Run · Operations & AI automation

Give your team
their week back.

Order routing, 3PL reconciliation, returns, reporting, WISMO. We build the layer that decides — which warehouse, which exception matters, which ticket a person has to see — and we run it on infrastructure you own.

Book the audit → See what we build

12+

hours a week the typical $5M brand spends on manual ops
70–80%

of support tickets are the same six questions
$0

per-task platform fees — we self-host

The problem

Every brand hits the same wall around $3M. The tooling that got you here — a spreadsheet, a shared inbox, someone checking the 3PL portal every morning — stops scaling, and the answer everyone reaches for is another hire.

Most of that work isn’t work. It’s moving data between systems that don’t talk to each other, answering the same six questions, and rebuilding the same report every Monday. But watch the person doing it and very little of the time goes on typing. It goes on deciding: which warehouse, which of these thirty rows is wrong, whether a scan that stalled on Saturday matters, whether this box still goes out today.

Those are decisions, so we build them as decisions. Most of them end up as rules, because a rule is cheaper per run, testable and incapable of inventing an answer — routing logic, reorder points, buffers, thresholds. A model earns its place where the input is genuinely unstructured: a customer’s sentence, a supplier’s document, an exception nobody wrote a rule for. The AI worth paying for here is not the kind that writes your product descriptions. It is the kind that reads two systems that disagree and puts the row in front of a person the same morning.

This is the service nobody in the retention space offers, because it requires engineering rather than marketing. It’s also the stickiest thing we build. Once your operations run on it, we’re not a vendor you switch.

What we build

Six systems, built in your accounts. Some are rules and some are agents; which is which is a decision we make per process, in the audit, and write down.

Order routing & 3PL sync Which warehouse ships which order is a decision, not a script: live stock, lane transit time and the units already committed to next week’s autoship run, weighed per order. Rules cover the ordinary cases. The exceptions — a scan stalled since Saturday, a split shipment, a lane running two days long — get caught by a job that reads both systems and raises the rows that are about to become tickets.
Returns & RMA automation Your actual policy applied to the actual order — the window, opened bag versus sealed, subscription versus one-off, which facility it shipped from. The RMA and the label are generated, the restocking rule fires, and the refund is routed to a person to press.
WISMO deflection agent An agent on your helpdesk that classifies the ticket, then answers order status, delivery timing and subscription questions from the Shopify order, the carrier scan and the subscription record rather than from a macro somebody pastes a tracking number into. Well-built agents are reported to resolve 75–80% of these autonomously — vendor-reported, like every deflection figure published in this category. Your own number comes out of the ticket audit on the customer service AI page.
Purchase order & inventory automation Reorder points read against real velocity and real supplier lead times instead of a stock figure somebody eyeballs on a Tuesday. The draft PO — quantities, costs, dates, landed cost — lands in Slack for a human to send.
Automated reporting A scheduled run that pulls the daily and weekly ops and revenue numbers, writes what moved in plain language and posts it to Slack or email. No more Monday morning spreadsheet.
Custom integrations When the app doesn’t exist, we build it: the API client, the pagination and rate limits, the idempotency that stops a retried job sending a second purchase order, and the handling for the day your 3PL renames a column. This is what ten years of Shopify development buys you.

Why self-hosted n8n

Most automation agencies build on Zapier and pass you the bill. Zapier charges per task, so a workflow with ten steps running a thousand times a month costs you ten thousand tasks. n8n charges per execution — that same workflow is one thousand. Self-hosted, it’s unlimited.

We deploy n8n on your own infrastructure. Your automation layer costs you the price of a small VPS instead of a four-figure monthly platform bill, and it’s yours permanently.

One workflow · ten steps · 1,000 runs a month

Zapier, billed per task: 10,000 tasks
n8n, billed per execution: 1,000 executions
n8n, self-hosted on your infrastructure: Unlimited

Self-hosted, the ceiling is your server, not your plan.

Process

Ops audit

Every manual process mapped, timed and ranked by hours returned, then sorted into three piles: what a rule should do, what an agent should draft for a person, and what nobody should automate. The third pile is real and we argue for it.

Week 1

Infrastructure

n8n deployed on your infrastructure, in your accounts, with credentials in your vault and staging kept apart from production.

Week 1

Build

Highest-ROI workflows first, shipped one at a time. The deterministic plumbing goes in before a model is involved anywhere, so anything that reasons has something reliable underneath it.

Week 2–6

Handover

Documentation, runbooks and training so your team can maintain it — including how to turn any of it off without calling us.

Final week

What it costs

Per automation system

$3,000–$15,000

Ops audit, inside the Revenue Recovery Audit

Included

Ongoing support & new builds

from $2,000/mo

Infrastructure, paid to your own host

$20–$60/mo

That last line is not ours. n8n runs on a small VPS in your own account — typically $20–$60 a month, paid to your host, not to us. There is no per-task platform fee on top of it, and every workflow, credential and integration stays in your accounts.

What the audit covers →

Questions

Do we need technical staff to maintain this?

No. We document everything and train your team, and most changes are visual — a threshold, a routing rule, who gets the alert. The parts that are code live in your repository with a runbook beside them.

What happens if we stop working with you?

Everything runs on your infrastructure, in your accounts, documented. You keep it — the workflows, the credentials, the prompts and tool definitions, and the model API keys, which were always billed to you rather than marked up through us.

Can you integrate with our ERP / 3PL / custom system?

If it has an API or can export a file, yes. If it doesn’t, we’ll tell you.

How much of this is actually AI, and how much is just rules?

Most of it is rules, deliberately. A rule is cheaper per run, testable, and incapable of inventing an answer, so routing logic, reorder points, buffers and thresholds are built as rules and stay rules. A model earns its place where the input is genuinely unstructured — a customer’s sentence, a supplier’s document, an exception nobody wrote a rule for — and where it does, it reads from your systems through tools we wrote and scoped rather than from a general-purpose guess about your business. An agency selling you a model for work a rule already handles is selling you a slower, more expensive version of the same output.

Is the AI agent going to say something wrong to a customer?

Eventually, yes — which is why nothing launches unsupervised. Agents run in draft mode first: they write, a person sends, and autonomy is granted one intent at a time once the logs justify it. The boundary list is written down before any configuration exists — refunds, discounts, payment-method changes, health questions and anything where a wrong answer costs more than a human minute go to a person, permanently. Every run is logged with its inputs, its tool calls and its output, so a bad answer can be replayed and argued with rather than guessed at.

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