AI for E-commerce
Businesses
Larger ecommerce operations have enough product, customer and service data to support useful AI, but the work crosses several platforms and owners. We start with a measurable commercial or operational problem and connect the systems behind it.
70.2%
CARTS ABANDONED (BAYMARD)
15×
AI-DRIVEN ORDERS, 2025 (SHOPIFY)
ACP & UCP
TWO AGENT-CHECKOUT STANDARDS, ONE RACE
AI assistants are becoming a sales channel
For twenty years, ecommerce meant a human typing into Google, landing on a product page, clicking buy. Most of it still is. Won't be for long.
ChatGPT spent autumn 2025 selling inside the chat via the Agentic Commerce Protocol that OpenAI built with Stripe. OpenAI pulled it in March to fix pricing, inventory sync and sales tax, but the protocol's open and the next attempt is already in flight. At NRF in January, Google, Shopify, Etsy, Wayfair, Target and Walmart fired back with the Universal Commerce Protocol, backed by Visa, Mastercard, Amex, Stripe and Adyen. Copilot and Gemini are wiring up too.
If your product data, pricing and policies aren't readable to an agent, you stop being on the shelf.
YESTERDAY'S STORE
- Built for Google blue links
- Product copy written once, then ignored
- Support inbox of "where's my order"
- Email blasts to the whole list
- A merchandiser hand-pinning hero rows
AGENT-READY STORE
- Feed an agent can shop on your behalf
- Descriptions rewritten as the catalogue changes
- Tier-one support handled before a human sees it
- Lifecycle emails that read like you wrote them
- Merchandising that re-ranks itself overnight
Store operations worth automating
Skip the "AI strategy" deck. Most online retailers get most of the upside from these five. Pick the one that's hurting most this month, build it, move on.
Product data
Titles, descriptions, attributes, images and alt text rewritten from your specs and reviews. Cleaned for Google Merchant Center, Meta, TikTok and the agent feeds.
Search & merchandising
On-site search that reads intent, collection pages that re-rank against today's stock and margins, and a quiet pricing assistant that flags the SKUs nobody's watching.
Checkout & agents
Total-cost-up-front, address autofill, smart shipping options, and the feeds and policies you need to be sellable inside ChatGPT, Copilot and Google AI Mode.
Service & returns
WISMO, sizing, exchanges, refunds and warranty triage handled by an agent that's read your policy, with the high-stakes calls escalated cleanly to a human.
Retention
Lifecycle email and SMS written per customer, not per segment. Win-backs that mention the dress they bought. A second AI watches the open rates and rewrites the bad ones overnight.
How we start
We review the catalogue, customer journey, connected services and data available through the commerce platform, then scope the workflow with the clearest measurable case.
The storefront keeps running while a contained workflow is tested against existing product, customer and service systems.
BOOK A STORE AUDITStore audit
Read access to your store, analytics, ESP, helpdesk and ad accounts. One-page report: where the money's leaking, where AI pays back fastest, what's already DMCC-compliant and what isn't.
Clean the catalogue
A product data loop that rewrites titles, descriptions and structured attributes from your specs, reviews and supplier sheets. The feed Google, Meta and the AI agents see stops being the limiter.
Launch the first revenue loop
On-site search, lifecycle email, or a service agent. We pick the one with the biggest known gap, launch it, plug it into your existing tools, and prove the lift before scoping the next.
Put the learning loop on top
A second AI watches each loop's outputs. When the recommender misses, the email flops or the support agent escalates badly, it works out why and updates the prompt, knowledge or rule overnight. By morning, that mistake doesn't repeat.
Recent changes in AI commerce
Three changes from the last twelve months that decide what your store looks like for the next five.
ChatGPT learned to check out, then paused.
OpenAI launched Instant Checkout with Etsy and a million-odd Shopify merchants including Glossier, SKIMS and Spanx, running on the Agentic Commerce Protocol built with Stripe. They pulled the feature in March 2026 over pricing, inventory and sales-tax accuracy. The protocol is still open. The next attempt is coming.
Google fired back.
At NRF, Google and Shopify launched the rival Universal Commerce Protocol with Etsy, Wayfair, Target and Walmart as co-developers, and Visa, Mastercard, Amex, Stripe and Adyen on the payment side. The agent web is now a two-horse race, and your catalogue is the asset.
DMCC Act in force.
The CMA can now fine up to 10% of global turnover for drip pricing, fake reviews and the dark patterns that used to be normal. The first eight investigations opened on 18 November 2025, targeting StubHub, viagogo, Wayfair and others. Subscription rules land later this year.
Sources: OpenAI, Stripe, Shopify, Google, CMA, Taylor Wessing, Cooley.
When this is worth discussing
We work best when there is a real operating problem, enough volume to measure and people from the affected teams who can make decisions.
Usually a good fit
- An established UK business, usually with annual revenue above £10m
- A repeated process with a known cost, delay, error rate or capacity problem
- A senior sponsor and a day-to-day owner who understand the work
- Access to the relevant staff, systems, sample records and security requirements
We may point you elsewhere
- A standard product already covers the process well
- The requirement is a one-off small build with no wider operating case
- There is no owner or access to the people and data needed to test the result
- The plan relies on AI making high-impact decisions with nobody responsible for review
Questions the buying team will ask
Isn't this just a Shopify app?
Sometimes. Native platform features should be tested first when they cover the requirement and data terms. Custom work is useful where the workflow crosses products, needs your own rules or must update another operating system.
We're on WooCommerce / BigCommerce / a bespoke stack. Still works?
Potentially. Shopify, WooCommerce, BigCommerce and custom stores offer different APIs and extension points. We review the catalogue quality, policies, integration limits and connected systems before confirming the build route.
Do we have to do all five loops?
No. Pick the one that hurts most. For most brands that's product data or service. Once one loop's paying for itself, the next is an easy call. We don't bundle.
Will the agent stuff drive sales for us?
Buyer behaviour through AI assistants is still developing, so we would not base the investment case on forecast channel volume. Clean product data, current prices, stock, returns terms and structured policies are still useful across search, marketplaces and assistant-led discovery.
What about returns, refunds, and the angry customer?
The service loop reads your policy, your order data and the carrier's tracking. It handles WISMO, sizing, exchanges and standard refunds inside the policy. Anything outside, anything where someone's angry, anything over a value threshold goes to a human with the context attached. You set the rules. We log every decision.
How do you keep us on the right side of the DMCC Act?
Total price up front, no surprise fees at checkout, no fake or incentivised reviews, no hidden subscription traps. The audit flags where you're exposed. The loops we build respect it by default. The CMA can fine 10% of global turnover, which is reason enough not to wing it.
How long, how much?
We define the storefront, data, platforms and commercial measure in scope before proposing a first phase. Any rollout depends on the test result and the operating team's ability to own it.
Who owns the prompts, data and code?
You do. That includes the source, prompts, evaluations and data. We can host it or hand it to your team. The AI runs on your provider accounts, so the bill and logs stay with you.
Talk to us about your store operations
Tell us the storefront, trading systems and measure the team wants to improve. We will identify a use case with enough data and volume to test, plus the customer and brand controls it needs.