S-05 · Operations
The spreadsheet holding
it all together
is on one laptop.
The ERP, the warehouse system, the customer’s article numbers and the vendor master hold different versions of the same thing. The mapping that reconciles them is maintained by one person who is sometimes on leave.
01 · The problem
Every order that arrives as a PDF is a translation job.
A retail chain will not move to EDI for a supplier your size, so each line is mapped to an internal code from a cross-reference only one person maintains. One mis-mapped line is a wrong pack size, a return, a credit note, and a delivery miss the customer quotes back at you.
02 · The work
What the week is actually made of.
- Re-keying customer purchase orders that arrive as PDF attachments and customer-format spreadsheets.
- Mapping customer article codes to internal ones from a cross-reference only one person maintains.
- Onboarding a vendor: certificates, bank details, duplicate checks — and the expiry dates nobody tracks after the first week.
- Chasing un-closed purchase orders and open work orders that nobody cancelled, before the close.
- Rebuilding the weekly report from three exports and a column of lookups, every week, from scratch.
03 · What we build
What the work looks like here.
- A-001 Workflow automation Structured capture of every order document, in whatever shape it arrives. lead time 2–6 weeks
- B-002 Custom AI agents The mapping rules written down and applied consistently — including the exceptions that currently live in one person’s head. lead time 3–8 weeks
- C-003 Bespoke integrations One record of what exists, reconciled across the ERP, the warehouse system and the customer’s own codes. lead time 1–4 weeks
04 · Delivered
+18%
conversion · D2C retail · pan-EU
A bridge between Shopify, a 12-year-old ERP, and a custom CRM — with an AI layer that pre-empts stock crises and drafts supplier emails.
05 · Where it stops
Where it stops.
- A new mapping rule. The system proposes it; the person who has always known decides.
- A substitution the customer has not agreed to, however obvious the match looks.
- Anything the scan data could be used to rank. Throughput is a measure of a process, not of a person.
06 · Governance
Scan data becomes worker monitoring the moment it is used to allocate.
Handheld and scan records identify individuals. They are used to improve a process and never to rate the person holding the device. And where a rule is automated, the exception knowledge that made it work gets written down rather than lost with the person who held it.
07 · Next