Our rules
The rules we work by, the daily rhythm, who may do what, and how the system is built. The same for every client.
We didn't start with a theory. Most of these began as real errors.
Structural errors are refused. Judgement calls are blocked only when two signals agree — a single odd value just gets a warning.
Late entries are normal. Books close after the night shift, then are reconciled and re-checked before 8 a.m.
That counts everything twice. A closed day is undone, re-closed, and the difference carried forward.
A new price applies from its date. A fixed mistake applies to the past too, so it never returns in old reports.
Backup → test copy → verify → apply, with a named approver. A guard refuses anything unrehearsed.
Every morning
The owner's work changes from assembling and tracing numbers to deciding the few that need judgement.
Yesterday is closed after the shift; reconciliation and checks run.
Explained, fixed by procedure, or raised with your operations lead.
Summary to the owner and ops team: yesterday's numbers and what needs a decision.
Missing forms, negative stock and data edits are monitored all day.
Controls
AI does the routine. But nothing touches live data without a recorded “go” from a named person.
A trigger: an incoming message, a schedule, or a number that's off.
A trigger: an incoming message, a schedule, or a number that's off.
AI prepares a proposed action, with its reasoning.
Approved at the level you set for that action.
Carried out in your systems, and logged.
How often suggestions are accepted, plus one outcome number.
Approval is set per action. Start strict, and raise it only once suggestions prove accurate.
Slide to set how far AI is allowed to act
Prepares a draft.
Approves before it runs.
Drafting an order from a WhatsApp message.
How far you want to be consulted is agreed at the start. Silence is not approval.
A guard refuses the live apply unless the same correction was rehearsed on the test copy within the last hour.
How we choose the fix
Every Petakan finding gets an Opportunity Score, then climbs from the bottom rung. We stop at the lowest rung that actually solves the problem.
Many problems are solved on rungs 1–4. That's good news: cheaper, faster, easier to maintain.
Does this step need to exist?
A duplicate report nobody reads.
Can hand-offs and approvals be reduced?
Three signatures for a small purchase.
Are the format and rules clear?
One order template for every channel.
Rules are fixed? Let a machine run them.
Daily sales recap from every marketplace.
Needs judgement? Only then use AI — with approval.
Reading free-text order messages into a draft order.
How we work with you
What happens in each phase, what you hold at the end, and when it counts as done.
Our team walks your workflow end to end, on site, with the people who do the work — every place stock or work sits, every hand-over, every way value is lost.
One-page process map, list of loss points, improvement recommendations, estimated hours and cost to save.
Done when: The workflow fits on one page and you agree with it.
After Petakan, the process map and recommendations are yours — continue or stop there. Target: first process live in 8–10 weeks.
Map the workflow onto the same template we use for every client. Agree treatment rules, the reports you need, the baseline and the approval level.
Model document, list of forms & reports, business case — signed off by your process owner.
Done when: Every movement on the map has an event type, a rule and a written approval level.
Set you up as your own tenant, load master data with scripts that check before writing, and prepare the forms — on top of the systems you already use.
System ready, forms tested, master data you have reviewed.
Done when: You have reviewed the master data and every form passes on the test copy.
Train the team in their language, run in parallel with the old way, switch on the morning checks. The first count sets the opening balances.
Live system, a team that actually uses it, the first weekly report.
Done when: Two clean closes in a row, and the first weekly report matches your own numbers.
Every morning yesterday is closed, checked and reported. Exceptions are fixed by procedure, then turned into new rules.
Daily, weekly and monthly reports — and a system that gets safer every month.
Done when: Ongoing — then add sites, flows and reports.
Under the hood
For your technology team: why the data structure is different, and why it's possible now.
Investigating anomalies and writing corrections used to need an analyst per client. Now an agent does the routine — making a daily close economical for mid-market businesses.
Tool boundaries, scoped permissions, approval gates and full logs let an agent work on real books without blind trust.
Reconciling every day per location is now cheap — not just at month-end.
Every staff member has a phone. Forms in their language put recording where the work happens.
From true numbers to decisions
We don't bolt dashboards onto dirty data. We make the operational data true first.
FAQ
Your data stays yours. Each playbook reads only what it needs, every action is logged, and memory is visible and deletable by you.
Every phase has a decision point. After Petakan, the process map and recommendations are yours — you can stop there.
The aim is to give the team's time back from repetitive work to work that needs judgement. People stay in charge.
Two weeks, fixed price. The process map and recommendations are yours.