ProductPilot builds the systems that run a business without anyone driving them — Discord and Slack bots, Google Sheets and GCP pipelines, n8n workflows, and LLM agent fleets that research, write and act. We built one for ourselves to see whether it held up. In seven months it produced over $80,000 in attributed sales.
Source: Amazon Attribution (US + CA), 1 Mar – 6 Oct 2026. Self-reported from a media property we own and operate — built and run entirely by one of our agent fleets.
The marketing fleet ran across 35 brands in home improvement, water treatment, audio, wellness and baby — products it promoted and sold through Amazon Attribution, without a person choosing them one by one.
Brand names and logos are the property of their respective owners. Listed to indicate products promoted via Amazon Attribution; no endorsement, sponsorship or partnership is implied. ProductPilot participates in the Amazon Associates programme.
“We rebuild the same report every Monday.”
Someone exports a CSV, pastes it into a sheet, fixes the formulas and posts a screenshot. Every week, forever. That is a pipeline and a bot, not a job.
“Our data lives in six different places.”
Ad platform here, orders there, spreadsheets everywhere, and no single view. We wire the sources into one warehouse and put the answer where your team already works — Slack, Discord, or a sheet.
“We bought the tools. Nobody runs them.”
Software gives you capability, not capacity. An agent fleet is the operator — it runs the workflow continuously and only asks for you at the approval gate.
Every workflow we build has the same shape, whether it is posting content, reconciling a warehouse or answering in Discord. Five stages pass work down the line, and nothing irreversible happens until someone approves it — the difference between automation you can defend and automation you find out about later. Release the queue below.
Before selling agent fleets to anyone else, we pointed one at a media property we own. It researched the category, drafted the content, held every draft for approval, published, and fed performance back into the next cycle — running on the same Sheets and GCP plumbing we build for clients. Month one returned $2,261. By month three it was $20,881, and it held above $15,000 a month through August.
Measured events from Amazon Attribution across the seven months the workflow ran.
Pricing indicative and shown for illustration. Final scope follows the Phase 01 audit.
Every one researched, drafted and published by the system described above — and approved by a person before it went live.
An automation studio. We build workflows that run without an operator — bots, pipelines, agent fleets. Marketing happens to be the domain where we proved it hardest, because attributed revenue is an unforgiving way to measure whether a system actually works. The same engineering builds your Discord bot or your warehouse.
Ours. We built and own the property that generated them, which means we carried the risk and can show you the raw Amazon Attribution export rather than an anonymised slide. The brands listed above are products we promoted through the Amazon Associates programme, not clients who paid us.
Specialised agents that hand work to each other rather than one model answering one prompt: one gathers, one processes, one routes to a human for approval, one executes, one measures and feeds the result back into the first. It runs continuously without a person driving it, but it cannot take an irreversible action without a person approving it.
We do, by default. Most teams keep it that way; some move it in-house after a quarter once they have seen the output quality. Either is fine, and moving it is a config change, not a rebuild.
Usually, yes. We build on Google Cloud, n8n, Slack, Discord, Notion, Sheets and whatever APIs your stack exposes. If you already pay for a tool that does the job, we will wire it up rather than rebuild it — and tell you when a tool is not worth keeping.
Yes. Code, infrastructure and documentation are yours. We can host and run it, or deploy into your own cloud account and hand over the keys. There is no proprietary layer you have to keep paying us for.
An hour for the Phase 01 audit, access to the systems the workflow touches, and one named person to approve the blueprint. After that, a weekly 30-minute review.
A single workflow is typically live in three weeks. Larger fleets run four to six. We put every build into shadow mode first — it runs and reports but takes no action — so you can check its judgement before it has any.
Tell us the thing your team redoes every week. We will tell you whether it is worth automating, roughly what it costs, and honestly if it is not.
contact@productpilot.es