AI systems for Shopify brands doing $3M–$30M

Four jobs. One machine.

Twelve systems, grouped by what they do to the business rather than by which department would buy them. Underneath all four sits the same stack: n8n running on your own infrastructure, MCP tool definitions we wrote for Shopify, your subscription platform, your helpdesk and your 3PL, and a model used only where a rule genuinely cannot do the job. We’re engineers, not a marketing agency — ten years of Shopify development is why we build the plumbing other retention agencies can only advise on.

Book your Revenue Recovery Audit → Run the numbers first →

Recover

Money you already earned, going out the door at the payment layer.

The decline code is already in the webhook payload and almost nobody reads it. We rebuild the ladder so the routing is decided per failure rather than per calendar: which codes get retried, how many times, on which days, and which ones skip the retries entirely and go straight to a one-tap card-update page. Around it sits pre-dunning before the charge that was always going to fail, and an account updater reconciled against the token Recharge, Skio, Smartrr or Stay AI actually bills — not just switched on at the processor. Nothing in this group refunds, credits or cancels on its own. Billing is the one place an automated mistake compounds instead of ending, so a person presses the button.

Retain

Money that repeats — if the second order happens.

Retention fails on timing, and timing is a prediction problem. We model consumption per SKU and per segment — a 60-serving tub at one scoop a day is a two-month reorder window, not the 30 days the app defaulted to — then time the reorder prompt, the education sequence and the billing cadence against that curve instead of against the charge date. The cancel flow reads the reason before it offers anything: “too much product” gets a cadence change, “too expensive” gets a smaller size, and a discount is the last door rather than the only one. Flows carry 41% of email revenue off about 5% of sends (Klaviyo benchmark data, 183,000+ brands). That is the ceiling this group is aimed at.

Run

Money you spend on people doing what a system should do.

The largest of the four, because it is where most of the AI actually is. The runtime is n8n on a VPS in your own hosting account, with MCP tool definitions we write for Shopify Admin, your subscription platform, your helpdesk, your 3PL and your warehouse — each one scoped and logged, so an agent reads what it needs and writes only what it should. On top of that: a support agent that answers “where is my box” from the fulfilment record and can skip the box rather than send a link; a nightly job that compares Shopify against the subscription app against the 3PL and reports the rows that disagree; catalogue enrichment written back through the Admin GraphQL API in batches you can roll back. Anything a rule can do, a rule does — deterministic workflows are cheaper per run, testable, and incapable of inventing an answer.

Reach

Money that never arrives, because the model never named you.

Two arguments about the same money. On the search side, a fixed set of forty to eighty prompts your buyers actually type, run across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews on a schedule and logged three ways — named, not named, or named alongside which competitors — with every cited source captured, because that list is next month’s work plan. Roughly 85% of brand mentions in AI answers originate on pages you do not own (AI-search citation research), which is why most of this is third-party work rather than rewriting your own pages. On the paid side the model is yours: the value attached to a purchase event stops being the first order’s total and starts describing the customer that order represents, so the optimiser is asked to find subscribers rather than first orders.

About those figures

Every figure on the cards above is a category benchmark, not a forecast for your account. The service page behind each card carries the source next to the number, including which ones are vendor-reported rather than independently measured — the distinction matters more in this category than in most, because a vendor’s customers and a vendor’s definition of “resolved” are doing a lot of the work in those numbers. Where nothing honest has been measured yet, the card shows an em dash and stays that way until it can be replaced with something we stand behind.

Ranges are for the build. The audit that scopes it is $1,500–$3,000, fixed, and credited in full against whatever you build afterwards. Most brands need two of the four verbs — the audit says which two, and in what order.

Maturity and saturation on each card are our own read of the 2026 automation landscape rather than a third-party index. We have an obvious interest in that assessment, so treat it as a point of view rather than a finding.

What we don’t do.

Stating the negative is faster than a discovery call, and it saves us both a week. The floor is published for the same reason: $3M+ annual revenue, on Shopify Plus or running a paid subscription platform.

  • Not the AI that writes your captions, or any content tool with an agency attached
  • Not an agent anywhere a wrong answer costs more than a human minute
  • Not a Klaviyo agency that only writes campaigns
  • Not a fit below $3M, or without repeat-purchase behaviour
  • Not a design or theme shop — even though we can build
  • Not a general “ecommerce growth” agency
  • Not the cheapest quote you will get, and not trying to be

Most “AI for ecommerce” is content generation. Ours runs the parts of the business nobody has time for — the retries, the reconciliations, the reorder timing, the tickets.

The second line in that list is the one worth arguing with, so here is the test behind it. It is not whether a model can do the job — it usually can. It is what a mistake costs. Where being wrong means a compliance problem, a claim you cannot make, a refund that should not have gone out or a subscriber who leaves, the expected value of automating it is negative even at a two-percent error rate. So agents draft and people send, refunds and discounts stay with a human, and anything sitting on data nobody has reconciled gets the reconciliation first and the agent second. We say that in the audit rather than sell the agent anyway.

What we are is three things almost nobody combines: a deep Shopify development background, so we build the integration and not just the email; a money-recovery focus, where the result is measured in recovered dollars rather than opens and clicks; and the automation layer itself — self-hosted n8n, custom agents and the MCP tool definitions underneath them, running on infrastructure you keep if we ever part ways.

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.

Book your Revenue Recovery Audit →Or run the numbers yourself →
Fee
$1,500–$3,000, fixed
Duration
Two weeks
Credited
In full, against any build
You supply
Read access + one 45-minute call