What I would do differently if I started this business over
# What I would do differently if I started this business over
Somebody asked me last month what I would change if I started over. The three answers I gave on the spot were the flattering kind. Work harder earlier. Worry less. Pick one thing. All three are free to say and none of them would have altered a decision I actually made.
So here is the version with the invoices open.
What I run, so you can weigh the advice: mobile proxy lines on carrier SIM cards, a rack of Android phones rented by the week, anti detect and multi account tooling, scraping, and content sites with a video pipeline underneath. A SIM is about ten dollars a month. A modem depreciates at roughly a dollar fifty. A handset costs two hundred dollars and either earns that back or does not. None of it was designed. It accumulated.
Owning the hardware is the decision I would defend hardest
The case against buying the metal is real. Capital tied up in things that break, and a person present on a Saturday because a bus powered hub browned out under thirty modems and dropped half a rack off the network. The fix cost about a hundred and fifty dollars.
What it bought me is a cost floor I can write on paper, one that does not move because someone in another company edited a pricing page on a Tuesday.
I have watched resellers lose their margin to exactly that. Their business was a spreadsheet built on a supplier's price, and the supplier owned the spreadsheet. Asset light is a fine strategy for whoever sells you the asset.
This is the position I get argued with about most, usually by people whose instincts formed in software, where the next unit costs nothing. Mine costs ten dollars a month forever, and that changes what a good decision looks like all the way down.
The other two I would repeat
Everything here that worked started as a tool I needed. The proxy lines because I could not stand what I was renting. The content engine because writing all of it by hand was taking three days a week. The things that went nowhere started the other way round: I read that a market existed and built for a customer I had never been.
And I documented processes early, badly, which paid off in a way I did not predict. The lanes I have handed over to software were, without exception, the ones that already existed as a written list of steps. The ones still sitting on me are the ones I never wrote down, because I am the only person who does them. I still have four or five, and each is a reason I check my phone on holiday.
Starting narrower, and the caveat that nearly cancels it
I run fifteen brands. In aggregate the portfolio is defensible and I will defend it. One channel dying threatens nothing. One supplier going bad threatens nothing. A spread of small lines has survived things that would have finished a single product company.
Each of the fifteen got about a fifteenth of my attention and almost none of them wanted a fifteenth. They wanted half of it for six months and then hardly any of it forever. I can name the three that pay for the rest, I could have named them two years ago, and I did nothing about it.
The sharper criticism is about why the second business existed at all. It was a hedge against the first one failing. That hedge cost more than the risk it insured, because the first one was never going to fail. It was going to be slow. Slow does not need a hedge. Slow needs patience. Patience is cheaper and much less entertaining.
Diversifying early is a way of postponing a decision while looking prudent. I did it for four years.
Charging earlier, because free users answer a different question
Free users are generous. They say encouraging things and ask for features. Every one of those signals comes from a person carrying none of the cost of being wrong about you.
The week the first customer paid, I got a question no free user had asked in months of using the thing. What happens at three in the morning when a line drops and I am asleep. That question is the product. I had spent a year building around the other ones.
Free also changes who shows up. A free tier fills with people evaluating a dozen tools this week, so the feedback is comparative. They will tell you your dashboard looks worse than someone else's. They will never tell you your failover is a lie.
The price does not have to be high. It has to be above zero. My rule now is that the first number should be slightly embarrassing to say out loud on a call.
Automating later than instinct says
This is the change that irritates me most, because it looks like the opposite of a mistake.
Automate a process before you have settled it and the automation freezes whichever version you happened to have that week.
I wrote a provisioning script when I had about a dozen lines. It generated a name for each port from a pattern I picked in twenty minutes on a Thursday. That name is now read by three systems: a dashboard groups on it, a reconciler matches on it, a billing job parses fields out of the middle of it. The name became a contract I signed before I understood what it described.
Changing it costs a week across three codebases for no visible improvement, so it is never the most important thing on any given Monday, so it never happens. That is the real cost. Not the script. The way the script hardened into a rule while I was looking elsewhere.
My test now is boredom of a specific kind. Bored of the same eleven steps in the same order means the process has stopped moving. Bored of the general area means I have not found the steps yet.
Having a model write the automation has made this worse. Building the wrong thing used to cost two days, and two days is enough friction to make a person think first. Now it costs forty minutes. The thinking was always the expensive part, and the cheap part got cheaper.
The help I should have hired two years earlier
I kept the dull half myself for a reason that sounds respectable. It was cheap to keep and expensive to explain.
Which is true. Explaining takes a week. Doing it takes ninety minutes. You pay the week once and the ninety minutes every week, and I was weighing the week against a single instance of ninety minutes. That is the arithmetic of a man who has already decided not to hire.
The less flattering half is that I was better at it than anyone I could have hired. Being the best person in the building at a job nobody should be doing is a comfortable place to hide, and it looks like diligence.
Some of that work now runs on a machine and people assume that closes the argument. It does not. The machine took transcription, tagging and first drafts. It has not taken the part where a person notices a customer is annoyed before he says so.
The things that look like mistakes and were tuition
Four clean regrets in a row is a story, not a record, so here is the counterweight.
Going wide is how I found out which of these actually pays, and there is no reading your way there. It also produced the one asset I would not trade: I understand what a multi account operator needs because I sell him lines, rent him phones and run the tooling myself.
The free year earned its keep too, even though almost none of the feedback did. That was the year the thing got reliable enough to charge for, and it got reliable because sixty odd people were hammering on it for nothing.
Why a retrospective is worth less than it sounds
A retrospective is written by the one person who already knows how it turned out.
I can tell you to start narrower because I know which one was worth keeping. Standing at the beginning with two ideas and no way to tell them apart, I could not have. The only reason I can now is that I ran both of them badly for four years. Every decision looks obvious in a mirror that shows one road.
Some of this is unfalsifiable, and I mean that literally. Charge earlier cannot be tested. There is no version of me who charged from the first week, and if there were, and he had stalled, he would be writing a post about pricing too early and killing his own distribution.
Same with start narrower. The operator who went narrow and picked the wrong narrow thing has a post titled I should have diversified, and his probably travels further than this one.
So take the four changes as a description of what I believe about my own case. One operator, one temperament, one country, seven years. That is an anecdote in a lesson's clothes.
The one item that is arithmetic
Own the part that is hard to replace, or at least the thing your price depends on. That has nothing to do with my temperament or my market. If the main input to your business is priced by somebody else, your margin is a line item on somebody else's agenda, and you learn what they decided when everyone else does.
A SIM card at ten dollars a month is a worse business than software in nearly every measurable way. It is also mine. I will take the worse business I control over the better one I am renting, and seven years of bad Saturdays have not moved me an inch on that.
Get new guides and videos first — join the Telegram channel.