// Scenario 06
An AI back office, signed off by a person
Not a chatbot and not “AI does everything”. A small team of AI specialists, each with one job, that prepares the work, and a person who approves anything a client will see.
- For
- Founders and small teams doing everything
- Already running in
- Casagros · Saturn Laboratory · SumoAnalysis · Naba Realty
- Scenario
- This is a scenario we deliver, not a past client result.
Why it matters in 2026
By 2026 most small companies have tried AI and hit the same wall: a general chatbot is helpful for one task at a time, and a fully automatic one eventually says something wrong to a client. What works is narrower: one specialist for replies, one for quotes, one for catalogue updates, one for reports, each with clear rules, your data and a human approval step. That is how we run our own studio, and it is the setup most small teams can afford instead of a new hire.
You need this if
- You or your team spend hours a week on replies, quotes, listings or reports that follow the same pattern.
- You tried a chatbot, but nobody trusts it with clients.
- Work piles up when one person is away.
- You know what a good answer looks like, but writing it takes time you do not have.
How we deliver it
1 Pick the jobs
We watch how work actually flows and choose the two or three repetitive jobs where a draft saves the most time and a mistake is easy to catch.
2 Build one specialist per job
Each specialist gets its rules, your tone, the data it may use and the data it must not touch. It prepares drafts; it never sends on its own.
3 Put a person in the loop
Drafts land in one queue. A person approves, edits or rejects each one, and every decision is logged, so you always know who sent what.
4 Improve every week
Edits and rejections become better rules. We track time saved and error rates, and add a new specialist only when the first ones earn their place.
What you get
- Two or three AI specialists, each for one repetitive job
- One approval queue for drafts, with a log of every decision
- Clear rules on what each specialist can and cannot do
- Weekly numbers on time saved and corrections made
- Documentation your team can run without us
What we will not promise
We will not connect an AI that sends messages, moves money or changes prices without a person approving it. Speed is not worth a mistake in front of a client.
Questions
Which AI models do you use?
We choose per job, among the leading providers, and can change models later without rebuilding the setup. Your data is used to do your work, not to train public models.
Does our data leave our tools?
Only what each job needs. We document which data each specialist sees, and keep sensitive data out where we can.
Is this the same as the OG engine you mention on this site?
Yes, it is the same approach: specialists with one job each, and a founder approving every deliverable. We build the same setup inside your company.