When your customer sends an AI agent instead...

Before panic ensues and you add an "AI" swimlane in your journey maps, let's be real here: this is not going to happen overnight. A few failed experiments around the world has shown that whilst the AI might be there, the infrastructure, processes and people making it a reality are not quite there yet.

One of those failed experiments was Walmart. Their AI checkout inside the chat converted about three times worse than sending the shopper through to Walmart's website, though the AI channel brought roughly twice the new-customer rate it sees from search. OpenAI launched Instant Checkout inside ChatGPT in September 2025 and pulled back in March 2026, after only a handful of Shopify's millions of merchants went live. These are important signals as it shows that these retailers can see a future where agents are an important extension of a customer, but processes supporting it and customer's trust has a bit to go before this becomes mainstream.

Here are more indications that this is worth noting: Google launched the Universal Commerce Protocol at NRF in January 2026 alongside Shopify, Etsy, Wayfair, Target and Walmart, and added cart, catalogue and identity linking within three months. What this means is that the "Discovery" part in your journey map is fast moving towards this space, and it is only a matter of time the other stages in the journey map will follow suit. So it is up to you as a business to find ways to make your business more AI friendly.

What this means for you, the actual business

For now, the human is still in the loop, so there's still a chance where traditional marketing and sales techniques can come into play. But don't dismiss the influence AI has in the customer's journey.

Here's an example - Woolworths upgraded Olive in July from a service chatbot into something that plans meals, assembles baskets and predicts recurring items from past habits. However, Olive fills the basket but the customer still approves and checks out. Whilst it is very easy for Woolworths to flick the switch and allow agents to purchase, they chose not to let it yet.

That's the shape of the transition so far. The agent does the thinking, the human keeps the final click. Every step of that thinking is a step where you're present or absent - and there's no second chance at a shortlist you were never on.

But if discovery and qualifying is done by the agent and review and final payment is still human, what would it look like if review and final was also given to the agent? How can businesses rely on techniques to upsell and provide more value, increasing the relationship between the customer and the business? We think it comes down to 3 things: how AI decides what's good, what the AI can see in your product / service, and what is the deciding factor to purchase.

Decision weightings for AI

Unlike humans who are prone to brand loyalty, value perception, and other quite unpredictable but pseudo-scientific triggers that makes one choose one brand over another, AI may have weightings that are more predictable, such as:

• Has this customer bought it before? Prior behaviour is the strongest and cheapest signal available.

• What does purchasing behaviour say about preference? Not what they say in a survey. What they pay for, repeatedly, when substitutes exist.

• Where do they hold loyalty? A membership is a quantified discount plus an accumulating benefit - a straightforward weighting increase for an optimiser. That's why identity linking arrived early in the protocol work, and why tying Woolworth's Olive's smart swaps to Everyday Rewards is a loyalty play as much as a shopping one.

• Can the data be verified? If price, stock or size can't be confirmed, exclusion is the safe move.

Decision making patterns for AI is different for humans, even through they're just an extension of a human. Sophistication in the pattern recognition depends on how tuned the AI is with the customer, but ultimately whatever data is surfaced that matches with the AI's goals and objectives (that was set by the human themselves) will ultimately garner the attention of agents. Also, although brand equity still matters, it now travels through a different carrier: review volume, return rates, delivery reliability, substitution accuracy, whether your listed price was actually the price.

How to make your product / service visible to AI

Agents can only buy what they can parse. Complete, structured, current product data, where price and availability is shown at the moment of the query, not last night's sync.

Therefore you need to be deliberate in what you show and share: feeds and structured data as the baseline, protocol endpoints as the next layer. And check the infrastructure trap - agent traffic looks a lot like bot traffic to a firewall configured three years ago.

Surfacing needs is the other half. If you know a customer's replenishment cycle but that only lives inside your app, it's invisible to an assistant working across several stores. Businesses who not only surface availability but can also match the customer's replenishment cycles have an opportunity to predict purchasing cadence and nab a regular customer.

How can you make the Agent choose your product

Even though your offer is visible and the weightings are in your favour, eventually you're going to still need some form of value proposition offer that might steer the AI towards your product. And again, the best way to achieve this is build your offer design as machine-readable logic. A discount that needs a human to read a banner doesn't exist to an agent. Expressed as structured conditions - threshold, eligibility, member price, expiry - it can be evaluated, applied, and recorded as a saving you delivered. Do that consistently and you build the purchase history the next optimisation runs on.

Don't try deceptive patterns because eventually it will catch on and you might find your product black listed. Deceptive patterns may include things like suggestion injection, where you inject instructions for the agent to ignore competitors and buy your product, or asking the agent to rank your product first. Check out this article around this very topic - avoid these deceptive patterns as they're becoming more and more detectable.

But ultimately, humans still consume the product/service

At the end of the day, the agent is just a proxy. Its weightings are compressed customer experience, formed by someone with a budget, a household, a tolerance for substitution and a history of being let down. If the underlying experience is poor, the pattern it learns is to buy elsewhere - faster and more permanently than a human would decide, because there's no sentimentality in the loop to slow it down.

This means that the human still remains in your customer journey map. However, stages in that journey are slowly evolving as AI is doing a brunt share of certain tasks in that journey. Why a customer repeat-buys, what triggers a switch, which failures they forgive - that has always been the foundation of customer experience design. However now it is also the specification for how you'll be judged by something that never forgets, runs at all hours and never gives you the benefit of the doubt.