This newsletter is going to be about a simple experiment:
What happens when you take a normal, boring service business and slowly start adding AI into the way it operates?
Not a software company. Not an AI startup.
A real-world service business with messy customers, workers, proposals, site visits, delays, Whatsapp messages, payments and unfinished projects.
That business is mint.
mint is a pool & landscaping company that I started in Bahrain. We mostly do villas: gardens, pools, outdoor spaces, etc.
we are now around 14 months into building it, and the first year has made one thing clear: before AI can make the business better, the business itself needs a strong operating system.
the funny thing is, I didn’t actually plan to be involved in the business much at first.
the original idea was to hire someone very experienced, let the business operate more or less normally, observe how it works, and then slowly figure out where AI could help. And on paper, that made sense. The person running it had 20+ years of experience in the industry, so I assumed the best thing I could do was stay out of the way for a while and let the business reveal itself.
That part worked.
Just not in the way I expected.
I only really started getting directly involved in the past few weeks, so a lot of this is still fresh. But once I got closer to the actual day-to-day, the biggest issue became obvious very quickly.
Getting projects was not the hardest part.
Delivering them cleanly was.
The pattern had been pretty clear: we would get a project, start it, then it would drag. Small issues would pile up. Clients would get annoyed. Close-out would take forever. Payments would get delayed. Commuication halts. And then I would end up getting pulled into the mess at the end.
After seeing this happen enough times, the problem became clear.
We had one person trying to handle too much: sales, proposals, client communication, project execution, labor coordination, complaints, and close-out.
The thing I eventually realized is that the person I had hired was actually very good at one part of the business: talking to customers, understanding what they wanted, painting the vision, and getting them excited about the project. That is a real skill.
But it is not the same skill as getting a project finished. Execution needs a very different personality: someone detail-oriented, organized, slightly annoying about follow-ups, and willing to keep pushing until every small item is closed. That was not him. So instead of forcing him to be good at both, I split the role and brought in someone whose job is simply to get projects across the finish line.
Mr. Experience (now Mr. Sales) is now focused on sales and client acquisition.
Once the client signs, the project moves to Mr. Operations: who manages the site, coordinates the workers, tracks progress, controls costs, and pushes the project to completion.
I also reduced some of our fixed overhead to make room for this without increasing monthly burn. Salaries that were too high for where the business is today were adjusted, freeing money up for hiring Mr. Operations.
That’s the boring part.
But I think it matters because AI will not fix a business that has no operating system. If projects are messy, responsibilities are unclear, and context is scattered across WhatsApp messages and people’s memories, then adding AI just gives you a faster mess.
The next step is proposals.
Right now, proposals take too long. For many jobs, Mr. Sales has to manually recreate the client’s whole outdoor area in AutoCAD just to show an idea, before we even know if the client is serious… which takes like a whole day, at least!
For early proposals, we could just use ChatGPT image gen to quickly create visual concepts, explain the design direction, and help the client understand what we’re thinking.
Then, once they commit & pay, we can spend time on proper drawings and detailed execution plans.
So my next target is simple:
send more proposals, faster, without wasting serious design time before the client is serious.
After that, I want every project to have its own simple context file.
Photos, measurements, client requirements, proposal notes, decisions, changes, site updates, payment status, open issues, handover items and Whatsapp conversation history (we do everything over Whatsapp).
Not a fancy app yet. Maybe just markdown files and folders.
The important thing is that the business starts capturing context properly.
Because once the context exists, I can unleash agents and have them actually help.
Whether it’s draftnig proposals, summarizing projects, preparing client updates, tracking open issues, handover lists, whatever..
Maybe, just maybe .. it could eventually help manage the entire project flow & communications with customers.
That’s the dream!
But first, the business needs to stop leaking context everywhere.
So that’s where mint is right now.
Step one: fix execution.
Step two: speed up proposals.
Step three: capture project context.
Then start layering AI on top.
See you next month
— Yazin