Poldermarketing
How can a startup build a repeatable marketing process with a small team?
A small team makes marketing repeatable by repeating one written loop: a customer insight, one bet, a prediction, the work, and a keep, change, or stop. Freeze one buyer and one bottleneck first. Document a motion only after it works twice. Automate a handoff someone has already done by hand. The test is that next Monday runs from the file.
Monday has to run from a document
A marketing process is repeatable when someone who missed the meeting can ship next week from a written brief, and the review ends in keep, change, or stop. A calendar, a tool stack, or a founder who remembers the steps is not that process.
Three files are the test: who is qualified, this week’s bet with its assumption and prediction, and last week’s decision. Matt Lerner, on First Round Review, treats the search as the job. He argues early growth at strong startups is lopsided, about 90 percent of it from 10 percent of what they tried, and that most experiments fail.
Freeze the buyer and the bottleneck first
Write the buyer and one bottleneck before you freeze a calendar. Do it once a few customers have come back.
Lerner’s founder-led playbook starts from a live product and some happy customers. If they exist and growth does not, the gap is go-to-market, not missing features. The founder cannot hand the search off yet: thinking without shipping, building from a hunch, and hiring an expert without the context are slow ways to learn. See when to hire the first growth marketer and what the founder should stop doing.
Five lines: role and situation, the outcome in their words, where they looked, qualified (fit, intent, the handoff), and the bottleneck. If doubling that step would not double the business, it is not the limit. About five recent buyers show a pattern. Refusals often say only that it was not for them, so ask buyers what they bought and where they looked. The north star is value delivered, not a registration. Use rough counts.
Separate a kickstart from an engine
Stay on an unscalable tactic until one loop carries most of the growth.
Lenny Rachitsky’s Racecar framework separates the motions. An engine feeds itself: search on Google, YouTube, or TikTok, paid distribution, sales, or virality, including invites, word of mouth, sharing, and incentivized referrals. A kickstart is unscalable work toward a first 1,000 users. In B2B he names three: people you know, where buyers already gather, and press. A turbo boost is a one-off spike. Use kickstarts, maybe a few boosts, until the engine carries the majority, then lubricants: conversion, activation, and retention, plus an occasional one-off boost. Start a second engine before the first one flattens, and keep tuning the first.
List the places this buyer already shows up, rank them by fit and by a test this team can run within two weeks, put at most three through the screen, and commit to the one that passes.
| Question | Keep it | Drop it or wait |
|---|---|---|
| Seen this buyer there? | Yes, in a call or community | Picked from a list |
| Signal inside one or two weeks? | Yes, or a smaller test exists | Only a readout months away |
| Who runs it this month? | Founder or one marketer | A specialist you lack |
| Kickstart or engine, and repeated? | Yes | A one-off boost, or it worked once because you were there |
Reopen only when the same effort stops producing the same conversations, which is saturation, not a reason to open several channels. Doing more without a full team depends on this screen.
Write the assumption you might be wrong about
Score the idea, then write the assumption a customer might prove false. A task with no prediction does not compound.
Five moves: gather insights and past tests, score, select, run, and write the lesson. Score the key driver the change would move: traffic, activation, conversion, retention, or referrals, not the task itself. Score impact from 1 to 5 and effort from 1 to 5. Drop what misses a driver. Start at the rate-limiting step. Test a hard idea through its riskiest assumption only: a landing page, a button for an unbuilt feature, a keyword check, or doing the service by hand. A long search program starts by checking demand and prospect quality.
The note has four lines.
- We believe they are not acting because of this reason.
- We will change one thing.
- This metric moves this way, by enough to matter.
- If right, we do this next. If wrong, we stop that. If neither result changes next week, do not run it.
A circular note only predicts the idea will work, such as a new headline and more demos. Name why they are not acting, such as a headline that misses the outcome they used. Use a metric deeper than a click, readable in one week or two. Parallel teams often run 3 to 5 tests a week. Two people who share every test run one, or a second on a different skill. Stay on one to three bottlenecks. A workflow comes after the note.
Keep a week two people can finish
One weekly meeting checks the milestone, the live bet, and the blockers, and the rest of the week ships that bet. A department calendar on two people makes activity, not a decision.
Emily Kramer’s operating cadence is dialed to the company. Each quarterly goal has one owner, who sets monthly milestones the week before the quarter, adjusts them monthly, and keeps them beside the tasks. The meeting nearest the first of the month checks milestones and blockers. Do not add a posting streak, a newsletter, and a paid channel in the same week unless one of them is the bet.
| Point in the week | The only job |
|---|---|
| Start | Ship the one bet. The owner is on the note. |
| Middle | One research conversation with a recent buyer, then replies and follow-up on people already warm. The next bet starts from that customer insight. |
| End | Read the result against the prediction. Write keep, change, or stop. Lock next week’s bet. |
Cap open work at one live bet, plus upkeep of a motion that already worked twice.
A second win earns the playbook
Write a playbook after the same motion has produced the outcome twice.
On the page: objective, audience, inputs, steps with an owner and a due day, and two or three numbers. Around it: one request path, one brief, one checklist, one board. Before a prospect sees it, agree the brief, finish the asset, match the tracking names, and tell whoever sells.
The handoff is a rule: fit, intent, required fields, owner, and speed. Once a month, read acceptance, time to first touch, and meetings. Before you add volume, low acceptance means you rewrite who is qualified, after checking the arrivals are the buyers you named. Do not automate a louder version.
Make the numbers comparable, then decide
Compare channels only when every public link uses the same three parameters, spelled the same way. A figure that cannot force keep, change, or stop stays off the review.
Google’s campaign URL guidance says to always set utm_source, utm_medium, and utm_campaign. Values are case sensitive. One lowercase scheme, one source per platform, one medium per channel, and a campaign value that matches the internal name. A missing parameter shows as (not set). Read them in Traffic acquisition. The landing-page query string omits them.
Management watches the north star, the bottleneck, and whether the work shipped. Operators watch broken forms, bad routes, missing tags, and stale numbers. The kill rule is the prediction. On a miss, isolate the first break: targeting, attention, readability, comprehension, trust, resonance, or the cost and benefit trade-off. For readability, show the line for three seconds and ask what it said. A drafted report still needs those definitions.
Automate the handoff after two manual runs
Automate a step only after someone has done it by hand at least twice and can describe a bad record. The tool repeats the handoff, while a person still picks the buyer, approves what a prospect reads, and decides if the channel continues.
Order: tags and the readout, then routing for a lead you can already qualify, then a follow-up someone has sent by hand. See when automation makes sense. If the company is established in the EU, or it is not and it offers goods or services to people in the EU or monitors them there, set a specified purpose and a lawful basis before the file is kept, and a contract before a tool processes it. A freelancer must not be the only login.
Ninety days, then a week without the author
Ninety days installs the files, then proves a week still runs when the author is away.
| Stretch | Done when |
|---|---|
| Days 1 to 30 | Buyer page written. About five recent customers interviewed. Funnel in rough counts. One channel passed the screen. CRM is the shared record. |
| Days 31 to 60 | Weekly meeting held. Every bet has the four lines and a kill rule. First playbook is follow-up you already do. One tag scheme. |
| Days 61 to 90 | One motion worked twice, with an owner. Checklist in use. Next milestones drafted. Help only for the step the playbook cannot cover. |
Pass when that week runs with the author away, before a full-time hire.
Keep the cadence off in four cases
Skip the cadence when you cannot name the buyer, nobody has come back, the offer changes after every call, or nobody will write a prediction. Also stop for a circular note, an ownerless campaign, a playbook for work you have not run, a tool that writes to a prospect or keeps a channel going, a board of registrations, or one freelancer login.
I leave the loop in your accounts
I am Piet Baudoin. I work alone as Poldermarketing, an eenmanszaak, fully remote, in Dutch and English.
The offer is all-in: one person for marketing, AI, and automation, for a freshly funded startup anywhere, from the first message to the first customers. You do not hire a separate specialist for each part. You can hire me freelance for a bounded project or for a few days a week. I am an AI-native growth marketer, strong in AI, content, automation, and workflows, and I build and run the work in your accounts. Google Ads and Meta Ads are newer for me: I set them up and review them. I am the wrong hire when the job is scaling those.
The free growth scan shows how the current site reads to a buyer. How I work is the shape of an engagement. Positioning and messaging is where I start when the words are still moving.
Questions people ask
How many tests should two people run in a week?
A team that can run different kinds of tests in parallel often aims for [three to five a week](https://review.firstround.com/the-secret-to-running-effective-growth-sprints-follow-this-process-to-learn-faster/). If the same two people must do every test, run one, or two only when they use different skills, such as a message change and a routing fix. Most well designed tests fail. A failure that names what you learned still finishes the loop. A week with several campaigns and no written prediction does not.
When is it too early for this cadence?
Too early when the product has no customers who came back, when you cannot name the buyer, or when the offer changes after every conversation. In that stretch the work is interviews and a manual motion, not a calendar. It is also too early when nobody will write a prediction before the work ships. A rhythm around a muddy offer only makes the mud move faster, and the Friday meeting becomes a debate.
What belongs in the experiment note?
Four lines: the customer assumption you might be wrong about, the test, the metric and the direction you expect, and what you will do if you are right or wrong. If neither result would change next week, do not run it. Add the audience, the channel, the owner, and the lesson after you read the result. Skip a note that only says the idea and that you hope it works.
Does a content calendar count as a process?
A calendar is a schedule. It becomes a process when each item has an owner, a customer assumption, a metric, and a decision to repeat or stop. One answer to a question from a sales call can be the whole week. A streak of posts with no readout is random work with dates on it. Write the checklist after that piece has produced the outcome twice.
When should a handoff be automated?
After a person has done that handoff twice and can say what a bad record looks like. Start with consistent link tags and a readout, then routing for a lead you already know how to qualify, then a follow-up someone has already sent by hand. Keep a person on anything a prospect will read and on the decision to continue a channel. A form that only creates a row and pings a chat does not need a new tool.