News + Thought
AI + The Purchasing Journey
For years, we’ve talked about faster and more efficient purchasing journeys. In 2026, AI is accelerating this change at an unprecedented pace - so what are the implications?
If you’re old enough to remember the advent of ecommerce, you’ll recall that the internet cut consumer decision-making from hours to minutes. The web made comparison easier. Mobile made buying constant.
But now - with AI - what’s happening now is different not merely by degree, but in kind: AI purchasing agents are starting to remove the consumer from large parts of the process altogether.
This probably sounds baffling at first glance, but the key difference is that people are no longer the ones who are actually doing most of the deciding. The result is that what once took around an hour in the pre-internet era can now collapse into something like 90 seconds.
In reality, the modern consumer journey is already heavily front-loaded. If you asked me to estimate for you, I’d probably say that around 95% of the decision is made before a purchase happens.
In the olden days, that used to mean people were spending time researching, comparing options, reading reviews, and slowly narrowing choices. Increasingly, it means something else: the decision is being formed in advance by systems that are filtering, ranking, and recommending long before a human ever reaches a checkout page.
The reason why this is such a crucial change to acknowledge is because it changes what competition looks like. In a world where AI agents are shaping decisions, visibility isn’t just about being ranked on a search engine or having a strong ad campaign. It’s about being understood, structured, and trusted by the systems doing the filtering.
Personalisation stops being a marketing tactic and becomes the primary moat. If an AI doesn’t surface you, you literally no longer exist in the decision set.
How do you ensure you're discovered on AI platforms?
Well, step one is understanding that AI agents don’t experience the buying journey the way humans do. The traditional funnel - discover, evaluate, decide, purchase - starts to collapse when you realise that an agent is handling all four stages simultaneously (and instantly).
Take discovery, for example: it stops being about keyword search, scrolling, or ad exposure. Instead, agents continuously scan structured product data alongside unstructured content like reviews, videos, and user-generated posts. They don’t “browse” per se, they filter constantly in the background, surfacing only items that match specific constraints and inferred preferences.
Evaluation also changes dramatically. Where consumers once opened multiple tabs, compared specs manually, and tried to interpret conflicting reviews, agents can instantly compare attributes across sellers, identify inconsistencies, flag counterfeit or low-trust listings, and cluster reviews by sentiment and theme.
What used to be a cognitive workload for the consumer becomes a routine processing task for the system
Finally, decision-making shifts even further. Instead of humans weighing up discounts, timing purchases around promotions, or trying to optimise for loyalty points and perceived deals, AI agents can simulate outcomes based on personal constraints.
Budget, sustainability preferences, warranty conditions, and even behavioural history can be factored into an optimal recommendation in seconds. To an outsider this still might look like “choice”, but in reality it’s often just the output of a constrained optimisation process.
For AI tools that support integrated purchases, the purchase itself becomes almost invisible. Instead of filling out forms, navigating checkout flows, and dealing with shipping uncertainty, agents complete transactions directly, using pre-approved payment methods and handling post-purchase tracking automatically. This reduces any kind of present friction to basically zero.
Sure, this might sound like it’s a long way off to some, but the infrastructure is already starting to form. New protocols are emerging that allow agents to operate across the internet in a far more unified way.
The most famous example is probably Anthropic’s Model Context Protocol (MCP), which enables agents to access real-time structured and unstructured data across the web more seamlessly, while OpenAI’s Agentic Commerce Protocol (ACP) is designed to support secure, end-to-end transactions initiated by agents.
Put in simple English, MCP helps agents see and interpret the world, while ACP helps them act within it. Together, they form the backbone of an “agentic” commerce layer where discovery and transaction are no longer separate steps.
The scale of what this enables is significant. Some estimates suggest AI agents could facilitate more than $8 trillion in online consumption by 2030. Today, they account for a small fraction of transactions (around 2% of online spend), but I’ve seen predictions that it could rise to roughly 25% within the next five years as consumers increasingly delegate decisions to intelligent systems. And to be honest, I find it hard to disagree.
AI search is already taking share from traditional search engines, and it’s been suggested that AI-driven search could grow from around 10% of global search traffic today to as much as 65% by 2030. As that happens, advertising naturally follows. AI-driven ads are expected to grow rapidly, with monetisation typically lagging adoption by around two years as platforms figure out how to commercialise the behaviour shift.
When you put all of this together, the broader implication becomes clearer: AI agents are not just optimising parts of the internet, they are beginning to sit between consumers and the internet itself.
Consumer behaviour is moving from active searching and comparing to passive delegation and acceptance of AI-driven outcomes.
This all sounds jolly and fresh and exciting, but within it hides a potentially uncomfortable truth. For a long time - certainly as long as I’ve been involved in the sphere - digital strategy has been built around persuading people.
Whether through messaging, branding, and increasingly sophisticated marketing funnels, the core concept remained the same.
But in an agent-driven world, the first audience is no longer human. It is algorithmic. Products need to be structured in a way that can be understood by machines. Reviews need to be interpretable at scale. Trust signals need to be machine-readable, not just emotionally persuasive. Pricing, availability, and differentiation need to make sense in a system where comparison is instantaneous and exhaustive.
I am slightly treading on the toes of the recent podcast that I've spoken on with Skywire, but the crux of the matter is that the actual question that we’re already hearing our clients ask is changing.
It’s no longer just “why would a customer choose us?”; it is now “would an AI agent choose us on behalf of the customer?”
And if the answer to that is no, the customer may never even reach you in the first place.
