Case study
Field Operations
Deploying an intelligence layer for an operation team of a $1.28 trillion USD consumer-electronics brand in Thailand
- Function
- Operations
- Industry
- Consumer electronics retail
- Company
- $1.28T Public MNC
- Coverage
- 1,400+ stores nationwide
- Team
- ~25 HQ + 1,000s field
Company profile
| area | detail |
|---|---|
| Field org | detail~25 HQ staff across data/reporting and coordination, field supervisors, and a 1,000s-strong sales/promoter force nationwide. |
| Workflows | detailSell-out tracking, market share sensing, stock and flooring, per-seller training, field feedback and follow-ups, promotion tracking, performance-drop diagnosis, and campaign rollout tracking — by store × model × week. |
| Current tools | detailVendor reporting (manual export), retailer exports, Power BI, LINE, Excel, Internal ERP. |
The challenge
Headquarters plans and executes from numbers: what is selling where, what is on display, what the field team is seeing — which stores to back, where to push stock, who needs coaching. But the data behind those decisions is fragmented and unreliable. Partner files arrive in different formats with inconsistent names across hundreds of stores; ~25 HQ staff re-type figures by hand every cycle, finish the deck the night before the presentation, and nobody can say where a number came from.
We redesigned how their operations team uses its tools: weeks on the ground mapping how the team actually works, then one central system the team runs on — so operational facts live as digital records legible to both AI and humans, instead of surviving only in a team member's memory of what happened.
Audit
An engineer embedded for 3 weeks, mapping files and processes and identified 9 core workflows with systemic patterns in every one.
| where we saw it | what it costs |
|---|---|
| Night-before reportingWeekly management deck | what it costs~25 HQ staff spending ~30% of each day stitching — finished the night before the presentation. |
| Rebuilt-from-scratch reportingMarket-sensing pack | what it costsHalf a day rebuilding from a raw export, every cycle: paste, pivot, shares, charts. |
| Mismatched inputsPartner sales files | what it costsDifferent formats + free-text product names; monthly vs daily cadence — reconciliation by hand. |
| Photos as data entryPromotion tracking | what it costsEvidence arrives as chat photos → product + price re-keyed from images, one by one. |
| Spreadsheet joinsCampaign rollout tracking | what it costsFiles joined by hand — plan, confirmations, orders/billing. |
| Numbers with no backupReported sales figures | what it costsNo second source; two reporting paths in disagreement, no way to check. |
| Hand-built dossiersPer-person training packs | what it costsRebuilt monthly by hand for a thousands-strong field team. |
| Meetings instead of answersPerformance-drop diagnosis | what it costsRun through meetings, often inconclusive — no data to settle it. |
Implementation
Month-by-month focus and results.
| focus | key results |
|---|---|
| Discovery + digital map | key resultsOnsite interviews mapped the workflows and data paths. Fifteen record types in one digital map; identity separate from labels, intentions separate from outcomes. |
| Local operations | key resultsParallel pilot went live with safe fallback. First weekly cycle with prepared figures: deck assembly down from a night-before crunch to a same-day review. |
| Data connections + reporting | key resultsEnd result is an intelligence layer anyone on the team can message — ask what is going on on the ground, get an answer from the digital map. E.g. “Which stores still have no confirmed display for the new lineup?” |
The architecture
We built a digital map of the business: real stores, sellers, products, and figures, each traceable to its source. The map mirrors how the operation actually works, so agents and people collaborate on the same shared picture.
On top of that map sits a shared workspace that runs on ops data: a team workbook and dashboard where the team works and collaborates in. The same agent answers in the company chat group too. Anyone can message the the agent and ask what is happening on the ground. The answer comes back with its source attached.
Thousands of sellers report into one place instead of being called for status, and vendor exports match to the same map instead of being re-typed — one digital map gives HQ a clear view of the ground.
Results
Every outcome the operation cares about, before and after the digital map went live: hours spent, work redone by hand, and where the numbers come from.
| outcome | before | after |
|---|---|---|
| Report preparation | before~25 staff × ~30% of each day stitching | afterPack generated from the map → 2 staff-hours review |
| Market-sensing pack | before2-day manual rebuild | afterShare + deltas computed on the map → < 1-hour review |
| Model and store mapping | beforeRe-keyed weekly | afterIdentity resolved once — codes, names, and aliases as one object |
| LINE triage | before100s - 1,000s of disorganised messages daily | afterMessages structured into observations with source + time |
| Display tracking | before10+ files → joined by hand | afterPlan vs confirmation vs billing as linked objects — gaps flagged |
| Where numbers come from | before0 sources | afterEvery figure carries source + time — ask why, get the lineage |
| Ground answers | beforeMeetings + calls to the field | afterAsk in chat → answered by reading the map |
| Follow-up ownership | beforeStatus chased across chats + memory | afterEvery follow-up named, owned, open until done |
| Training dossiers | beforePC book rebuilt monthly by hand | afterPer-seller dossiers generated from the map |
| Conflicting figures | beforeTwo paths disagree, no way to check | afterConflicts surfaced with both sources attached |
| Launch visibility | beforeLaunch state unknown until meetings | afterEvery store-model launch state visible + confirmed |
ROI
Recovered capacity converted to value at conservative loaded rates, discounted to year one and limited to directly measurable outcomes. The range reflects how sensitive each estimate is to its assumptions.
| value driver | basis | estimate |
|---|---|---|
| Reporting capacity recovered | basis~25 staff × ~30% day saved × loaded rate | estimate฿1.1–1.9M / year |
| Follow-up coordination recovered | basisStatus re-asking: daily → none | estimate฿300–500K team-year |
| Re-keying eliminated | basisWeekly reformatting → one-time mappings | estimate฿100–200K |
| Crunch and overtime avoided | basisNight-before assembly: every cycle → none | estimate฿100–150K |
| Error exposure reduced | basisBlind trust → sourced figures + spot-checks | estimate฿150–200K |
| Conservative total, year one | basis฿400–600/hour loaded × year-one discount × measurable only. Capacity value, not cash. | estimate฿2.1–3.0M |
Deploy custom AI in your business
We'll discuss how your company works today and the ways we can help you work faster and better.