Services
List capabilities in one place and cases in another, and the reader has to connect them. So each area carries its own work directly underneath.
Start Here
Written as problem names rather than project names. Each one moves to the area that handles it.
Readiness
Whether the process is defined, and whether core operational data lives in a system of its own. These two set the starting point, and the starting point decides the first project.
| Current state | What it looks like | First project |
|---|---|---|
| Neither a defined process nor a system to hold the data | Requests scatter across email and chat; records end up in spreadsheets | Intake and history first. A shared intake queue so every request gets a number |
| A system exists, but the process was never defined | Headquarters supplied a system that does not match local work, so workarounds multiply | Define the process. Buying more software is not the answer |
| Both in place | Systems are producing data and the inputs and outputs of each step are clear | Attach automation and AI directly |
The second row is the one most often found at US subsidiaries of Korean companies. Headquarters does not know local conditions, and locally there is nobody to close the gap. That is the seat we have occupied for sixteen years.
01
Deciding what to do first. You do not need requirements written up.
Adding a tool to undefined work adds one more workaround. First we settle what goes in and what comes out, and count where and how often hands are involved. Most engagements start here, and stopping here still leaves something behind.
The stated problem was slow shipping and no visibility. The two systems of record were already exchanging data automatically; to see progress you had to look at a scheduling calendar instead, because someone changed a colour by hand at every step.
Result
Classified as a layer to remove rather than improve — evidenced by counting five manual updates per shipment.
02
Intake, approval and billing — the daily work, moved inside a system.
When the rules live in someone’s head they get recalculated every month, and a mistake is invisible until later. We move the rules into data so calculation becomes lookup, and reduce the points a person must judge to the one that needs judgement. Web screens, mobile apps and store release sit in the same scope.
Join date, contract term, sibling discounts and mid-term changes overlapped, so the amount differed for every member. The rules were absorbed into three columns rather than code branches — join date, billing date, contract end.
Result
The judgement point went from every member every month to one first cycle. Records without a payment method drop out of automatic processing and stay on a list, so nothing fails quietly.
Receipts arrived by email and the same figures were copied into spreadsheets and then accounting. The path from capture to posting was fixed at nine steps, with the validation gate placed before approval.
Result
The re-entry step is gone. While any exception is open an item cannot reach approval, so nothing already approved has to be reversed.
03
Read from the systems you already have, so the floor and the office see the same data.
Often a system already knows the state and a person is copying it across. That layer is something to remove, not improve. The first connection is read-only, and the source for each metric is fixed in writing before anything is built.
Rather than improve the manual calendar, we removed it. The systems already knew the state, so we read them directly and put the result on a floor board and an office screen. Trucks register on the timetable when they are scheduled, and a supervisor assigns the stage to prepare.
Result
Manual updates per shipment: five to none. Dock duplicate entry: three or more people on station, to one supervisor assigning stages. Five sites across three time zones aggregated on a common interval. Loading order follows the screen rather than the driver’s experience.
04
The logic stays fixed; only its authority to act moves up.
“What if it decides wrong” cannot be answered with a guarantee. It is answered with an order: the same logic goes from record-only, to confirm-after-a-delay, to immediate. Agreement does not decide when it is switched on — agreement with human decisions does.
The logic was completed first, then its authority split into three stages. Stage one records only, sitting alongside the human decisions for comparison; stage two confirms automatically after a set interval, with a window for a person to stop it.
Result
The decision to switch on is made from data. Criteria live in configuration, changes carry a delay and an effective date, and the number of records that would come out differently is counted before release.
We answer first whether this is a situation that warrants a review, or whether it is not the moment.
