Marketing Data Analyst and Agent Engineer

We are hiring one very specific person. If you have spent the last 1 to 3 years building agents that ingest, process and structure marketing and sales data, and you can demo them, read on.

Seeda builds marketing measurement models for consumer brands. We have spent four years making the company AI-first, function by function. The one thing still too manual is the data supply chain into the model, and that is where the errors live.

The role: Marketing Data Analyst and Agent Engineer. Remote, Sydney overlap, full-time, reporting to our Head of Data Platform.

These are the real problems, unsolved today, across four live accounts:

1. Radio and TV schedules arrive as one file per station and get retyped by hand. 14,516 rows for one client's radio. Offline bookings arrive as 189 signed PDFs.
2. The agency media plan, the invoice and the platform connector disagree on every number. We found a 20 percent gap that turned out to be fees, eight months in.
3. A confectionery brand tracks retailer promotions in its own format. Get the structure wrong and the model confuses promotion effects with media lift.
4. About 40 percent of rows fall through our channel mapping into a human review queue, and the rulings never get written down.
5. A platform feed expired and nobody noticed for a year.
6. A spend line in a closed period moved overnight. We only know because a person remembered yesterday's number.
7. A revenue export of 58,879 rows arrives with the tagging column reading Untagged.
8. Our first LLM extraction invented figures on live client work. The rule now: scripts compute, humans judge exceptions, nobody retypes, and no LLM invents a number.

You have solved a material share of this somewhere else, with the current agentic stack rather than a chain of prompts, for people who depended on the output. You know where it broke and what you changed.

Must have: 1 to 3 years building agentic data pipelines in production for marketing, advertising or sales data. Something you can demo live. Python and SQL you would let us read. Fluent written English. Working day overlapping Sydney hours by at least five.

Big advantage: you did this inside a marketing data company, a media agency data team, an ad platform, a retailer's insights function or a measurement vendor.

To apply, no cover letter. Send two things to [Confidential Information]:

1. A recording, ten minutes or less, of your agents processing real marketing or sales data. What they did, what they got wrong, what you changed.
2. Your CV or a repo link.

If the recording is real, the next step is a paid working session on one of our actual station files and one of our actual promotion calendars.

We know this describes a very small number of people. If you are one of them, I would like to talk this week.