How we work
How a Luxin engagement actually goes.
Luxin, an AI consulting firm, works in eight steps over one to three months, with a person approving every output along the way. Here is what happens, who does what, and how long each part takes.
It takes eight steps.
- 01
We start with two questions.
No survey. No audit deck. We sit down for a coffee and ask two things: what do you understand about AI, and how does the business actually run day to day. We listen first, then we look at where the repeating work lives.
- You
- Tell us how the work actually happens today, warts and all.
- We
- Listen, ask, and write down the repeating work as we hear it.
- How long
- An hour, over coffee
- 02
You watch a real system work.
Marc shows the system he runs his own company on: the morning brief, the drafted emails, the approval queue. Then he maps it onto your operations, task by task. Seeing it beats hearing about it.
- You
- Say what you would want to see happen with your own inbox and paperwork.
- We
- Show, not tell, and translate it onto what you already do.
- How long
- The same hour
- 03
We work out what the current way costs.
Your hours, your hourly rate, your numbers. Every figure has its assumption printed next to it, so you can change one and watch everything below it move. The math is written so you can run it yourself.
- You
- Agree an hourly rate you believe, and correct any number that looks wrong.
- We
- Build the cost of the current way, conservatively, and show our working.
- How long
- A day or two after the coffee
- 04
You get a plain-English proposal.
Each project fixes one thing you named, in the order that makes sense: admin first, then growth. One to three paths, each with a fixed price and the payback next to it. We say which one we would pick.
- You
- Pick a path, or tell us the numbers are wrong.
- We
- Write it so the office manager, the partner, or your accountant can read it in ten minutes.
- How long
- Usually within the week
- 05
We build one thing first.
The baby step. One task, the one that eats the most time and happens most often, built on your own accounts at a fixed price. You see it running on your own information before you decide anything larger.
- You
- Twenty minutes with the person who does the work today, and a few real samples.
- We
- Write down how the work is done as plain-English instructions, then build the system that follows them.
- How long
- One to two weeks
- 06
You approve everything.
Work comes in. The system does the part that is the same every time and stops. Drafts land in your drafts folder, dates wait for a one-tap yes, invoices hold for a signature. It is about 90% right at handover, and it climbs as you correct it, because the corrections are the training.
- You
- Read, change what you want, and say yes. Or do not.
- We
- Watch the first weeks, fix what you flag, and never let anything send on its own.
- How long
- Every day, from the first day it is live
- 07
Your information stays where it is.
Builds run inside your own Microsoft 365, Google, or accounting software, under your own logins, on commercial AI terms that never train anyone’s models. Access is scoped to the one task, so a system that drafts invoices cannot read anything else. If a build needs a server, it is one you rent, in your name.
- You
- Set up the logins with us at kickoff. Passwords go in a password manager, never in email.
- We
- Pranav reviews how every build handles your data before it goes live.
- How long
- Set up at kickoff
- 08
You own it when we are done.
The accounts stay yours. The code and the written instructions are yours. We hand over a plain-English runbook, walk your team through it, and write down what to do if something looks wrong. If Marc were unavailable tomorrow, someone else could run it from the runbook.
- You
- Decide whether you want us on call. Most small changes you can make yourself.
- We
- Thirty days of fixes included, then $100 an hour or $250 a month if you want us on call.
- How long
- One walkthrough, then it is yours
The mechanism
Nothing goes out without you.
A system watches the work coming in and does the part that is the same every time. Then it stops and waits for you. You read it, change what you want, and approve. Only then does anything move.
It gets more accurate as you correct it. The corrections are the training.
Three terms, so nothing here is jargon.
- A skill file
- is just a standard operating procedure, an SOP, written for AI: a page of plain-English instructions that says how a task is done in your business. The system follows it, and you can read and change it.
- A server you rent
- is like an apartment you rent, always on. It runs the system overnight so nobody’s laptop has to be awake. It is in your name, and the host bills you directly, at cost.
- The approval loop
- is the rule that the system drafts and a person decides. Nothing sends, deletes, or goes on the calendar until you have said yes.
What a full build looks like on a calendar.
The shape of a real five-project proposal, with the names taken out. A single pilot is the first two rows.
- Day 0
Signed agreement and deposit: two clicks, one payment link.
- Week 1
Kickoff: twenty minutes with the person who does the work today, samples in hand, logins set up together.
- Week 2
The first system is live on your accounts. You start approving real work.
- Weeks 3–6
The next projects, one at a time, each live before the next one starts.
- Week 8
Handover: runbook, team walkthrough, running costs in your own accounts. You own it.
What a kickoff needs from you.
The signed agreement and the deposit.
Twenty minutes with the person who does the work today, and one example they were happy with.
Two or three real samples: an invoice, a report, a client folder.
A login for the system, scoped to the one task, set up together at kickoff.
Tell me what’s eating the most time.
Thirty minutes, no deck. You describe how the work actually happens, and I tell you plainly whether there is something worth building.