From 30 People to 5: How I Rebuilt My Company With AI
A year ago, my company had more than 30 people on the payroll. Today we are five. The business kept growing through the change, and we did not lose a single brand or customer relationship along the way.
I want to be careful about how that sounds. This is not a story about cutting people to make a spreadsheet look better. It is a story about a company that had grown into the wrong shape, and about a founder with no engineering background who rebuilt the way the work gets done.
Some context first. In 2016 I co-founded Fine Rituals. We take international beauty and wellness brands into Israel and Asia Pacific, handling regulatory registration, distribution, retail and e-commerce. Over the years we have worked with more than 150 brands. Before that, I was VP of Sales and Business Development, International, at Groupon, and Managing Director of Gamba, an apps company that raised Seed and Series A rounds.
I have managed large teams for most of my career. For a long time I believed, like most founders, that headcount is how you scale. This past year taught me otherwise.
The team that nearly broke the company
Growth brings work, and work brings hiring. Every new brand meant more emails, more follow-ups, more documents, more spreadsheets, and more meetings about the spreadsheets. We hired to keep up. Then we hired people to coordinate the people we had hired.
From the outside it looked like success. From the inside, the warning signs were piling up:
I spent my week inside the company instead of with our brands. Internal meetings, status updates and handovers ate the hours that should have gone to partners and new deals.
The same information lived in five places. A brand's latest status sat in someone's inbox, someone else's notes and a CRM field that nobody had updated.
Response times got slower as the team got bigger. More people meant more handoffs, and every handoff added a delay.
Payroll grew faster than the work that actually moved deals forward. We were paying for coordination, not for progress.
None of this was anyone's fault. The people were good. The structure was wrong. A distribution business runs on hundreds of repeated, careful steps, and we had assigned a person to almost every one of them.
One question for every role: a person or a system?
So I did something uncomfortable. I went through every role in the company, and every recurring task inside each role, and asked one question:
Does this need a person, or does it need a system?
To answer it honestly, I split the work into two groups.
Work that needs a person:
Negotiation, and every final decision on pricing and terms
Relationships with brand founders and retail partners
Anything that carries a signature, legal or regulatory weight
Judgment calls where the right answer depends on context no document holds
Work that needs a system:
First drafts of replies to routine emails
Follow-up reminders and keeping deal status current in the CRM
Daily and weekly reports for the team
Checking incoming contracts against our standard terms
Market research before we build a sales strategy for a new brand
Preparing first drafts of local-language product labels from a brand's artwork
When I finished the exercise, the picture was clear. A large share of our hours was going to work that followed the same steps every single time. That work did not need more people. It needed a better method, written down once and run reliably.
Building AI agents as a non-engineer
I am not an engineer. I did not hire a development agency, and I did not buy an enterprise platform. I designed and built our AI agents myself, and I run them from my own computer.
What made this possible was not technical skill. It was the thing I had spent twenty years doing as a manager: writing a clear brief.
Each agent has one job and a written playbook. I wrote those playbooks the same way I would brief a new hire on their first day: what good looks like, which mistakes to avoid, which examples to follow, and when to stop and ask. One agent drafts email replies in our voice. Another keeps our deal pipeline current and flags the deals that need a push. Another checks contracts. Another prepares research before a strategy call.
Two habits made the biggest difference.
First, when an agent gets something wrong, I fix the rule, not just the output. Correcting a single draft solves today's problem. Correcting the playbook solves it for every draft that follows. Over a few months, those small corrections built up into a system that knows how we work.
Second, I hold one rule above all others: AI drafts, a person signs. Nothing leaves the company to a brand, a customer or an authority without a person reading it and approving it. The agents do the preparation. People make the decisions and own the result. That rule is why I can trust the system, and it is why our partners can trust us.
Keeping every relationship through the change
The part I worried about most was not the technology. It was our relationships. In distribution, trust is the product. A brand gives you its name in a new market, and it needs to know that someone is looking after it.
Here is what protected those relationships:
Brands felt the change as faster answers, not fewer people. Because routine preparation now happens automatically, the people who remain spend their time on real conversations instead of admin.
One consistent voice. Every message is drafted to the same standard and reviewed by a person, so the quality brands receive is steadier than when it depended on who happened to be handling the account that week.
Nothing important fell through the gaps. Follow-ups that used to rely on someone's memory are now tracked every day. Deals stopped going quiet.
We made the human changes with respect. Reducing a team is painful, and the people who leave deserve honesty, time and care. A company that handles this badly pays for it in its reputation long after the savings show up.
Three takeaways you can apply this month
If you run a growing company and feel that headcount is rising faster than results, here is where I would start.
1. Map one week of work before you buy any tool.
Ask each person to list what they actually spent their hours on for one week: emails, reports, research, follow-ups, meetings. The map will show you, very quickly, which work is judgment and which is repetition. Most founders are surprised by how much is repetition.
2. Write the playbook before you automate anything.
AI is only as good as the instructions it follows. If you cannot explain a task clearly to a new hire, you cannot hand it to a system either. Writing the method down has a second benefit: the knowledge stays inside the company instead of leaving with whoever happened to hold it.
3. Keep a human signature on everything that leaves the company.
Let AI prepare, research and draft. Keep people on decisions, relationships and anything with your name on it. This is not a limitation of the technology. It is how you keep the trust that your business is built on.
What I would tell a founder starting now
The hardest part of this year was not building agents. It was letting go of the belief that a bigger team means a stronger company. For most of my career, headcount was the scoreboard. It turns out the better scoreboard is simpler: how fast you respond, how well you keep your promises, and how much of your own time goes to the work only you can do.
A small team with clear methods and good systems can carry far more than its size suggests. It took me twenty years and one very demanding year to learn that. I hope it takes you less.
About the author
Dana Zilberstein is co-founder and CEO of Fine Rituals, which has taken more than 150 international beauty and wellness brands into Israel and Asia Pacific since 2016. She was previously VP of Sales and Business Development, International, at Groupon, and Managing Director of Gamba, an apps company that raised Seed and Series A rounds. In one year she rebuilt Fine Rituals from more than 30 people to five, running it on AI agents she designed herself while the business kept growing. Her rule: AI drafts, a person signs. More at danazilberstein.com.