked by bad targeting, talk to your AI partner to make sure you can check in frequently on actual customer experiences, beyond receiving general reports. And when you do look at reports, consider the data in sections. Don’t just look at the average. An 80% completion rate means that 20% abandoned the experience. Determine what led people down the wrong path, and have a plan for how you’ll react if they post an embarrassing photo of their experience on Instagram.

Next story loading loading..