Answer Engines Aren't Eating Retail. They're Doing Retail's Job.
So when AI also comes for every grid, form, and dashboard in retail software, it’s time to think more fundamentally about the problems you solve and not the things you sell.
Source: Adobe Stock. AI generated. When AI helps your customers better than you do.
I struggled to figure out the order to tackle things this week. Do I talk about retail’s UI challenges, and then more generically about how AI is changing the way we think about UI/UX? Or the other way around?
I settled on the other way around: generic UI/UX evolution (revolution?), followed by what it means for retail. It leaves the really punchy part – that retailers too easily lose the thread on what all that UI is for anyway – to the second half.
Ultimately, it’s the more important point. For all I know we’ll all have chips implanted in our brains and stop talking out loud all together in the next 20 years. So AI’s impact on UX is, as I note below, really the prelude to the opening act, not anything truly definitive about what’s next.
But for retailers to get on that learning curve, they must move beyond thinking of themselves as companies that sell stuff. I use that phrase “sell stuff” on purpose to make it the most generic, practically derogatory summation of retail that I can come up with. This is the least of what retail could do.
When you don’t have to bend to the constraints of what the technology can do, when technology can bend to match what you need, then you need to stop thinking about why this is a dropdown instead of a radio button. You need to start thinking about the problems you solve – where the purchase is a sign that you’re successfully solving problems and not the outcome itself.
Yeah AI is killing the report and the dashboard and the grid. But that’s the least important part of what it’s doing when it destroys UX as we know it.
Let’s get into it!
Retail Economic Indicators
First, the usual pit stop on our gradual descent into an economic hellscape.
We got a new read on inflation in April, and it was not good. It wasn’t even that inflation is back up over 3%, it was that the producer price index was at 6%. Economists are predicting that April will be the last good month for consumer spending, as PPE inflation become CPE inflation. Part of the reason why? Because inflation is now officially outpacing wage gains in the market.
I feel like I’ve been saying “Winter is coming!” since December 2021, and it still hasn’t arrived. I don’t personally remember the oil shocks of the 1970’s (though I will admit I was alive then), but I feel like anyone my age or younger are all Summer children who have no concept of winter (sorry for the extended Game of Thrones reference, if you have no idea what I’m talking about). Even though the US is unlikely to have lines at gas stations, we are all about to find out what oil dependency really costs us.
AI & UX: Change Will Come
In the meantime, we have to soldier onward in the quest for the impact of technology on retail’s future, in part because I’ve been wrong before and I could be wrong now (about consumers’ willingness to spend). And the technology progression – and consumers’ and employees’ adoption of it – does not stop, even if the economy tanks.
So let’s start with the tech part of it.
What kicked off thinking about the future of UX in an AI-driven world for me was this article that looks at 10 UI patterns that may not survive. The ones that jumped out at me are of course the ones that show up in the solutions I look after in my own job: static dashboards and pre-built reports – something I’ve been thinking about a lot lately, but also manual data entry forms, manual search and filter selectors, and CRUD interfaces with heavy data tables.
In inventory management, picture a morning dashboard that shows an allocator the top things they need to care about today. One thing might be a worklist of decisions that need to be made, one may be to review a recommended action and approve it, one may be an alert on a margin trend. Tomorrow, it may be three completely different things. One thing that needs attention may need a bar chart. One may need a grid that highlights something that is out of bounds compared to other items. One may pull up a new allocation order to send items to stores for the allocator to approve.
The UI is generated based on what the allocator needs to pay attention to, and further generated depending on what actions the allocator needs to take. Maybe for each specific thing there are some UI’s we recognize today – a grid, an input form. But they don’t really exist until the allocator needs to see them.
This is a future where maybe the idea or the template exists, and AI chooses which one to use, but I don’t think it’s even as structured as that. The AI will learn how to best present the challenge or the options based on a combination of the problem itself, the options, and how the particular user it’s interacting with likes to think.
Now try a store. I’m still not 100% convinced that store associates are going to spend a lot of time talking to their devices. I can be convinced otherwise, but at the moment I see too much noise, too much confusion with customers asking “are you talking to me?” and ear pieces that need to be yours and yours alone, with all the challenges of “I forgot mine” or “I lost mine” that comes with that.
Anyway, the interaction definitely works best with talking, so let’s pretend that Apple’s sub-vocalization patent comes to AirPods quickly. Imagine just saying “Hey Tillie (because we’ll need a cute name for ‘Till’), add the black off-the-shoulder sweater to the cart for Nikki”. The cart is there, but no one needs to see it, let alone have it up on the screen, just to add to cart.
And Tillie can always talk back – “what size sweater?” or “there’s a BOGO on sweaters, do you want to add another?”
This won’t end up with a blank screen with just a prompt box, either. Google released a really cool interaction method called Magic Pointer. You put your mouse pointer on something on your screen and the LLM understands the context. “Tell me more about this” gives you a different answer when your pointer is hovering over a sales number vs. hovering over an inventory bar chart.
Thinking Machines released several videos showing much more conversational interactions versus typical chat and response. And that’s just getting started.
As the article above states, “The best interface is the one that disappears.” As I’m sitting here still typing out my thoughts, with my memorized shortcuts for italicizing and bolding and saving, I’m very self-consciously aware that we have learned the interface and not the other way around. But one thing that AI can definitely do quick and easily is generate a UI. So why should those of us in software spend so much of our time trying to build one?
I was thinking about it as “dynamic UI” but it really could go as far as “dynamic software” – where AI generates the capability you need to do something along with the UI you need to do it. On the fly.
This will hit reporting and analytics first, where the question is “show me” and not “do this”. But “do this” won’t be far behind.
Interlude: What About All That “AI is too expensive to survive” Stuff?
I have to take a minute here to say two things:
One, I still believe we are not paying what AI costs. Saying that I could see a future where UI and even the underlying capabilities are generated on the fly sounds like that completely contradicts my concerns about costs and I readily admit that. I am still very concerned about costs. Open source is coming up as some possible alternative, but I have lived through trying to run enterprise software entirely on open source and all we did was end up externalizing all the costs we would have paid a vendor. So color me skeptical about that option.
Two, we have to live in both worlds: what is possible, with no consideration for how much it costs (just in case someone figures out how to make it way cheaper, or, I don’t know, OpenAI takes down Microsoft and Oracle with it and ChatGPT ends up owned by the US government and we pay for it like electricity) – and also thinking about AI’s future knowing that we do not have a clean path towards cost-effective AI use today.
Having an AI repeatedly generate the same chart for multiple users over and over again sounds wasteful in a way. I don’t know how we’ll manage that, but I bet you someone is thinking about that right now.
AI & UX: Retail’s Future
I’ve given two examples of how people interacting with their solutions may change in retail. But now I want to take a step back and think about this more broadly.
When I look at how technology is going to impact the retail enterprise, I start with consumer adoption of technology. For retail, how are consumers changing their shopping behaviors or how is technology changing what consumers expect from retailers?
Next, I look at how employees adopt technology – and this is really from the point of view of, I use this at home, how can I use it at work. LLM adoption followed this pattern for sure because OpenAI went wide and basically free, so consumers were able to think of all kinds of ways they could use a platform LLM with the only constraint what they were willing to try. And companies had to scramble to catch up when those consumers wanted to do the same for their work – with big issues around proprietary company data making it into model training, for example.
For retailers as the companies in question, they have a choice. They can help enable customers and employees or not. Retailers can be a boulder in the stream and sit there stuck while customers and employees alike just flow around them. They can engage and collaborate and try to help. And/or they can stay out of the way, watch where the stream flows and come along to pave it to make the stream flow more smoothly.
What retailers must remember, though, is your role in all of this. Consumers engage with retailers because they need something or they want something. And sometimes it’s not the product that they want, that’s just a side effect of the outcome they’re seeking, a concept embodied by retailtainment.
While retailers have always been in it to sell stuff, as consumer wants and needs have grown more sophisticated over time, so have what retailers need to do in order to successfully sell stuff. That means you’re not a grocery store. You are family meal enablement. You are solving the problem of how to feed a household of people on whatever budget and time constraints they have set for themselves. This is solving a problem, not selling them food.
You are not a clothing retailer. You are “a summer beach day in a shirt” purveyor (Tommy Bahama, for example), “a person who loves enjoying the outdoors” enabler (REI, for example), a stylish work professional wardrober (Ann Taylor, for example). In some of these cases, the retailer is solving a problem (what do I wear to work?) and sometimes the retailer is feeding a want, a desire: I want to feel like I’m on the beach even if I’m stuck at home and it’s pouring rain outside.
It gets way too easy for retailers to forget that they are fundamentally about solving problems, even if the problem is “I’m bored.” When retailers talk about the customer journey and really only mean the purchase journey, they have forgotten. They have equated “selling stuff” with solving problems.
So when I read articles that hype how AI is hollowing out the eCommerce website, I am generally in disagreement. Not that AI is not eating discovery and selection – it totally is. But eCommerce sites were never good at selling stuff, forget about solving problems. And stores generally have been much better at both. I will absolutely argue that one of the reasons why stores have a 25-30% conversion rate and eCommerce sites have a 3-5% conversion rate is because store associates can help solve problems and not just sell stuff.
Thinking about how to revamp your website so that it can be groped by agents more easily is not addressing the fundamental issue. What answer engine platform LLMs are doing is not cannibalizing discovery and selection. They are solving consumers’ problems. Can I shout that for emphasis? They are SOLVING CONSUMERS’ PROBLEMS!!
Making your website agentic-ready in no way helps you stay relevant to consumers. For whatever slice of brand positioning you have staked out, what problems are you solving? What needs and wants and desires are you enabling? How can you help consumers solve these things? If you let go of trying so hard to sell stuff (I know, I know) and treat it as a side effect instead of the main objective, then the world will open up before you, with all kinds of ideas.
Because what is happening here is not that answer engines are giving consumers the same thing that retailer websites have always provided, only somehow better. Answer engines are solving problems. It’s actually better to think of answer engines more like store associates than a reimagined web UI.
Ask the right question, and you have a much better frame for understanding how the water is flowing. Are you going to be a boulder, stuck in the streambed while consumers and employees pass you by? Are you going to get in on the learning curve with your customers and employees and see how you can help them along? Are you going to watch which way the water flows and come in behind to follow their lead?
What Have We Learned This Week?
During the Internet Age, recombinant innovation was the point where the hockey stick really shot upward on what was possible. The internet was cool, but when you combined it with mobile and webhooks, people started putting things that had been around for a while together in brand new ways and invented whole new business models. Uber as the most obvious example.
We have not entered the recombinant innovation point for the AI Age yet. Rethinking UI is starting to feel like the prelude to the opening act, though. When I think of the acres of grids encapsulated in merchandising solutions, for example, and I think of someone just saying, JARVIS style, “Hey, we’re going to move up introducing the new summer line by two weeks. Leave the old spring stuff in stores that look like they’ll sell through it in time but have the rest of the stores pack it up to send to outlets next week.” And then have that analysis run, and the transfer orders created and sent out?
And then have a progress report on those transfer orders get automatically added to that person’s dashboard to monitor until it’s all done?
And as a software provider I don’t have to spend any time thinking about the right way to display any of that data? Or build out endless ways of enabling the what-if analysis on a grid?
It doesn’t matter if we can afford that future today or not. It’s worth building towards.
Program Note: Next week is Memorial Day weekend, which means the Pulse Report will come out on Tuesday, not Monday.
Until then!
Nikki
I write this weekly article as much to force myself to make sense of what I read on a daily basis as to say anything about it publicly. I will continue to do so - for free - for as long as I can sustain it.
If you like what you’ve read and you haven’t subscribed, please do! It helps me know I’m saying interesting things.
If you’re already subscribed but you can think of someone who would enjoy reading this, please share the article.
Or better yet, share the whole publication and let them get in on the weekly fun!
Thanks for reading, and thanks for your feedback and support!