My role
Product Manager, owning product strategy and the alpha scope (MBA practicum)
Team
Four co-founders, with CTO Itai Biton as the only developer
Timeline
May to October 2026, alongside a full-time job
Outcome
A focused alpha live in production on 9 Sep 2026, 16 weeks after I joined
~$1.3B
a year in American home purchases in Israel, the first market I sized
6-8 → 1
separate portals on the roadmap replaced by one person who owns the deal
16 weeks
from my first day to the alpha in production
The problem and why it mattered
If you live in New York and want to buy an apartment in Jerusalem, you end up hiring six people you’ve never met, from a broker to a tax consultant. None of them owns the deal, so the buyer project-manages the most expensive thing they’ll ever buy, often in a language they don’t read.
REOS.Global was built to sell that purchase as one managed service. When I joined as product manager in May 2026, the company was pre-seed with a live product, and the plan was six to eight separate portals over twelve months, all built by one developer, CTO Itai Biton.
Sizing the first market from the bottom up, from public counts of foreign buyers, I got roughly 930 American purchases a year, around $1.3 billion. That’s a lot of money moving through very few deals, enough to carry one deal flow that works but nowhere near enough for six portals built ahead of demand.
I had six to eight hours a week next to my job at PostHog and an MBA, nobody reported to me, and most decisions happened in Hebrew voice notes. Nothing was written down either, so the alpha requirements document I wrote became the company’s first.
What I changed
One person who owns the deal
After walking every screen as a buyer, I recommended starting with English-speaking buyers in Jerusalem. Then on 16 June, in person, I argued that a buyer can’t judge a lawyer or a mortgage advisor from a profile, so they should never have to pick one. One named account manager should own the deal end to end, by hand first and automated later.
Itai adopted the model in that meeting.
Championship, I adopt it today.
Itai Biton, CTO, on the concierge model, 16 June 2026 (translated from Hebrew)
Later research backed the call, since three of the five nearest competitors had built portals ahead of demand and none of them survived. A person doesn’t scale, though, and that was the trade I accepted. To pay it down, I pushed to track the account manager’s work from day one, because how that person works is the spec for whatever gets automated later.
Specs for every stage of the deal
In July I shipped the onboarding spec, replacing a long multi-step form with one intake screen of five defaulted questions.
In August Itai asked what the leanest alpha had to ship to work fully. I wrote a master requirements document over six smaller ones, organised around the six stages of a deal, marking every step as software or a person. I marked what already existed myself, so the founders corrected a list instead of filling in a blank one.
Scope cut to what one developer could ship
I went through Itai’s own page list line by line, marking each page in or out with one line of why. He then handed me the scope meeting, and the buyer portal came out cut to five features. A week later I turned it into a build queue of five milestones and 12 issues, with a definition of done.
The calls I made and what they cost
The homepage pitched an AI assistant as the advisor. I argued against an AI that could give a wrong tax or legal answer when trust is what we’re selling, so in the alpha the AI chases documents and scheduling and a named person owns every decision. We gave up the flashier demo to keep that liability with a person.
I also recommended showing real listings and real costs before asking for an account, trading away early leads to put the value first. The passwordless sign-in from that spec, Google or an email link, is what production uses today.
And I checked the numbers. The all-in cost estimate applied the Israeli resident’s purchase tax to foreign buyers and understated what they’d pay, so fixing it moved to the front of the build. Six parallel research passes with AI agents also reversed my own pricing recommendation the same day, because nobody in this market charges the buyer for coordination.
What happened
By 28 August every milestone was closed and the alpha was merged, with nothing dropped from the plan. On 9 September it went to production, 16 weeks after my first day, and anyone can sign up and use it today.
The queue mattered less than that sounds, though. Six of the 12 issues were already done on work Itai hadn’t pushed yet, so what the queue really gave him was a definition of done and a clean way to release.
Measurement didn’t land either. The events behind my July metric were never built, and the seven I wrote, tests first, weren’t merged before my work ended on 22 September. There’s no revenue yet and no purchase has gone end to end, and the next build is aimed at brokerages, where I’d argued the first money could come from.
What I’d do differently
I’d read the code before writing the spec, to see how far along Itai already was. I’d get into a customer conversation in week one and keep asking until I’m in, because I asked twice in June and let it drop.
And I’d make the analytics events part of the definition of done, so an issue can’t close before its number can be read.
