Stefanus Wong — Operating Record

01 — THE LEDGER

Every claim
with a number
attached to it.

FIG. 1.1 — Sixteen years, plottedFive tenures · four markets · one continuous record

Ordered by size of commercial consequence, not by recency. Currencies as reported at the time. Every line is defensible in a reference call.

Intervention	Result	Where
Proprietary premium ad network, built and scaled	SGD $100K in 30 days → $2M / yr	SPH
Cross-market search advertising strategy	SGD $22M over 3 yrs · +22% YoY	Yahoo
Data Performance department, built from zero	HKD $600K net-new, year one	DDB HK
First-party data & segmentation, HKD $2M+ budget	ROI +120% · CPR −60% in 2 qtrs	Jardines
Reporting pipeline rebuild	7 days → 3 days (−57%)	DDB HK
KPI re-engineering, cross-functional conflict	Revenue +150% · SLA 5d → 3d	Yahoo India
Zero-budget organic strategy, 40+ markets	50K → 3.2M (+3,100%)	Greenpeace
Print-to-digital migration, eight luxury titles	8 of 8 live · on time · on budget	SPH
B2B demand generation overhaul	0.3% → 1.0% conversion, 12 mths	New Energy Nexus
Enterprise recruitment programme, 4 markets	Sign-up conversion +120% / 3 yrs	Jardines
Multi-agent automation of daily commerce ops	Near-zero marginal cost per task	WONGSO

02 — ARTIFACT I · THE ARCHITECTURE BLUEPRINT

An org chart
for software
that works
while I sleep.

WONGSO is a real company — Hong Kong entity, Indonesian supply chain, 190+ SKUs, four currency markets. Headcount: one.

The system below is what makes that arithmetic work. Three engineering decisions moved it from demo to production: cost tiering, retry ceilings, and a governance gate that will not let an agent do anything irreversible without me.

Most AI marketing stories are pilots. This one carries a P&L, an inventory position and regulatory exposure — which is why it has an audit trail.

FIG. 2.1 — MULTI-AGENT OPERATING TOPOLOGY WONGSO.STORE · IN PRODUCTION

FIG. 2.2 — SUB-AGENT OUTPUT CONTRACTEXCERPT

<!-- rejected at parse if any node is absent -->
<agent_return id="catalog-compliance" run="0714-02">
  <status>complete</status>
  <tier>amber</tier>
  <scope>190 SKU · GMC feed US/SG/AU/UK</scope>
  <findings>
    <item sev="2">14 SKU miscategorised cat-3</item>
    <item sev="1">GTIN absent, 6 SKU</item>
  </findings>
  <actions_taken>recategorised 14 · feed resubmit</actions_taken>
  <escalate>false</escalate>
  <cost model="haiku" tokens="18.4k"/>
</agent_return>

Structured output is not a formatting preference. It is the reason an agent's work can be audited, costed and rejected — the same reason finance will not accept a narrative in place of a line item.

FIG. 2.3 — ESCALATION LOGEXTRACT

Tier	Event	Disposition
AUTO	Meta title rewrite, 41 product pages	Shipped. Logged. Reversible in one commit.
AUTO	Currency feed refresh, AU + UK	Shipped. Diff attached to log.
NOTIFY	14 SKU recategorised, GMC cat-3 → cat-1	Acted, then reported. Reviewed same day.
NOTIFY	Pinterest catalog re-sync, 109 items	Acted. Approval confirmed downstream.
HUMAN	Merchant compliance review, account-level	Halted. Audit, remediation, staged resubmit — by hand.
HUMAN	Sales channel closure (permanent policy block)	Halted. Fix-vs-pivot priced. Pivot approved.
HUMAN	Any price change, any market	Standing rule. No agent touches price.

Tier is assigned by reversibility and blast radius, never by the agent's confidence. An agent that is certain about an irreversible action is precisely the case the gate exists for.

What transfers

The store is not the asset. The asset is the governance layer, the unit-cost model and the failure-mode map — the three things every enterprise AI programme discovers it needs about eighteen months after the pilot. I have already paid for that learning with my own capital, which means it arrives at your organisation pre-tested rather than theoretical.

03 — ARTIFACT II · TURNAROUND CASE FILES

Three systems
that were
broken before
I arrived.

Each file runs the same four movements: the failure state, the diagnosis, the intervention, and what showed up in the P&L. No narrative without a number at the end of it.

Mandate

Eight luxury print titles, no digital presence, and a newsroom that did not want one.

Singapore Press Holdings · 2013–2015 · Digital Operations & Key Partnership Manager

Before / after

FIG. 3.1 — SPH, 24 monthsOn time · within budget

Failure state

Harper's Bazaar, L'Officiel and six sister titles were generating declining print ad revenue with no digital inventory to sell against. The newsroom read "digital transformation" as a threat to craft. Previous attempts had been announced from the top and quietly ignored.

Diagnosis

Two failures stacked. Commercially, every proposed launch plan assumed traffic would arrive first and revenue second — eighteen months of cost before a single dollar. Culturally, the migration was being sold to journalists as a mandate rather than a capability, so the most senior and most influential staff were the most resistant.

Intervention

Three tracks, deliberately sequenced. Technical: I built the SEO architecture myself — sitemaps, site trees, crawl governance — so all eight properties launched search-ready instead of retrofitted. Human: I designed capability training pitched at print craft and converted the loudest sceptics first, so adoption spread laterally rather than downward. Commercial: rather than wait for organic traffic, I architected APEX — a premium ad network pooling our luxury inventory with external publishers including The Diplomat — so the new properties carried a revenue line from launch week.

P&L result

8 of 8 properties live, on time, within budget, 24 months
SGD $100K APEX revenue in the first 30 days
SGD $2M APEX annual run rate at scale
+200% LuxuryInsider.com organic traffic in 12 months
8 publications of editorial workflow modernised in year one

Transferable

Transformation fails on culture, not technology. Convert the sceptics first, and give the new platform a P&L from day one so it survives the second budget cycle.

Mandate

A creative agency flying blind on data, losing pitches to competitors who were not.

DDB Hong Kong · 2021–2022 · Data Performance Lead

Before / after

FIG. 3.3 — DDB Hong Kong, first yearDepartment built from zero

Failure state

Campaign reporting took seven days — arriving too late to change anything it described. Data sat at the end of the process as a post-mortem. Pitches were being lost to data-literate competitors, and there was no internal capability to build against.

Diagnosis

The brief was to build a data function, but building it as an internal cost centre would have guaranteed its death in the first downturn. The real problem was position, not tooling: data was being consumed at the wrong end of the workflow and had no commercial constituency inside the agency to defend it.

Intervention

I rebuilt the reporting pipeline end to end on cost-effective analytics tooling, then moved data to the front of the creative process instead of the post-mortem. Then I made it commercial — packaged the capability as a billable service and took it to flagship clients: Jardines, Allianz Global Investors, Manulife. For Jardines I ran first-party data analysis and audience segmentation across a media budget above HKD $2M annually. And I trained the entire agency rather than only my team, so data literacy became house style rather than a department.

P&L result

HKD $600K net-new agency revenue, year one
−57% reporting turnaround (7 days → 3) in the first quarter
+120% Jardines portfolio ROI · −60% Facebook cost-per-result, two quarters
+4% MoM sustained organic traffic, Allianz Global Investors
80% new-business win rate on data-backed proposals
Jardines and Allianz later retained me independently

Transferable

Do not build reporting systems. Build decision systems, then make them a profit centre — a capability that bills is a capability that survives the next restructure.

04 — ARTIFACT III · TECHOPS & DATA PIPELINE

Four currencies,
one catalog,
no operations
team.

FIG. 4.0 — Cross-border feed topologyOne catalog · five destinations · zero operations headcount

Cross-border commerce fails in the plumbing: feed rejections, currency drift, category misclassification, stock ghosting between markets. The stack exists to make those failures visible before a customer finds them.

Market	Currency	Feed surface	Localisation handled
United States	USD	Google Merchant Center	Pricing · tax display · sizing · cat 1–4 taxonomy
Singapore	SGD	Google Merchant Center	Pricing · GST treatment · shipping bands
Australia	AUD	Google Merchant Center	Pricing · GST · apparel sizing · returns terms
United Kingdom + EU	GBP	Google Merchant Center	Pricing · VAT · EU compliance categorisation
Social commerce	Multi	Pinterest Catalogs	Domain claim · 109-item approved catalog · ad approval

WONGSO product photography — **PLATE II** — Catalog photography, one of 190+Feed-compliant across four markets

Storefront

Shopify Liquid — mobile-first template work, dynamic upsell logic, JSON-LD schema, technical SEO built in rather than bolted on.

Inventory backend

AppSheet — custom database routing cross-border stock between Indonesia and Hong Kong. 0% stock discrepancy across O2O retail.

Feed layer

GMC · Pinterest · Meta Commerce — four-currency product feeds with automated compliance categorisation across 190+ SKUs.

Retention

Klaviyo — lifecycle flows and an affiliate programme, contributing to 25% YoY revenue uplift.

Measurement

GA4 · Looker Studio · SEMrush — one reporting surface, built so an agent can read it as reliably as I can.

Knowledge layer

Living state docs · SOPs · session logs — any agent or human resumes any workstream cold in minutes. The same discipline as agency handover design.

FIG. 4.1 — INCIDENT · SUNK-COST DISCIPLINECLOSED IN 5 DAYS

02 Jul — Incumbent social commerce channel returns a permanent regional policy block. Not a bug; a jurisdiction exclusion, no appeal path.

03 Jul — Fix-versus-pivot priced. Cost of pursuing an exception: indefinite, with no probability estimate available. Decision: kill it.

04–06 Jul — Catalog re-cut for an alternative surface. Domain claim filed, 109 items submitted.

07 Jul — Replacement channel live and ad-approved. Total elapsed: five days.

Included deliberately. A portfolio without a failure in it is a brochure.

FIG. 4.2 — COMPLIANCE PROGRAMMERESOLVED

A platform-level merchant compliance review is an account-existential event: feeds suspended, paid surfaces dark, revenue at zero until cleared.

Method — systematic audit of all 190+ SKUs against the platform taxonomy, product-data remediation at source rather than at feed level, then staged re-submission in tranches so a second rejection could be isolated to a specific batch instead of restarting the clock.

Governance — tier-red throughout. No agent action; human sign-off on every submission.

Platform compliance is programme management, not a support ticket.

05 — THE PRACTICE

How I engage,
and what
it costs you
to find out.

Stefanus Wong — **STEFANUS WONG**Hong Kong

Clients to date: Jardines · Allianz Global Investors · Greenpeace Southeast Asia · New Energy Nexus · Mr. Ripley · Magifarm.

Available for permanent leadership mandates in Hong Kong, Singapore and Bangkok, and for the engagement modes opposite on a remote or hybrid basis.

MODE 01

Growth diagnostic — fixed scope, fixed fee

Two to three weeks. I take your funnel, feed infrastructure, martech stack and reporting pipeline apart, and hand back a prioritised intervention list with cost, effort and expected commercial impact against each line. You keep the document whether or not you keep me. Most engagements should start here; several should end here.

MODE 02

AI operations audit & governance build

For teams already spending on AI tooling without a decision-rights framework, a cost model or a failure-mode map. I map what your agents and automations are actually permitted to do, price what they cost per task, install the escalation tiers, and write the SOPs your team will still be running twelve months later. Almost nobody offers this work, because almost nobody has run it in production.

MODE 03

Fractional Head of Digital Growth

Retained, two to three days a week. I own the growth number, build the operating cadence, hire and train the permanent team, and design my own redundancy. Suited to companies that need director-level judgement before they can justify a director-level salary.

MODE 04

Campaign & launch systems — D2C and crowdfunding

End-to-end launch infrastructure for product businesses: Shopify build and conversion logic, Klaviyo lifecycle and reactivation, multi-market feed setup, campaign asset systems, and the reusable template library that stops every launch starting from zero. Built on my own catalog first, then on yours.

Sixteen years
of evidence,
not adjectives.

Every claim
with a number
attached to it.

An org chart
for software
that works
while I sleep.

Most AI marketing stories are pilots. This one carries a P&L, an inventory position and regulatory exposure — which is why it has an audit trail.

Three systems
that were
broken before
I arrived.

Eight luxury print titles, no digital presence, and a newsroom that did not want one.

A regional office in open conflict, and no authority over either side of it.

A creative agency flying blind on data, losing pitches to competitors who were not.

Four currencies,
one catalog,
no operations
team.

Storefront

Inventory backend

Feed layer

Retention

Measurement

Knowledge layer

How I engage,
and what
it costs you
to find out.

Growth diagnostic — fixed scope, fixed fee

AI operations audit & governance build

Fractional Head of Digital Growth

Campaign & launch systems — D2C and crowdfunding

Every claimwith a numberattached to it.

An org chartfor softwarethat workswhile I sleep.

Most AI marketing stories are pilots. This one carries a P&L, an inventory position and regulatory exposure — which is why it has an audit trail.

Three systemsthat werebroken beforeI arrived.

Eight luxury print titles, no digital presence, and a newsroom that did not want one.

A regional office in open conflict, and no authority over either side of it.

A creative agency flying blind on data, losing pitches to competitors who were not.

Four currencies,one catalog,no operationsteam.

Storefront

Inventory backend

Feed layer

Retention

Measurement

Knowledge layer

How I engage,and whatit costs youto find out.

Growth diagnostic — fixed scope, fixed fee

AI operations audit & governance build

Fractional Head of Digital Growth

Campaign & launch systems — D2C and crowdfunding

Every claim
with a number
attached to it.

An org chart
for software
that works
while I sleep.

Three systems
that were
broken before
I arrived.

Four currencies,
one catalog,
no operations
team.

How I engage,
and what
it costs you
to find out.