Marijn Booman/go-to-market operator/AI systems

From a business on a map to a reply in the inbox.

I design and run the systems that do go-to-market — discovery, qualification, outreach, follow-up — and keep a person on the calls that matter. Two worked examples below, and the stack I run on myself.

Operating rules

Six rules the systems are built on

Not slogans. Each one is a decision that shows up in the wiring — usually in what runs before what.

Spend last: Every stage filters before the next one costs money. The free checks run first; the paid lookups only ever see the leads that survived them. Ordering is most of the cost control.
Automate the work, not the judgment: The model does the first pass and writes down its reasoning. A person still decides, and can see exactly what the model was thinking when it gets one wrong.
Use the signals other people skip: Photos and customer reviews sit in public on every map listing and tell you how a business actually operates — retail or wholesale, size of premises, who it serves. Most tools only read the website.
Deliverability is a design constraint: How the mail is sent decides whether it arrives. Low volume from a real mailbox, genuinely written, gets read where a campaign blast goes to a tab nobody opens.
Enrichment is a guess until verified: Addresses come back with a confidence score, not certainty. Some mail servers accept anything, which makes a wrong address look correct. Verify against the company's own domain or drop it.
Pick one system and go deep: A GTM stack has ten things worth building. Building three of them badly helps nobody; building one properly changes the quarter.

Case study

A lead pipeline, map to inbox

gig.solja.one — local services marketplace

Situation

The marketplace needed a steady flow of the right local businesses to talk to. Buying a contact list doesn't fix that: the lists go stale, most of the names are a poor match, and everyone else in the market is emailing them anyway. Sales was spending most of the week looking for people to talk to instead of talking to them.

What I did

Designed a pipeline that does the finding and the groundwork and hands a person a short, qualified list with an email already drafted. Built with an engineering partner — the GTM design is the part I own: who we target, how a lead earns its way to outreach, how the mail actually goes out.

Figure 1 — the pipeline

Seven stages from a map search to a sent email. Everything left of the dashed gate is free; the paid lookups — scraping, people, enrichment — only ever run on leads that already passed the review. Replies come back into the system, get sorted by sentiment, and set the next touch.

Stage	What happens	Runs on
1 · discover	Every business in a city that matches your keywords, with address, category, rating, reviews and photos.	Google Maps API
2 · qualify	A model reads the photos and reviews against a written description of the ideal customer, and returns a reason for every call.	Vision + text model
3 · scrape	Reads the company's own site for a public inbox; retries through a residential proxy when a site blocks readers.	context.dev
4 · find people	Decision-makers and their verified addresses — at least two per company, so one holiday doesn't stall the thread.	Apollo.io · Apify
5 · store	Everything deduped and linked back to the original map record, so one dataset drives the map, the outreach and the follow-up.	In-house CRM
6 · draft	One email per contact, in the sender's own style learned from past mail. Automatic checks catch leftover placeholders.	Writing model
7 · send	Out through the sender's real mailbox — threads properly, sits in Sent, no campaign headers. A pixel records opens.	Gmail API

What broke, and the fix

Catch-all domains. Addresses "verified" against the wrong company because the mail server accepts anything. Fix: check the address resolves to the company's own domain, and drop the ambiguous ones rather than guess.

Sites that block readers. Some company sites refuse automated requests. Fix: a fallback pass through a residential proxy, so the request looks like an ordinary consumer connection.

Open tracking counting me. The pixel fired on the sender's own view of the message. Fix: filter the sender's client, and treat opens as a rough prioritisation signal, never a number to report.

Each one needed its own fix. That's the half of the work a diagram never shows.

Case study

Selling location intelligence

marketlyzer.ai — location analytics

Situation

Marketlyzer has genuinely good data: score any address before you sign the lease, track every branch of a chain, map a territory with contacts and new-entrant alerts. What it didn't have was a way to sell it. The US market is saturated — Placer.ai owns the category and enterprises already run a dozen analytics tools. Europe has no equivalent and a real appetite for the data. That gap was the opening.

What I did

One vertical, not ten. Started with gyms, because the strongest proof was there: the model predicted membership for a Swedish chain's new sites to about 80% accuracy — 1,829 predicted against 1,800 actual — before a single one opened.
Sold the wedge, not the dashboard. Enterprises don't want another dashboard; they want a number. The pitch became: give us fifty streets, we'll tell you the revenue per branch, and which fifty to skip.
Two buyers, two decks. Lease managers get traffic and turnover by category. Mall directors get solution and payback. Land one or two brands as a reference, then expand store by store — not through top-down procurement, which runs one to two years.
Wired the CRM to the product. The CRM sits on the Marketlyzer backend, so the same record drives the map, the outreach and the follow-up. Outreach runs from a company address, automation-first and A/B tested; demos are where a person takes over and closes.
Priced it so the mall makes money. A package for the mall, brands pay separately to join the brand-mix tool, the mall takes a cut. The tool becomes a revenue line, not a cost.

Figure 2 — the motion

One wedge — the forecast — feeds two decks for two buyers. Land a reference brand, then grow sideways. The usual alternative, selling top-down into procurement, is a one-to-two-year cycle; this motion routes around it.

What happened

Early days, and the numbers are soft: a case study landed with Inditex, and outreach was replying at roughly 15–20%. The target the motion is built to hit is one closed enterprise deal within thirty days of the funnel going live.

My own stack

I don't pitch systems I don't run

A few of the ones pointed at my own week:

Figure 3 — the chief-of-staff loop

Meetings, calendar and inbox go in; a brief, a set of drafts and calendar holds come out. The state file records what's handled so the next morning's run doesn't raise the same thing twice. Built this one recently — first run cleared a month of meeting notes into a single briefing.

Chief-of-staff loop. Every meeting is transcribed, filed, and read by an agent that writes a daily brief: who I owe a reply, what I said I'd send, which meetings should move, which emails I forgot. Figure 3.
Three outreach streams, always on. An agent runs three parallel searches — target roles, recruiters, client leads — and nudges me on anything that's gone quiet, with a weekly follow-up sweep.
Pipeline in the terminal. Mail pipes into a sheet the agent can read; it flags unresponsive threads and surfaces the next action each morning.
Origami: trialled, cut. Automated founder outreach. Fast, but too cold without better filtering, so I switched it off. Worth saying — knowing what to stop running is part of the job.

The pattern

Every one of these is the same shape

Intelligence — who exists, and who is actually a fit
Targeting — a human checkpoint, with the model's reasoning attached
Outreach — one-to-one mail at machine scale, from a real mailbox
Pipeline — one record, deduped, linked back to the source
Follow-up loop — reply in, sorted by sentiment, next touch scheduled
Measurement — honest signals, used to prioritise, not to report

The machine does the finding. A person still decides. That's the whole design.

About

The human behind the systems

I've spent the last 15+ years building, launching and scaling businesses across Europe and Asia — usually in situations where there wasn't an existing playbook.

I'm Dutch, grew up in Switzerland, and have lived and worked across Beijing, Shanghai, Singapore, Kuala Lumpur, Tokyo and Barcelona.

That journey shaped how I work today: understand a market quickly, find the opportunity, connect the right people, build a system around the problem, test it in the real world and keep iterating.

Long before AI agents, much of my career was already about building systems — for market expansion, partnerships, growth, operations and commercial execution.

A few chapters along the way

25+ markets launched and scaled. Worked across Asia, Europe, the Middle East and other emerging markets, helping businesses go from initial market entry to repeatable commercial operations.
HotelQuickly — 0 → $150M GMV. Co-founded HotelQuickly, one of Asia-Pacific's early mobile-only hotel booking platforms. We grew from an idea into a business operating across 18 countries, raised approximately $15M and ultimately exited. Read about HotelQuickly on TechCrunch →
iFlix — 3 → 24 markets. Helped expand iFlix across Asia, the Middle East and emerging markets as the streaming business grew from approximately 1.2M to 25M users, using partnerships with telcos and local market operators as a major distribution engine. Read about iFlix in Variety →
FreakOut — building ahead of an IPO. Led regional expansion for Japanese adtech company FreakOut, opening operations across markets including Thailand, Turkey, Indonesia, Taiwan, Malaysia and Singapore. FreakOut went on to become one of Japan's notable adtech IPO stories. Read about FreakOut's IPO →
Ferryhopper — first profitability in 7 years. As Commercial Director, helped Ferryhopper reach its first profitable year, while growing revenue by roughly 40% YoY, increasing B2B GMV by almost 200%, and developing major strategic distribution partnerships including Omio.
Open English — €3M new net revenue + 2 M&A projects. Led expansion initiatives across Spain, Portugal, Turkey and India, generating approximately €3M in new net revenue while also supporting two M&A projects: English Ninjas in Turkey and Enguru in India.
$3.6M crowdfunded across 9 campaigns. Built and ran nine Kickstarter and Indiegogo campaigns with a 100% funding success rate, helping founders turn early concepts into commercially validated products.
A few examples:

2 exits. 1 IPO. Multiple zero-to-one journeys.

Across travel, streaming, education, adtech, fintech and marketplaces, the common thread has rarely been the industry. It has been figuring out how to make something work where the answer isn't obvious yet.

Today, I'm applying that same mindset to AI. Instead of only building teams, processes and commercial operations, I can now build agents, automations, workflows, data layers and interconnected systems that research, analyse, create and execute alongside me.

The tools have changed. The way I think about building hasn't.

AI isn't replacing my experience. It's allowing me to turn that experience into systems that can operate at scale.