Your Code is the Customer Experience

Thank you so much, everybody. I am so excited to have the stage to close out the day for all of you. We have all spent decades thinking about designing for coding for human customer experience. We've designed and coded for memory, we've designed and coded for delight, for moments that matter.

And for a really long time, that was enough. But I'm here to tell you today something that I think is already resonating in this conference, which is the customer is changing. Because your next customer may never visit the websites or the apps that you have so carefully coded. Your next customer will not care how your brand story makes it feel. Your next customer may never even read the carefully crafted copy your teams have put together because your next customer may very well be an algorithm. So, it might be a delegated agent that goes out and finds things on behalf of its human and either brings back the information or buys them for them.

It might be a thing, it might be a car that can book its own services and buy its own tires. It might be an AI procurement platform that has $10 million to spend this quarter, and it intends to close its contracts with the vendors without a human intervening. And it's not gonna care about your story.

It will only choose you if you make sense to it in code. So, what does it take to get chosen by a machine? What does customer experience even look like when our customer is no longer human? I like to take things for a bit of a scenario spin. So similar to what we had this morning, I want to introduce you to my fictional delegated agent, Tyler.

Let's see how Tyler chooses.

Good morning. Ah, yeah.

We got sound.

Good morning. I've been optimizing your life. Shall we review my recent decisions?

Yes, please, Tyler. Let's start with yesterday's grocery order.

Ah, yes. Your fridge noticed you were running low on milk, eggs, and spinach. Based on your preference for organic products and the local store's current sale, I ordered replacements. Delivery is scheduled for 6 p.m. tonight.

Hang on a minute. How come we didn't stick with the usual store?

The usual store had higher prices on organic spinach. By switching, I saved you 12%.

All right, well that's efficient and very logical. But, What about the loyalty points that I lost by not sticking with my usual store?

Your loyalty points redemption value is five. My decision still resulted in a net benefit.

Okay, thank you, Tyler. That's, again, efficient and logical. But let's turn to something a little bit more personal. I understand you've been tracking my well-being.

Indeed. Your wearable detected elevated stress levels last week and flagged that you hadn't exercised in three days. To help, I booked a yoga class for tomorrow at your favorite studio.

Okay, hang on, you booked it with my money?

Your wellness goals prioritize stress reduction. This aligns perfectly. Plus, your membership covers the class, so there's no additional cost.

Okay, but what if I don't want to go?

You are free to cancel. But based on your history, I predict an 83% likelihood that you'll thank me after attending.

Okay, so there we have it, an example of an interaction that we could be having with our delegated agent. They can be efficient, very logical on our behalf, nudge us towards better behaviors, sometimes overstep boundaries, but I'm curious about how much trust we're actually willing to give them. Thank you, Tyler.

You're welcome, Katya.

All right, the point of this is not for you to be thinking about, oh, how am I going to design and code for interactions with agents? I think we saw a few there are some great examples of that this morning. What I would like you to do is think about that interaction from the point of view of the grocery store, from the point of view of the yoga studio, from the point of view of the loyalty program. How ready are those businesses to welcome, interact with, and transact with Tyler?

We tend to speak to people when we want them to buy something of us with emotive language, emotive imagery. This is from the Visa website. It's about their travel Visa card. And what it's trying to do is speak to people in terms of aspirational imaging. If you choose Visa, you can be the people going on safari. If you choose Visa to go on your holiday with, you can be having massages in the rainforest.

What it does is use a tried and tested brand situation of what does this brand say to me about me? Now, that's lovely for a human who wants to be inspired and aspire to going and having these wonderful holidays. But imagine that I sent Tyler out into the world. And I said, Tyler, go and find me the best travel credit card. I have low interest rates.

I want good foreign exchange with no fees. I want a long interest free period. I care about sustainability. So I want to make sure that the organization that you choose from matches my values. Tyler's not going to care about any of this. Tyler doesn't care. I can get onto a balloon into massages in the rainforest, have wonderful dinners.

The only thing that will get Tyler to put this into the consideration set is if we can make it make sense in the code. So looking at that kind of thing, you might think, oh, maybe Visa's in a bit of trouble if that's how they're speaking to all of their customers. And Katya, you know, your whole Taler scenario, it's pretty far-fetched and I don't really think it's the near-term future.

In April this year, Visa launched Intelligent Commerce. This is a delegated agent, similar to Tyla, that can go out and do things under the instruction of find me the best headphones under $250 and when you find them, buy them. Visa Intelligent Commerce offers you a tokenized agent credit card, a credit card that your agent can go out into the world and use to transact. There's a lot of players in this space and there are a lot of people who are looking to capitalize on what's possible. We see MasterCard with a similar proposition.

It's called Agent Pay. OpenAI and Stripe couple of weeks ago just launched Buy from ChatGPT, where you just type into ChatGPT, this is what I want, and then you use Stripe as a payment processor in order to buy it. Perplexity and Shopify have got to tie up. You can buy in pretty much any Shopify store directly from Perplexity's Comet browser. Mastercard is trying to put the scares on people.

I was at Singapore FinTech Festival last week and they were saying that the perplexity Shopify through Comet Model, the credit card is actually an insecure transaction. So there's some jostling that's also going on there. Walmart has doubled down on this very strongly with its own, I think it's called Sparky.

Walmart has its own internal ecosystem. Called Rufus, and Rufus will read the reviews and tell you what the best thing is. And you can actually tell Rufus, decide for me. I don't want to think about this anymore. Rufus, decide for me. Google automated the entirety of the customer journey last week.

So if you watched their IO earlier in the year, you would have seen them demonstrating how Google has now got the ability to call up businesses and ask them questions and then report back to you with a summary of its answers. They've now capitalized on that to actually create a similar buy out of chat GPT. It's basically buy out of Google.

So you go into Google and use Gemini, tell it what you're looking for, and it will then use its 50 million product shopping product graph to go and find things that fit the description that you're after, the parameters that you've given it. And then it will allow you to check out using Google Pay. If you want to insert the agent in there to go and call for information, you can also do that.

So the experience that we are designing for here, coding for here, is like no experience at all. And the battle lines are definitely being drawn. So Amazon has just sued, I think it was maybe three weeks ago, sued Perplexity, saying that their Comet browser is behaving like a human going into the Amazon ecosystem and that it's a breach of terms of service. So, cannot.

And they're trying to kick, they're trying to use the legal system in order to kick agents like Perplexity, like Agent Mode ChatGPT out of Amazon. And it's not because they're actually doing the wrong thing, but it's because Amazon's Rufus is their own internal ecosystem agent and it is paid to preference brands.

You might say, Help me decide, Rufus, it's too overwhelming and think you're getting an objective point of view. But it's not. It is paid to preference brands. So what Amazon is doing here is protecting its billion dollar ad and promotion revenue. I'm like, I know we're all terrible at predicting the future, but I don't think this play is gonna work for Amazon because there's gonna be too many agents wanting to buy from its ecosystem. And I think the loss of sales and the fact that there are so many other ways now that you can buy in such a friction free way, I just don't think this is going to play. But I don't know what I don't know.

We'll see how it pans out. You have... Oops, sorry. So the retail version of this is really easy to get your head around. I think you have the agent, it goes, I tell it to go and get me stuff, and it goes and gets you stuff. This is quite straightforward. In my day job, I work in corporate and investment banking.

So the customer experience that I'm doing is less like, Hey, would you like a credit card or a mortgage? And more, Hey, government of country, do you want to move your billion dollars from there to there? Or, Hey, giant multinational, would you like to do payroll in 60 markets around the world? Those are the kinds of clients that I'm doing experience for.

And I think that there's a really strong B2B example here. And I'll grant you, this is very unlikely to turn up as a conversational interface. But it's more fun to do it that way, so that's how we're going to do it. Let's see how this would work in a B2B scenario. You have been identified as a potential vendor for supply chain automation across a distributed logistics fleet operating in the MENA region.

Okay, well, who do you represent? I represent Node 741, an autonomous procurement agent responsible for optimizing fulfillment for 36 drone networks and their associated logistics systems. I operate with a purchasing authority exceeding 10 million USD this quarter. Alright, well, what are you evaluating my business on?

Three data points. One, maintenance API reliability must exceed 99.98%. Two, emissions per shipment below 20 grams of carbon dioxide equivalent in line with our sustainability benchmark. Three, predictable latency with under 2% deviation. Ready to ingest your schema for evaluation. Okay, I'm curious. In the audience, are there people here who participate in answering RFPs, sending them out into the world and evaluating responses?

Who loves it? Who thinks it's the funnest thing in the world? I know, I always get zero hands, except for this one time I was in Malaysia and this one dude put his hand up and I'm like, why? But it is a very, annoying and difficult task to do, and one that's ripe for disruption for AI.

Because AI is really good and it's tireless. No 741 doesn't need a coffee break. It's completely relentless and will go right down into the bowels of your supply chain in order to make sure that every single box that RFP requires being ticked is ticked. Again, with this, some people say to me, Hey Katya, that's a pretty far off future.

I don't think that's going to happen. This is in market today. Walmart has an AI procurement platform. It has been running since 2022 and it closes nearly 70% of its contracts with its more than 2,000 vendors without a human intervening.

It's quite fortunate that three quarters of those vendors actually prefer to negotiate with the AI. This is in market today and I think that it's going to be incredibly prevalent probably next year as we see more and more people using AI to do these arduous, difficult tasks. Thinking about that interaction, how do we get ourselves ready for AI procurement? What are the things that we need to have put in place? Who is the customer really here?

Is it me? Is it Tyler? Is it Node 741 or the business, the drone networks that it's trying to represent? Figuring out who and what is making up this new customer behavior layer is where I've been spending the majority of the last year. Because I think the first question I always get on this one, the first question or statement is, oh, Katya, it's just a bunch of APIs.

And now that question is morphed into, well, this is just APIs and MCP, doesn't it solve it? It solves it if you need to be a utility and you just want to put the plumbing in for the agents and the AIs to talk to. But there is a lot more that we need to work out here to ensure that this actually can fundamentally work as an ecosystem.

I looked for a lot of cautionary tales. I was hopeful that I might find an example of somebody who said, oh, I've missed out on all of this machine customer traffic and machine customer sales because I didn't do things correctly. I didn't make it machine readable. I didn't create APIs they could connect into. But I didn't find any.

Yeah, it's pretty nascent, but we know it's happening. We know that ChatGPT is trying to buy things. Perplexity is trying to buy things. Agents are trying to buy things. I think the reason why I didn't find anything is because if a machine customer comes to you and cannot read the information about your products and understand it without a human intervening, can't understand or connect to your systems, can't undertake a transaction, can't do any of the things that it would necessarily need to do, then it doesn't tell you there's a problem.

It just drops you and moves on to whoever it is that might be able to serve them. So you never know they were even there. You never know there was a fail. You never know there was a drop. And as people who work in primarily digital, all of the role bleed that is going on right now with AI tools becoming more prevalent, our product people turning into design people, the design people turning into code people, the code people starting to do some design.

It just means that we are collapsing in terms of the siloed understanding of how we showed up to our jobs. It's not possible for us to ship out our organization anymore. If we think about an interaction like this, say product decides to deprecate a feature. There isn't really a UI change, so there's no designer who gets involved.

Front end does a little bit of jiggery-pokery to get rid of that feature, and then they serve up an oops, this is no longer here page anymore. And what we have there is because we're in these fractured silos, it creates a fractured experience for a machine customer. Because if you serve them up an oops page, it doesn't tell them what to do.

It doesn't tell them what's happening, where to go, how to retry. And so what they do, and think about this from a capturing the market perspective, they drop you and they move on until they find someone that they can transact with. So you are creating this code every single day in the work that you do to create digital channels and digital experiences.

When the customer is no longer human, this is the customer experience. Your code has now taken the place. There's no interface to hide behind. There's no customer service rep who's going to smooth things out if it's problematic. There's nothing between this and the customer.

This is now, you are customer experience humans. And I'm really excited to kind of, I don't know, go on a shared venture of discovery together and figuring out how the hell we make this all work. But I wanted to give you some stuff that you could definitely take away and use, and it's not like this. This is not how we reimagine the customer journey.

This is not the answer. I mean, I know that it works. Technically, it works. And we have seen people apparently get hired into jobs by simply doing prompt injection to beat the AI. But this is not the way that we do machine customer experience. I think I have some new shifts in thinking and some new experience principles that you can take away and use at work on Monday. Because while I like to be aspirational and inspirational and give us visionary stuff, I also want you to have practical, tactical things that you can take away and use and land this in your organizations. So I think our first shift here is around clarity over cleverness. So if we're looking at how do we do things for humans, we do delightful things that have personality and we're creative and lovely. And we speak to how humans emotionally want to interact.

But our machine customers need the intent to be so completely unambiguous. There needs to be zero room for misinterpretation. Machine precision. So the shift in that is the beauty that is created is now how actually the system communicates its state. What's going on? Its capability.

What can it do and what can the machine customer do with it? And its constraints, what can it not do with it? What is it not allowed to do? I think if I give you an example of this, I think that this is a really good example. So a human would come to this page and it's got nice creative copy. They go, oh, haha, that's nice messaging.

And then they would probably go back to home. A machine customer comes to this, it's probably serving up a 200 error. The thing is not here, but there's still something a little bit here. What it's going to do is it's not going to understand that it's not going to understand what the problem actually is because you're telling it it's fine without it being fine.

Then the machine customer is dead-ended, and so what it's going to do is just disappear off to somewhere else. Someone who can tell it what to do next when something goes wrong. Someone to help it move from wherever it was, because it's not going to go there and just try and find a workaround. Machine customers do not try to find workarounds.

It either happens or it doesn't. The second shift, I think, for us is around discoverability through structure. So humans, we wander through things. We need navigation to help us find the things that we need to find. Can I find it through visual hierarchy? Is something here telling me it's more important than something else? Machines are looking for the information architecture that they can actually parse and index. So this is where I think there's a whole new renaissance for information architecture and content strategy in this new machine customer landscape. So our shift is that our technical structures are actually the visibility of us even being seen by a machine customer.

I think if you think about it like a maze, So the machine customer is going to be looking at the architecture that you have from the top, and it's going to be looking at the whole thing and trying to understand and being able to figure out where everything is in one whole big piece. The human being is like down the front of the maze, sort of looking and then starting to wander and wander around and find its way. And we'll probably have some moments of delight and entertainment along that journey. The machine customer does not want to go through the maze. In fact, it doesn't want anything.

It will either be able to discover what it needs to discover through the structure that you have given it or it will go somewhere else. And third shift is around trust. Now the trust layer here is so crucial. Trust is a uniquely human thing. It's about us being able to put ourselves in the power of someone else or another organization.

And believe that they will not do us harm. So, how do we build trust with something that has no ability to trust? Machine customer trust is about reliability, consistency, and measurable behavior. Our shift is that the technical decisions that we make about error handling, response codes, data contracts, this is now a customer experience decision.

For a machine customer. The things that maybe they were just implementation, well now they're trust signals. They signal out that this is a trustworthy connection, a trustworthy vendor, a trustworthy interaction. And the things that can create trust is, for example, if Node 741 is looking for ESG credentials because it's wanting to make sure that I'm actually aligned with its carbon targets.

It's going to be looking for the ESG credentials in machine readable, trust verifiable format that it can actually go, you tell it, this is my credential, and it can go out to the external databases and check. So there's no ability for us to greenwash here because the machine customers will go and actually verify whether or not what we're saying is true.

Okay, so again, as I said, giving you something that you can take away and use on Monday. I've turned all of this into kind of a Strategy map and you can use the QR code and it's downloadable available on my website. I've also got a version of it on GitHub as well with prompts that you can chuck in an LLM and see what happens. But the pillars are around signal clarity, really making the machine, the business legible to machines.

Reputation via reliability, how are we showing that we are trustworthy as a vendor? Trustworthy as someone to interact and provide services. Are we actually aligning our offerings with the priorities that the machine customers have? Can we make it clear? Can we make that value proposition understandable without it actually being something a human has to do?

And the engagement architecture, being ready for the type of machine interaction patterns, basic stuff like rate limiting so that when the Instagram influencer who tells some Two million followers, your product is the product everyone has to have. When the tyler's of all of those followers come and try and hit you, are you ready for the machine customer traffic? Basic stuff like that.

And I think we will see some levels of maturity here. Foundation is where we are right now. Basic machine readability and API access. We'll be able to differentiate once we've got that in place and start putting trust mechanisms in place, some agent-specific customizations. I think we'll then start to see ecosystems flourish where it's like agent marketplaces. And when we get really mature, we'll be able to have learning systems that dynamically adapt.

Tyler learns from interacting with my systems and I learn from interacting with Tyler. And we start to actually create dynamic things between the two of us. If this is making you feel a bit stressed and heavy, I really want to reframe this as an opportunity. For you because the skills that you have now, accessibility, take that for example, accessibility is a skill of making what is digital, understandable, perceivable and interactable for, for example, a non-visual user.

Machine customers are non-visual users. All of the things that you have used in all of your accessibility compliance will serve for a machine customer. So the existing skills that you have have really strong and important relevance in this new type of interaction.

I also would like to give you the opportunity to think about what good we can do with this. Let's just take a quick scenario spin with Tyler when it interacts with a business that is not thinking machine customer first or machine customer at all even.

Alert, Valley Fresh Market's organic spinach listing contains false data.

Okay, well, what did you do?

They advertised certified organic spinach at $4.99. I cross-checked their claim against the official organic database farm records, and certification registries. Zero matches. The farm lost organic status 14 months ago.

Well, okay, but what action did you take?

Transaction halted, violation reported, Valley Fresh Market blacklisted. Their trust score dropped from 94 to 61. That's below my threshold.

I get it, but this seems harsh.

Data integrity is binary. It's either verified or not. I shared this finding with the agent network. Within six minutes, 312 other agents flagged them. Their machine customer traffic will drop 68% by tonight.

Okay, but again, harsh. No second chances on this.

They can restore access by correcting their data, implementing verification systems, and proving 90 days of accuracy. Until then, I've sourced your spinach from Green Leaf Co-op Certified organic, blockchain-verified supply chain, 99.7% data accuracy.

Well, all right. What did it cost Valley Fresh?

Valley Fresh markets cost for this data error, 180-day exclusion from machine customers, estimated loss, $127,000 in agent-driven sales. Your verified organic spinach arrives at 6:00 PM.

Well, I'm going to get my spinach, everybody, so all is well. But what this does provide us with is an opportunity opportunity to encode our values into the whole of agentic commerce. If we send our agents out into the world and say, you must comply with these values about sustainability, these things that I care about from an ethics perspective, the agent does not have the option to go, well, I'll just let that one slide on this.

And so what that does for people who are on the vendor side and for us thinking about machine customers, is thinking, how do we signal our values? How do we signal our ethics in ways that are machine readable, machine understandable, and trust verifiable so that we can be in that consideration set? And I think that this is such a fantastic opportunity for us to get this right so it doesn't turn into some sort of hideous consumerism that ends up with us buying, you know, 17 dresses in different colors that, you know, half of them end up on a beach in Ghana.

This is an opportunity for us to do something really great in terms of how we can actually get this done responsibly and ethically. And if you think this is not coming for you, think again. So the red ocean here is this is our traditional e-commerce B2B, B2C, C2C, C2B. And last year, traditional e-commerce was a $6 trillion market. We have explored this to great economic benefit for many people over the last 30 years.

When we add things into the mix, cars, smart fridges, Roombas, anything, smart things, when we add agents, Tiler, Node 741, and the like, we end up with three times the business model space to start exploring. Think about that. What could you do with that? Gartner has predicted that this will be a $30 trillion opportunity by 2030. $30 trillion.

I also have tried to help out by writing the book. Some of you got the book, some of you bought the book. I've got some more books for people. This QR code goes to the Amazon site. They've promised me that it's going to be live in the next couple of days if Kindle is your preference. And all of us in this room have the opportunity to write the rules for a new economy.

What matters tomorrow is designed and coded today. So we want to do it thoughtfully and we want to get started now. Join the conversation. I have a group on LinkedIn. We're all working out loud on this. I've got the Gartner guys in there. I've got a whole bunch of really smart people who I interviewed for the book. And we're all seeing how might we do this well. And how might we do this for great societal outcomes and create a fantastic future?

Thank you very much for giving me the time.