AI commerce consultancy
We make it easier to buy.
The best purchases feel like nothing happened. Underneath, data, models and agents decide what to show, what to charge and when to stay quiet. We design the moment, then build the engine behind it.
Agentic commerce, X-rayed
It's 4 p.m. on Sunday and the bake sale is at 8 a.m. I have no flour, no time and one kid who just remembered. I asked my assistant. Twelve minutes later, a box was on its way.
-
She asked for nut-free supplies by 8 a.m. The agent picked this store.
It reads the catalog as data and picks the store whose every answer it can check.
-
Found because the page says nut-free, not just “yummy mix.”
Product data rewritten as clear attributes, so people and agents can both filter on it. It used to say “yummy mix!!”
-
A piping set she hadn't thought of. No coupon she didn't need.
Add-ons picked to fit this cart. A 15% code was held back: she was buying anyway, so it would have cost $15.75 for a sale that was already happening.
-
The price on screen is the price at the door.
Duties and taxes are worked out from official tables before she pays. Nothing to settle when the box arrives.
-
Paid in four seconds, one-handed.
The same easy path for thumbs, keyboards, screen readers and agents.
-
Wrong sprinkles? Refunded before the bake sale.
Every return gets a quiet risk score, and odd patterns get a second look. If a regular's orders start to fade, someone reaches out first.
Six layers. Pick one.
01 · The moments
Revenue is won or lost in five moments
Each one has a part the customer feels and a part they never see. We design both.
- 01
Discover
She asked her agent for nut-free baking supplies by morning, and it picked yours.
An agent-ready catalog: structured product data, feeds in ACP and UCP. - 02
Basket
The piping set she hadn't thought of.
Bundles built per customer, priced in real time. - 03
Checkout
No surprise at the total, and no coupon she didn't need.
Recovery aimed at the people a nudge would change. Landed cost shown before payment. - 04
Returns
Refunded before she got back to the car.
Return-risk scoring, with a threshold your team controls. - 05
Renewal
The account that never quietly left.
Order-rhythm models, and save plays that run inside your CRM.
And every step works for keyboards, screen readers and agents. 07 · Accessibility compliance automation
02 · Point of view
The button was never the hard part
Not long ago you could buy things inside ChatGPT. Within six months, OpenAI stepped back from its own checkout. The agents had arrived. Most of the stores they visited weren't ready for them.
OpenAI and Stripe launched Instant Checkout on an open standard, the Agentic Commerce Protocol. The launch looked like the story. Then OpenAI moved ChatGPT shopping back toward discovery and comparison, saying the first version didn't offer the flexibility it wanted. Analysts pointed at the unglamorous parts: onboarding merchants, keeping product data accurate, handling baskets with more than one item, carrying loyalty programs across.
All of that is engine work: clean data, honest pricing, operations that hold up when a machine is the customer. And the traffic kept coming. Adobe measured visits from generative AI tools to US retail sites up 693% across a holiday season, and those visitors converted better than everyone else. The agents are here. The open question is which stores they can buy from.
Growth in traffic from generative AI tools to US retail sites over a holiday season (Adobe).
How much better AI-referred visitors converted than other traffic in the same season (Adobe).
From the launch of Instant Checkout in ChatGPT to OpenAI's shift back toward discovery (Stripe, Retail Gazette).
03 · Play
Hold the coupon.
Six shoppers just left their carts. You have one 15% coupon for each. Give it and you might hand over margin for nothing. Hold it and you might lose the sale. Beat the engine.
Swipe right to give. Swipe left to hold. Arrow keys work too.
Spot the shoppers a coupon would actually change.
Every shopper looks the same from the outside. The engine can see who a coupon would change. You get one look at its read per shopper, if you ask.
04 · Field guides
Eight problems. Eight fixes. Each one measured.
Every playbook starts the moment a shopper hesitates and ends with a number you can check. This is how we think, written down.
playbooks
moments that decide the sale
rule: nothing scales without a holdout
Cross-sell & bundle generation
Product graph · offer generator · per-customer pricing
Cart abandonment recovery
Uplift model · reason inference · message per person
Selling to shopping agents
Structured catalog · agent feeds · protocol fit
Catalog enrichment
Attribute extraction · rewrite and translate · human review
B2B contract churn prediction
Order-rhythm model · revenue-at-risk · CRM save plays
Return fraud detection
Return-risk score · tunable threshold · case review
Accessibility compliance automation
Scan · remediate · regression monitoring
Cross-border tariff & duty automation
Classification assist · rules engine · landed cost at checkout
Principle
Human is the future of AI.
The engines we build do the counting, scoring and drafting. People decide what the business stands for, set the limits, and handle the cases that matter most.
People set the rules
Every play runs on a threshold your team controls, and your team can change it.
People take the hard cases
Return holds, large orders and upset customers go to a person, with the evidence in front of them.
People check the result
Nothing scales until a holdout shows it works and someone accountable signs off.
05 · Start here
Start with one moment.
Most clients begin with a diagnostic. Fixed scope, fixed fee, and an answer you can defend, even when the answer is no.
S1 · AI Commerce Diagnostic
Three plays worth funding, with the numbers behind them
What you get
A ranked shortlist of AI plays for your funnel. Each comes with a business case on your numbers, the data it needs and a 90-day plan.
How fast
Three to four weeks from kickoff to readout.
What it takes from you
Access to your data and a few hours of your team's time each week.
If the numbers say no
We say so. Nothing gets scaled until a holdout shows it works.
Under the hood
A funnel model built on your own data. Leaks sized stage by stage. Use cases scored on value, feasibility, risk and time to proof. Build or buy, play by play.
06 · Library
Read the thinking.
Articles, white papers and free tools on AI where revenue is won or lost. New pieces also go out on LinkedIn.
- Article
- Article
- Article
- Article
-
Checklist
Agent-readiness checklist
Twenty yes-or-no questions to test whether AI shopping agents can find, compare and buy your products. Score your catalog in an afternoon.
-
Template
AI commerce opportunity map
A one-page template for mapping AI use cases across your funnel, from discovery to renewal, and scoring them on value, feasibility, risk and time to proof.
Next step
Bring one number you want to move.
Thirty minutes with the founder. You'll leave with a straight answer on whether AI is the right tool for it.
AI is… the Future.