Recovering tens of millions for a state government, spending almost nothing to do it
Somewhere inside most state governments there is a number nobody is actually checking. A fee, a levy, calculated entirely from what an industry chooses to report about itself. The figure gets accepted every year, and the size of the gap between what is reported and what is real stays invisible, because nobody built a way to see it.
I found one of those gaps this year, in a sector I know well, in a state levy that depends entirely on operators declaring their own numbers honestly. I did not have a large budget to throw at it, and I did not have twenty years of relationships inside that government either. What I had was the idea, a developer I trust, and a decision about how to spend what little I did have.
What I actually had to work with
Four things, and not one of them was money in any real quantity.
The first was code, and speed. A small system, built lean, that can watch an entire stateās worth of booking activity and compare it against what is being declared, a job that used to require an army of auditors and still would not get done properly. One developer and a few months bought the core of it. What compressed those months further was a wider circle of independent developers I already knew well enough to call in favours from, which is its own kind of asset built up over years and worth nothing on a balance sheet.
The second was an insight nobody had acted on. The data needed to catch this gap already existed, sitting inside booking platforms, available the whole time. Nobody had connected it to what the state was collecting. Being the person who makes that connection first is its own kind of asset, and it costs nothing but attention.
The third did not come from this project at all. It came from years spent inside state government digital projects before this one, building things for a state administration. I already knew, walking in, roughly where this kind of data actually lives, which entities hold the licensing and registration records that matter, and who the real gatekeepers are versus who just looks like one on an organisation chart. That is not something you can look up. You get it by having sat inside the machine before, filed the actual paperwork, and dealt with the specific people who make a state government move or stall. It is the reason the early conversations took weeks instead of years, and it is why I knew which letters to ask for and who needed to sign them before anyone told me.
The fourth was the one I did not have myself: trust inside this particular government, built over decades, that I could not manufacture no matter how good the software was or how much of the machine I already understood. So instead of spending the next fifteen years trying to build that access myself, or letting the idea die because I lacked it, I structured a partnership and traded a fair share of the upside for someone elseās twenty years of standing. That is not a consolation prize. That is buying, with equity instead of cash, the one input I could not produce on my own.
Why this is not a build story
It would be easy to tell this as a story about the software, and I do not think that is the interesting part.
A developer optimises for the elegance of the system. A founder optimises for shipping it. Neither of those decisions is what makes this worth anything to a state government. What makes it worth anything is a set of decisions that have nothing to do with code: which asymmetry to go after, who to trust with a slice of the outcome, how to structure a deal so somebody elseās incentives point the same direction as mine, and how much of my own capital to actually risk chasing it, which in this case was very little.
That is capital allocation, not engineering. Capital allocation, in the broadest sense, is closer to the actual job at the top of a company than most people admit. Building the thing is table stakes. The scarcer skill is recognising a gap that is worth real money to an institution far larger than you are, and then assembling access, trust and code you mostly do not personally own into an outcome you could never have produced by writing more code yourself.
There is a timing piece worth being honest about too. A few years ago, matching messy, inconsistent booking data against declared numbers at this scale would have been a much bigger and slower engineering job on its own, regardless of who was doing it. That gap has narrowed a lot recently, and I leaned on that. I am not going to describe exactly how, because that is the one detail of this actually worth keeping close, but the honest version is that timing did some of the work persistence used to have to do alone. As far as I know, nothing quite like this is running in any other state in Malaysia yet. That will not stay true for long, which is its own argument for moving now rather than waiting for the idea to feel more finished.
What the gap is actually worth
I am not going to publish the real figures, because the exact number would help identify a negotiation that is still open. But the shape of it is easy enough to describe.
A mid sized tourism state can easily carry several thousand active short term rental listings on platforms like Airbnb at any given time, sitting alongside the hotels and resorts that already report through more visible channels. Most of those listings are either underreporting what they actually earn, or not registered anywhere at all. Even a modest, single digit contribution per night, collected consistently instead of patchily, compounds fast once you multiply it across a full year of bookings. Once you add the properties that are declaring nothing whatsoever, it is not hard to get to a figure in the tens of millions of ringgit a year, sitting inside a single state, that nobody is currently collecting. That is the scale of gap a lot of state treasuries are sitting on without knowing it, and it is why a small, cheaply built piece of software pointed at the right number can matter more than its size suggests.
The honest state of it
Right now the concept is in front of a state government for review, with real interest in taking it further. It is not signed, and I am not going to pretend otherwise. Bureaucracies do not move like customers do, and I am still learning what patience actually looks like when the counterparty is a government rather than a client with a budget cycle I understand.
If it lands, the number recovered is not a rounding error for a state treasury. It is real money, collected accurately for the first time, that can go back into the exact programmes the levy was supposed to fund. If it does not land, I still learned something a lot of operators never test directly: how much you can create with almost no capital of your own, if you are precise about which scarce thing you are actually short of, and honest about what you are willing to trade to get it.
That, more than the software, is the part I would want a board asking me about.
It is also not the first time I have built something a government never asked for and then gone and found the door myself. I doubt it will be the last.