AI systems for Shopify brands doing $3M–$30M
We qualify on complexity, not category.
The AI growth company for ecommerce. We don’t start by asking what is in the box. We ask whether the money arrives on a schedule, how many systems have to agree before an order ships, what you already pay Shopify, and how much of the operation still runs out of a shared inbox. Those four questions decide whether there is anything for a system to read — a decline pattern to route against, a consumption curve to model, a queue of tickets that are the same six questions in different words.
Four situations, one set of mechanics.
Most brands we work with are in two or three of these at once — a subscription brand on Plus that also sells on Amazon is the common shape, not the exception. Start with whichever one you recognised first. Each page names the signal that puts you in it, and what we would build first.
Subscription brands
You bill on a schedule, so every failure repeats monthly. Signal Recharge, Skio, Smartrr, Stay AI or Loop installed See the detail →Shopify Plus brands
You pay for the enterprise tier. Use the parts you're paying for. Signal Shopify Plus, from $2,300/mo See the detail →Multi-channel brands
DTC plus Amazon, TikTok Shop or retail — and nothing reconciles. Signal Selling on two or more channels See the detail →Brands scaling past $3M
The tooling that got you here stopped fitting somewhere around $3M. Signal First ops or retention hire posted See the detail →Why situation beats category.
A generalist agency sees a subscription as a billing setting. What actually predicts whether this work applies is not what you sell — it is whether the money arrives on a schedule, how many systems have to agree before an order ships, and whether there is enough margin in the business to pay for a fix and still be ahead.
Two brands selling completely unrelated products — a daily supplement in a 30-day bottle and a household refill people reorder every six weeks — have the same retry ladder problem, the same month-three cliff where the result is slow to show, and the same cancel flow with no pause option. Two brands inside one category, one on Shopify Basic doing $900k and one on Plus doing $18M with a marketplace channel attached, have almost nothing in common operationally. Category is the wrong axis.
The second reason is compounding, and it is sharper now than it used to be. Every teardown, benchmark and integration we publish makes the next build faster and better argued — and so does every system. The decline-code routing written for one Recharge account is the same routing on the next one. The MCP tool definitions that let an agent read a Shopify order, a Smartrr subscription record and a 3PL tracking event are the same definitions in the next build, with the credentials swapped. That only works when the operating model repeats. It does: a Recharge retry ladder is a Recharge retry ladder whatever is in the box.
What does not transfer is the judgement — which decline codes are worth retrying in your account, what your cancel reasons actually mean, which intents an agent must never answer for you. That part is built per brand, in writing, and it is most of what the audit is for.
The profile.
Read this as a list of inputs rather than a list of credentials: each row is something a system needs in front of it before it can do anything useful. If most of it describes you, the audit will find something. If none of it does, we say so on the first call rather than sell you a build.
| Attribute | What we look for |
|---|---|
| Revenue | $3M–$30M annual |
| Platform | Shopify or Shopify Plus |
| Model | Subscription, or high-repurchase replenishment |
| Stack signals | A paid subscription platform — Recharge, Skio, Smartrr, Stay AI or Loop — plus Klaviyo, a paid helpdesk and a 3PL |
| Complexity | Multi-channel: DTC plus Amazon, TikTok Shop or retail. 100+ SKUs. Fulfilment outsourced. |
| Team | 5–40 people, often with a first Head of Operations or Head of Retention just posted |
| Buyer | Founder, Head of Growth, Head of Retention or Head of Ops |
| What it sounds like | “We run eleven tools and I still can’t tell you profit by channel.” |
Two or three rows off is normal — nobody matches a profile exactly, and the stack signals matter more than the round numbers. The revenue line is the one that actually decides it, and it is published for a reason: below it, this work costs more than it returns.
Who this is not for.
We publish the floor rather than make you find it on a call — $3M+ annual revenue, on Shopify Plus or running a paid subscription platform. It is there so a bad fit can rule itself out in ten seconds, and so the people above it know the conversation is worth having.
- Under about $2M in revenue At the margins this category runs on, a five-figure build is a large share of a year’s profit. That is the wrong trade for you, and we would rather say so than take the money.
- No repeat-purchase behaviour If people buy once and never come back by design, there is no retention curve to work on and most of what we build has nothing to attach to. It is also the case that a system with nothing repeating in front of it has nothing to learn from — no consumption curve to model, no decline pattern to route against, no cohort to read.
- Wanting an agent on top of data nobody trusts An agent reading a catalogue where a third of the fields are stale will produce confident, fluent, wrong answers considerably faster than a person could — and the same is true of a 3PL feed nobody reconciles and a subscription record that disagrees with Shopify. We will sell you the reconciliation first and the agent second. If the reconciliation is not the job you want done, we are the wrong firm.
- Looking for the cheapest freelancer Our published ranges start at $2,000 and run past $25,000. If price is the deciding variable, somebody else is a better fit — plainly meant, not a negotiating position.
- Wanting a redesign as the main deliverable We build infrastructure: billing, flows, reporting, automation, the agents on top of them. A brand or website redesign is a different job and a different agency.
And what we are not, at any size
- Not an “AI for ecommerce” content tool — what we generate is catalogue data and drafted replies a person approves, not a campaign calendar
- Not a Klaviyo agency that only writes campaigns
- Not a fit for brands without repeat-purchase behaviour
- Not a design or theme shop, even though we can build one
- Not a general “ecommerce growth” agency
We don’t do brand design, and we don’t take work below the floor. Saying so up front costs us a handful of enquiries a month and saves everybody a wasted call.
There is one more boundary worth publishing here rather than discovering later. Refunds, credits, discounts, payment-method changes and cancellations that trigger a refund are prepared by a system and pressed by a person, in every engagement, on every plan. An agent may assemble the whole thing — the order, the policy, the history, the draft — but money leaving your account is a human action. That is a deliberate boundary rather than a technical limit, and it does not move for a bigger budget.
What we build for them
Find out what you’re losing.
Before you commit to anything, we tell you exactly what you’re losing and what it costs to stop it. Two weeks. Fixed fee. Credited in full against any build you go ahead with.
- Fee
- $1,500–$3,000, fixed
- Duration
- Two weeks
- Credited
- In full, against any build
- You supply
- Read access + one 45-minute call