Artificial intelligence is beginning to move beyond helping consumers discover products toward taking a more active role in purchasing decisions. Klaviyo’s recent Singapore research found that 76 percent of shoppers had used AI to help find or decide on a purchase, while 51 percent said they would be open to AI agents making routine repurchases on their behalf.
That shift could change more than how products are discovered. If software increasingly compares options, interacts with brand systems and eventually transacts on a consumer’s behalf, websites and advertising may no longer be the only interfaces brands need to optimize. Product data, APIs, customer identity and the systems sitting behind the storefront could become increasingly important parts of the customer experience.
The transition also raises questions around consent, accountability and human oversight. Klaviyo’s research found that 58 percent of shoppers would share more personal information in return for more relevant recommendations, underscoring the tension between useful personalization and data collection that customers may consider intrusive.
In this TNGlobal Q&A, Brian Kealey, Managing Director for APAC at Klaviyo, discusses what agent-led commerce could mean for customer relationships, first-party data and loyalty, where human judgment should remain explicit, and which parts of an AI-driven customer strategy can realistically be standardized across Asia.
Klaviyo’s Singapore research found that 76 percent of shoppers have recently used AI to help find or decide on a purchase. As discovery shifts from search and browsing toward AI assistants, what changes for brands that have traditionally relied on websites, marketplaces and advertising to influence the customer journey?
As AI becomes a normal part of how people shop, discovery stops being something a brand can shape stage by stage. When an assistant is doing the finding and fetching, it becomes the thing that shapes the customer’s first impression of a product, not you.
When someone asks an assistant to compare options or build a shortlist, the brand is no longer framing the first look; the assistant is. Working from whatever it can find about the product, it raises the bar on basics such as whether your product data and pricing are accurate enough for a machine to represent you well.
It makes the relationships you actually own far more valuable because, as third-party signals keep fading, first-party data is what lets you stay relevant when the discovery layer is no longer yours. In an agent-led world, that direct relationship is what keeps you in the conversation at all.
More than half of respondents said they would be open to AI agents making routine repurchases on their behalf. If software begins making some purchasing decisions for consumers, how should brands think differently about earning consideration and loyalty when the immediate “customer” may increasingly be an agent rather than a person browsing a storefront?
The temptation will be to win the agent’s attention the same way brands have always competed for a shopper’s, and I am not sure that translates cleanly. What the 51 percent are pointing to is routine, low-consideration buying. Those are repeat orders people already make without much thought, and that is where agents show up first.
In this model, the shopper is increasingly acting through an agent rather than dealing with the brand directly, so the real risk is commoditization. The task shifts from persuading a person in the moment to becoming the option the agent keeps choosing, and that comes from being consistent over time and from having a product someone already picked once and was happy with.
There’s a second layer to this that most brands aren’t building for yet. Right now, the assumption is that a shopper’s agent is coming to read your website or your product feed, so you optimize to be legible to it. Soon enough, that agent won’t be reading your site at all. It will be talking directly to whatever runs your side: a pricing engine, a loyalty system, a customer service bot.
That conversation happens entirely machine-to-machine, and it’s a different problem to solve. The API and the data structure become the storefront, not the website. Brands that treat this as a marketing problem will lose to the ones treating it as a plumbing problem.
Loyalty becomes less of a campaign you run and more of a place you hold in someone’s automated routine. Even as more of the buying gets automated, the brand has to stay the owner of that customer relationship rather than handing it wholesale to whichever agent sits in the middle.
The research also found that 58 percent of shoppers would share more personal data in return for more relevant recommendations. How should brands interpret that without treating it as blanket consent? Where should the boundaries sit between useful personalization and data collection that becomes intrusive or disproportionate?
I would read that as a conditional yes rather than open permission, and the condition matters as much as the headline. Nearly half of shoppers say the thing they most want is for brands to stop sending them irrelevant messages, and a similar share want to be treated as loyal customers rather than transactions. People are willing to share more, but they are doing it in exchange for relevance they can actually feel.
So the line I would draw is around usefulness. When data clearly improves the next interaction, customers tend to experience it as good service. When it is collected with no obvious benefit to them, or it surfaces somewhere they did not expect, it starts to feel like surveillance instead.
We always talk about treating every customer as if they were the only one, and the other side of that ambition is using the data responsibly and well.
A practical test is whether a brand could explain to a customer why it holds a particular piece of data and show how it improved their experience. If the answer is no, the data is being kept for the brand’s benefit, and that is where trust erodes.
AI-driven customer engagement depends heavily on the quality and context of first-party data. What are the most common data problems that prevent brands from delivering genuinely useful personalization, particularly when customer interactions are spread across e-commerce, email, messaging, service and physical channels?
In my experience, what looks like an AI problem is often a data problem that AI has simply made impossible to ignore. Generic AI produces generic results because it does not know your brand or customer history. All that information lives in your data, so the quality of the data sets the ceiling on how good the AI can be.
The data problem shows up in a few ways, the most common being a fragmented view of the customer. The same person appears as a browser on the website, an email address in your marketing tool, a phone number in messaging channels, and a support ticket in the service desk. None of it is joined up.
So a brand could know a great deal about its customers in aggregate while knowing very little about any one of them at the moment it matters.
The second issue is timing. Personalization built on yesterday’s data misses what someone did an hour ago, which is often the most telling thing they did. Take a shopper who browses a category this morning, leaves a cart full, and then starts a return in the afternoon. A system working off last night’s snapshot sees none of it and keeps pushing a generic promotion while the signals that would have won a sale, or told you to hold off, go to waste.
The customer just sees a brand that isn’t paying attention, which is the kind of message they tell us they want to stop getting.
Finally, one that catches a lot of brands out is that service data usually sits apart from marketing. Someone who just had a frustrating support experience receives an upbeat upsell soon after, simply because the two systems were never designed to talk to each other.
Most of these are orchestration failures, and sorting this out is really a leadership decision before it is a technology one. It comes down to committing to a single view of the customer that every channel can both read from and contribute to.
Personalization is often measured through conversion, click-through or revenue. What other indicators should brands watch to determine whether AI-driven engagement is actually improving the customer relationship rather than simply making marketing more efficient?
Conversion and revenue tell you that a message worked on the day. What they do not tell you is whether the relationship behind it is in good shape, and that is what I would watch more closely.
The measure we treat as the real sign of brand health is customer lifetime value because it looks at the whole relationship instead of just the last click. If it’s climbing, the engagement is building something. If it’s flat while you send more and more, you are borrowing against future patience.
Alongside that, watch the signals customers send when patience wears thin. Unsubscribe and opt-out rates by segment are the most honest feedback you will get on relevance. The time between purchases tells you whether you are building a habit or landing the occasional one-off.
The trap is assuming growth comes from more campaigns. It doesn’t anymore.
What people want most is for brands to stop sending things that do not matter to them, so a program that becomes more efficient while also becoming more irrelevant is winning on paper but losing the customer.
Asia is fragmented across languages, cultures, commerce platforms, messaging channels and privacy regimes. Which parts of an AI-driven CRM strategy can realistically be standardized across the region, and which parts still need to be designed market by market?
Standardize how you understand the customer and localize how you show up for them.
The part that can and should be consistent is the data foundation. How you identify a customer and bring their history together needs to hold across the region because, if every market builds its own version, you lose the ability to learn anything at scale.
That single, real-time view of the customer is something I would always want standardized. It’s also what lets you face outward, not just inward. The same unified foundation that ends your internal silos is what allows you to expose your brand safely to the agents that will increasingly transact on behalf of the customer.
Get that plumbing right once, regionally, and you can plug into agent-led commerce as it arrives. Get it wrong, and you’ll be retrofitting it market by market, after the fact, under pressure.
What has to stay local is the expression of it. Channel preferences vary from market to market, with WhatsApp carrying the conversation that email carries somewhere else, and the etiquette around each is different as well.
Language is not only translation; it’s also tone and knowing if and when it’s even appropriate to get in touch.
Privacy laws are set locally, so consent and data handling must be designed around local law rather than one template. If you try to standardize the whole journey, you usually end up with something that feels slightly foreign everywhere at once.
As more customer journeys become automated, where should human judgment remain explicit? Are there particular decisions around sensitive customer data, complaints, high-value purchases or unusual behavior, for example, where brands should avoid allowing an AI system to act autonomously?
Brands should define the outcomes and set the guardrails agents operate within. Automation takes the routine; people stay responsible for the exceptional, and most of the skill is in judging which is which.
Human judgment needs to stay in the picture wherever getting it wrong is costly or personal. Anything with emotion attached, like a complaint or service failure, benefits from having a person because the customer is looking for acknowledgment as much as they are looking for resolution.
High-value or irreversible purchases are similar, where people want to know a human is within reach even when most of the process runs itself. Anything unusual enough to suggest fraud or distress is far better paused for a person to look at than acted on automatically because those are the moments where a confident system can do real damage.
There’s a version of this that’s easy to miss because it doesn’t involve a customer at all. When a shopper’s agent negotiates directly with a brand’s pricing or fulfillment system, there is no person on either side of that exchange to catch a bad outcome in the moment.
That is exactly where the guardrails matter most: hard limits the brand’s own systems cannot be talked past, and a way to confirm the agent on the other end is actually acting for who it claims to be.
I’d trust a system less for how fast it negotiates and more for whether it refuses to move past a line it was told not to cross.
The design question is more about where you place the handover and whether the brand keeps control over how the system shows up. A model that acts with confidence on a case it should have flagged worries me a good deal more than one that flags too often.
Looking ahead, what do you think will be the biggest practical change when AI moves from helping consumers research products to actually acting on their behalf? What new questions around consent, accountability, auditability or consumer control will brands need to solve before agent-led commerce becomes mainstream?
Something I keep coming back to is that, before long, the “customer” a brand is engaging with may not be a person at all; it could be software acting on someone’s behalf.
Yet almost everything in marketing and commerce has been built around the assumption that a human can be persuaded. Our co-founder describes this as commerce entering a phase where software does not just execute tasks but makes decisions, and that perfectly captures the scale of it.
Consent becomes more involved because the customer is no longer approving a single purchase. Instead, they are granting standing permission to act on their behalf, and they will want to set and adjust the limits of that easily.
Right now, accountability is unsettled even in the simple case where an agent buys the wrong thing on a person’s behalf. It gets harder still once the exchange itself is machine-to-machine: a brand’s system negotiating with a shopper’s agent with nobody watching in real time.
At that point, the question isn’t “Did the brand or the customer get it wrong?” It’s “Which of the two systems in that exchange got it wrong?” Almost nobody can answer that yet.
Auditability has to cover that handshake too, not just the moment a person is involved. My sense is that agent-led commerce will move faster than the rules meant to keep it in check, so brands that start building for transparency and customer control now will be the ones that consumers feel comfortable trusting once it becomes the norm.
Editor’s note: This Q&A has been lightly edited for clarity and TNGlobal house style. The substance of the interviewee’s responses has been preserved.
Share your perspective: TNGlobal welcomes contributed insights and expert commentary from across Asia’s technology and innovation ecosystem. Submit a contribution for editorial consideration, or explore more conversations in our TNGlobal INSIDER and TNGlobal Q&A and Interviews archive.
Maya’s Charee Lanza on building an AI-first workforce without losing human judgment [Q&A]