The year our own back office broke
Millionyse builds software for other businesses, and automates the processes that run behind it. Teams in Australia and Sri Lanka, with clients across five primary countries.
By 2023 it had enough clients that its own admin started failing. Invoices went out late, or not at all. Software and domain renewals lapsed and cost us in fees. Projects went quiet and died, not because anyone did the work badly, but because the onboarding and the paperwork around it outran the people available to do them.
A company that automates other people's processes was losing money because its own were a mess. That is the part we would rather not write down, and it is also the reason any of this exists.
We tried the obvious thing first
We hired. And it worked. It genuinely helped, and we would tell any business in the same position to do exactly the same thing.
But hiring solves the problem at the speed of a person. Every new client meant more of the same work, and the only lever was more hours from more people. It was an answer with a ceiling, and from where we were standing we could see the ceiling.
So we built our own tools
Not a product. Just tools for the jobs that hurt most. The digital footprint work came first, the setup every new client needs before anything else can begin. Instead of somebody doing it by hand for the tenth time that month, the tool did it. Work that had taken hours took minutes.
And it still was not enough, because everything we had built was about doing the same work faster. The work itself never went anywhere. Somebody still had to notice it was time, start it, check it, and carry the result to wherever it needed to go next. We had automated the typing and kept the job.
Then we tried handing the whole thing to AI
When the models got good enough, which they had not been when the agency started, we did what everyone else did. We gave an AI the data and asked it for the outcome.
It hallucinated.
Not dramatically, which is the dangerous part. A figure slightly wrong. A detail confidently invented. An outcome that looked entirely reasonable until you checked it against the record, and by then it had already gone out. Survivable on a draft social post. Not survivable on a client's records, a signed document, or anything that leaves the building with your name on it.
That is where our approach came from. Not from a whiteboard. From watching a model guess on work that mattered. What we do instead.
What everyone actually wanted
Neither we nor our clients wanted another dashboard to learn. What both sides were asking for was someone to do the execution. Some of our clients had started hiring virtual assistants, which told us something we had been slow to see. They were not shopping for better software. Hiring a person was simply the only way anyone had ever been able to buy the outcome.
The pattern never changed. Different industries, different sizes, the same bottleneck every time.
Who we are building for
We are building for businesses past the point where one person can hold everything in their head, and nowhere near the point where they can hire a back office to take it off them. That is most growing businesses, for most of their life.
For a real AI employee, the AI is the smallest part. The automation and the infrastructure underneath are what decide whether it survives contact with a business that has actual money moving through it. We are not racing to be first. We are racing to be the one still standing when the novelty wears off.
Start with one thing it can take off your hands.
You don't have to move your business across. Pick the job that annoys you most, the chasing, the invoicing, the inbox and let it take that one first.