This year’s AI Business Week Digital Symposium may be over, but there’s still so much to discuss from the sessions. I’ve been thinking back on the retail and e-commerce session I joined to talk about the progress the industry is making in AI.
You can watch the session on demand now, to hear the full panel with Chuck Martin, Editor of Strategic Content for AI Business, Eden Zoller, Chief Analyst for Digital Consumer Service at Omdia, and Errol Koolmeister, Product Area Lead Engineer for H&M’s AI Foundation.
Last time, I talked about how retailers can help build trust with their customers, encouraging them to use AI-powered self-service. Now, I’d like to pull out one of the excellent audience submissions from the Q&A session. It highlighted one of the most important considerations for retailers when they prepare to launch a new AI tool: when’s the right time to hit the button and deploy it to your customers?
Don’t deploy too early…
There are a lot of brands out there that are deploying poorly executed AI tools that don’t really answer customer’s needs or end up adding friction to the experience—the opposite of what they want to achieve. And often, all they needed was a few more weeks of development and testing.
There’s a real risk to deploying your AI too early. If it’s not ready to face the public, there’s a laundry list of things that could go wrong. A customer service tool that can’t handle complex queries could put extra pressure on your human agents instead of relieving it. A new AI-driven feature—like visual search, for example—might struggle with demand as eager customers try it out. And, at its most basic, the tool might just break.
Eden cautioned patience, especially for customer-facing AI tools. One bad experience with AI is often enough to put a customer off using it for life. It doesn’t matter if you fix the issues or add new features; once they’ve decided they don’t want to use it, you’ll have an incredibly hard time tempting them back. And if your AI isn’t used, it’s a wasted investment.
…but don’t wait for a perfect product
However, if you’re looking for a completed AI project that’s ready for use and you never need to touch again, you’re going to have a long, long wait.
I’m fond of the phrase “no chatbot survives contact with the customer”. Regardless of how much work you put into your AI before it goes live, your customers are always going to find issues you couldn’t predict (or even replicate) in development. Errol and Eden both talked about the importance of live testing with real customers—you don’t know what you don’t know until someone breaks your tool in an unexpected way. That’s when you get the real opportunities to finetune your AI to your customers’ real needs, not just what you expect them to want.
In the end, you need to be pragmatic about when you choose to deploy to your customers. You need to be able to guarantee a level of functionality that makes it worthwhile for customers to use the tool—but it’s never going to be perfect right out of the gate.
AI needs to be an agile process of design and redesign, and that’s what a lot of organizations fail to embrace. In the race to become as proactive as possible, leaders forget that this needs to be a reactive process that responds to customers’ needs.
Catch up with the full session
There were lots of other great questions from the audience—and Eden and Errol’s presentations both gave a really interesting overview of the state of AI in retail, how it’s expanding through the industry, and why it’s important for brands to build trust with their customers.