Nine in ten floor questions are "Do we have it, and where is it?"

One of my first AI projects was a voice assistant for team members on the shop floor of a national discount retailer. We built it thinking people would use it to ask how to do their jobs well, things like the refund policy, how to run a specific process, or even information about their pay or what time they could go on a break.

We were very surprised when we saw that nine in ten questions were some version of "Do we have it, and where is it?"

Before the assistant, the answer to that question was often "We don't have it", even when the product was just one aisle over, and the customer would leave without it. Everything spent bringing them into the store leaves with them. Now think about how many times that question gets asked across all of your stores on a single Saturday.

The plan versus what happened

The problem that we were trying to solve was supporting team members who had just started in the business to be able to do their jobs as well as possible and receive real-time coaching, even when managers were not around. We had a mantra: How can we support someone who is on their third shift to work as well as someone that has been with us for three years?

This is still a meaningful problem to solve in any retailer, but once we deployed the tool, we found that there was a more pressing need than task support.

Product location was on the list early, as most junior team members don't yet know the location of products, especially in a big-box store. But it was a hard one to build: disconnected data sources, such as product location and stock levels, and even store maps, made us pick a safer implementation. Instead of helping people navigate to a location, we decided to provide simple straight answers. "Do we have it in stock and which aisle is the product in?" We thought that experienced team members on the shop floor all day wouldn't be turning to an AI assistant to figure out where the products were. We were very wrong. From junior casual team members on their first shift to veterans with more than 10 years in the store, product location became the favourite feature in the app straight away.

But what really sold it was customer service. Think about how it plays out on a busy Saturday. A customer walks up to a team member and asks for a product. Even if that team member knows exactly where it lives in the store, they still need to check whether it's in stock, because the shelf might be empty and the rest of it could be sitting somewhere in the stockroom. It doesn't matter if you've been in stores for 30 years, nobody can keep the location and availability of every product in their head. So the team member has two options: go and check, which can take minutes while the customer waits, or give the safe answer, which is usually "We don't have it."

The answer it replaced

When people talk about AI in stores, the conversation usually goes straight to jobs, and to the idea that the technology will replace the people on the floor. In our case, nobody's role was replaced by the assistant. The team members on the floor kept doing the same job, and what changed was the answer they could give when a customer asked for something.

"We don't have it" doesn't come from people being lazy. If you don't know where a product is, you can lose critical minutes looking for it, and once you go into the stockroom it gets even harder. Some of the internal systems are also not as friendly as head office might think, and checking stock in another store can take several clicks across several screens. So "We don't have it" becomes the polite way to end the conversation. With the assistant in their ear, the same team member could give the customer a straight answer in a couple of seconds and walk them to the aisle. That is what we meant by augmenting team members: a casual on their third shift could answer like someone who had been with us for three years, and our veterans had one less thing to carry in their heads.

Why it had to be voice

At the time we started to build the team members' system, the prevailing use of the technology was via chat, or basically typing and getting your answers back on the screen. Early on, when testing the first few prototypes with team members, we noticed that they were very adamant about not texting in front of customers. They didn't want to provide that experience to customers and preferred a voice interface, where they could press a button on the earpiece and ask a question (exactly the same behaviour that they have with their push-to-talk radios).

Voice costs more than a chatbot to build and, above all, to run, but we pushed on. When designing for complex change like this, especially considering the introduction of brand new technology, it's always important to start with what people actually want to use. In this case, voice was showing a high degree of use even in early tests. We decided to double down on this bet: people wanted to use it, so we built for that first and worked out the tech and the cost later.

What this means for AI on a busy floor

I often hear that AI doesn't belong in a fast-paced customer service job like retail. I compare it with the arrival of mobile devices (or portable data terminals, PDTs) on the shop floor.

In plenty of stores, a team member still greets you and walks back to a central computer to look something up. Larger retailers put devices in people's hands so they could answer on the spot. Years ago the worry on the ground was that devices would make service worse. They didn't. But as the stockroom and the extra screens show, the device alone doesn't get the answer, so many customer questions still go unanswered. AI is the same step, with more capability. Getting through several systems to a complex answer used to need fast hands and the right screen. A well-built assistant does it in a couple of seconds.

The next question I usually get is about location data, and my recommendation is to start where you are. The first step is to bring your data to the surface and put it in your team's hands, and then fix it from there. But I also have to be honest about what happens when you do that. If your stock and location data isn't reliable, AI will only expose it more, because it can reach that data in seconds and bring the errors to the surface every time someone asks a question. Then the next step is a serious look at your data: whether your systems know where products actually live, whether they're named and grouped the same way in every store, and whether the stock number on the screen is what's really on the shelf. The better the data, the better the outcome, and if you get it right, the experience can be close to flawless.

So when you look at AI for your stores, judge it by what happens in front of the customer. For us, the value showed up the moment a team member stopped guessing and started answering, and the customer who would have heard "We don't have it" was walked to the right aisle and left with the product. That is where AI becomes a no-brainer, so start with the outcome you want for that customer and put AI to work on it.