Operations
Sheet 5/12 · Medium–High · 4–16 weeksDispatch runs on phone calls, stock is counted on paper, and the daily plan lives in one head.
- The morning plan is made by one person from memory and a WhatsApp group; when they are off, it stalls.
- Drivers, technicians or field staff report by voice note; someone transcribes them later, or never.
- Stock is counted on paper and reconciled at month-end; stock-outs and over-orders both happen.
- Delivery notes, job cards and checklists are photographed and lost in a chat.
- Exceptions (a broken truck, a late supplier) ripple through the day because nobody sees them early.
- Turn voice notes and photos from the field into structured records (job, site, time, quantity, issue) automatically.
- Flag exceptions early from the records it now has — late jobs, stock below threshold, missed checklists.
- Draft the daily plan from rules you set, for a human to adjust.
- Answer "where is the job / the truck / the stock" for staff over WhatsApp.
- Plan on data that does not exist. If job cards are paper and nobody types them, start with capture, not planning.
- Optimise routes or schedules well without clean addresses, times and constraints — Harare addresses are often descriptions.
- Run during an outage without a queue; field capture must work offline and sync later.
- Replace the supervisor's judgement; it gives the supervisor the facts by 07:30.
| Effort band | Medium–High · 4–16 weeks (audit first) |
|---|---|
| Data you need | Job/delivery records; stock list with thresholds; staff and vehicle lists; supplier lead times |
| Integration points | WhatsApp (field), POS/inventory, fleet trackers, accounting |
| People | Operations lead as owner; field staff as capturers; supervisor as approver |
| Measure | Records captured same day; exceptions caught before noon; stock-outs; plan time |
| Zimbabwe constraint | Offline-first capture; power and network outages; descriptive addresses; data-bundle cost for field staff |
| Code | Shape | Fit | Why |
|---|---|---|---|
| S-01 | Audit | Primary | Operations problems are usually data problems. Map the flow and count the records before anyone quotes a build. |
| S-02 | Pilot | Primary | Six weeks of field capture (voice/photo → record) on one team proves whether the data can be had. |
| S-03 | Implementation | Secondary | Exception alerts, daily plan drafts and stock thresholds built on the captured data. |
| S-04 | Integration | Secondary | POS, fleet trackers, fuel cards, supplier systems and the accounting package all hold pieces of the picture. |
| S-06 | Managed support | Later | Thresholds, routes and rules drift with the business and the season. |
- Which records do you need us to capture, by whom, on which device, and does capture work offline with a 2G signal?
- Show a voice note in Shona becoming a structured job record; show what happens when it is unclear.
- Which exceptions will be flagged in month one, and who receives them?
- How does the system handle a day when the fleet tracker or POS is unreachable?
- Who owns the tracker, POS and supplier connections, and what happens if we change any of them?
- What is the staff time per day on our side to keep the data clean?
The full bank of 60 questions, grouped by risk, is at /vendor-questions/.
Start with capture, not with planning
Every vendor wants to sell the planning layer — the dashboard, the optimiser. In most Zimbabwean operations the planning layer would run on records that do not exist yet. The order that works is: (1) capture from the field with the tools staff already use (a WhatsApp voice note, a photo of a delivery note); (2) structure it automatically; (3) flag exceptions; (4) only then draft plans. An audit (S-01) that counts how many job cards are complete on a normal Tuesday will tell you where you are on that ladder.
Offline is the design constraint
ZESA reported 138 consecutive days without load shedding by May 2026 and aims to end it by year-end, which is welcome — but rural sites, fibre cuts and data-bundle habits still mean field devices are offline for part of every day. Capture must queue on the phone and sync later; the AI processing runs when the record arrives. Ask the vendor to demonstrate airplane mode.
What “structured” means, concretely
A driver’s voice note — “Ndasvika kuMutare, delivery yaChikwanha yapera, but 3 bags dzakabvaruka” — becomes:
| Field | Value | Confidence |
|---|---|---|
| Site | Mutare | High |
| Customer | Chikwanha | Medium — confirm spelling |
| Status | Delivered | High |
| Exception | 3 bags damaged | High |
| Follow-up | Credit note / replacement | Rule-based |
The confidence column is not decoration. Anything below “High” should route to a person. If a vendor’s demo has no confidence column, ask why.