Portfolio

Brand and growth marketing in Singapore. A national trading campaign measured against matched controls, four AI pipelines built with Claude Code, and a three-day CBD roadshow run end-to-end.

A city skyline at night, a tall lit communications tower rising above a harbour bridge

Nathan Wong · Brand & growth marketing · Singapore

I run campaigns and build the tooling that measures them.

// campaigns · measurement pipelines · MAS-compliant activations

A national trading campaign, a three-day product activation in the CBD, and the automation and analytics behind both. The figures below are masked, the method is not.

National trading campaign +53.7pts Percentage points by which registrants beat matched controls on trades per day, a difference-in-differences result that held under every robustness test.
CBD activation · 3 days ~8 in 10 App downloads during the roadshow that converted straight to a new investment portfolio.
AI operations 4 Shipped pipelines in a quarter, landing on C-suite and Head-of-PR desks.

Case 01 · National trading campaign

The part most marketers skip, proving it wasn't just a good month.

  • Campaign website
  • Daily draw tool
  • Ops automation
  • Prize fulfilment
  • Impact measurement

A months-long registration-and-trading campaign tied to a major sports season. Clients opted in and were rewarded for trading activity. I worked across the whole lifecycle, helped build the opt-in website, turned a colleague's rough internal tool into the live on-brand draw device, built the Python automation that ran the daily draws and got winner notifications out on time, coordinated getting the jerseys to winners, and built the measurement pipeline that answered the only question leadership cares about, did it work, and how do we know?

~2 in 3
of everyone who registered went on to actually trade.
+65%
more trades per day for opted-in clients against their own three-month pre-campaign baseline.
+12%
the rise for matched lookalike clients over the same window, so the market wasn't carrying it.
Difference-in-differences · trades per day vs own baseline

Registrants against a 1:1 matched control

Every opted-in trader paired with a lookalike non-registrant, same investor tier, near-identical prior trading, drawn from the full client pool. Each group measured against its own baseline first, then group against group.

Registrantsopted into the campaign
+65%
Matched controllookalike non-registrants
+12%
scale 0% to +70%, change in trades per day against each group's own 3-month baseline
+53.7 pts the registrant lead on trades per day, and +51.0 pts on traded value. The true campaign effect, isolated from anything the whole market was doing.
  • The gap held under every robustness test, including stripping each side's single largest whale account.
  • The typical control client slightly cut back over the same window.
  • So the lift was specific to people in the campaign, not the market being busy.

Why you can trust the numbers

I didn't run the analysis once and read off a result. I built the measurement as a reproducible system, every figure rebuilt from the source exports the same way each time, then verified it myself, figure by figure. Just as important, I designed the baseline to catch its own errors before they became conclusions, so the checks for a bad row, an overlapping group, or a number that doesn't reconcile all run up front, and a mistake surfaces early instead of quietly ending up in a headline.

Stated plainly, it is a matched observational study rather than a randomised trial, and registrants self-selected.

Case 02 · The tooling behind the campaigns

I don't wait on engineering or agencies for ops and intelligence, I build them.

Four pipelines built and run with Claude Code, landing in front of leadership, C-suite and the Head of PR.

A · Automation

Hands-off entry automation

The daily campaign entry process used to be manual file-wrangling. I built an end-to-end pipeline that pulls the day's web-form submissions, fetches the matching same-day dashboard export straight from the inbox, sorts everything into the right folder, runs the processing tool, and appends the clean result to a shared sheet, untouched by human hands, handling login credentials safely so nothing sensitive is ever exposed.

Outcome: a daily ops task that ate time now runs itself and cannot fat-finger a file.
B · Intelligence

Competitor ad intelligence

A pipeline that pulls competitors' live ad creatives from the official ad library, a stable source rather than brittle scraping, reads the text and video inside each one, and uses AI to tag its marketing angle, across a corpus of roughly 800 ads. Re-tagging materially corrected the analysis, revealing that a category we had read as open whitespace was actually a competitor's home turf. Packaged as a dashboard a non-technical colleague can run.

Outcome: competitor positioning read from real ad data instead of guesswork.
C · Intelligence

Brand perception intelligence

Rebuilt a brand-perception pipeline with Claude Code after upstream API changes broke the original integration, pulling from Sprout Social, Meltwater and app-store reviews and classifying sentiment consistently so results stay comparable run to run. It drafts a near-finished report in a single pass; I verify every thread by hand before it goes out fortnightly to the heads of investing and PR, who decide what we act on, and from there to the C-suite.

Outcome: leadership gets a regular brand-health read without a paid social-listening seat.
D · Modelling

Competitive share-of-voice

A weighted share-of-voice model comparing our media coverage against key competitors across a news and podcast catalogue, with coverage scored by publication quality, delivered to the Head of PR and Comms.

Outcome: competitive PR positioning quantified instead of eyeballed.
Plus · Regulatory rigour

Ran MAS-compliant campaign workflows end-to-end, from KOL disclosure frameworks to advertising assessments, so every activation met financial-services marketing regulations. Not a tool, the competency that makes all of the above shippable in a regulated market.

Case 03 · Physical roadshow · Raffles Place

Vendors, staff, merch, and a booth in the CBD for three days straight.

A three-day product-launch roadshow (27–29 April 2026) in the heart of Raffles Place, co-funded with a global asset-management partner. A guaranteed-win booth challenge was tied to mandatory product sign-up, backed by tiered rewards and thousands of flyers across CBD lunch spots.

What I owned, the operational execution layer that made the event actually happen:

  • VenueBooking and signed licensee lease for the CBD site.
  • VendorsAV, power upgrade, merch supplier and partner printing, all coordinated.
  • MerchSourcing, purchase orders, and delivery logistics for umbrellas, cups and fans.
  • StaffingHired and briefed 10 part-time promoters, taking hiring over directly when the original owner left mid-project.
  • FlowBooth script, part-timer briefing, and event timeline, planned jointly with a teammate.
  • OnsiteThere setup-through-teardown all three days, handling real-time issues and keeping the floor running.
~8 in 10
app downloads that converted to a new investment portfolio, a high booth-to-signup rate.
100s
new sign-ups and downloads over the three days.
6-fig
(USD) in new client funding driven during the activation.
rated
Google Play 3.6→4.2 and Apple 4.4→4.5 over the same period. Company-wide app ratings, context rather than a result I attribute to the booth.
Honest framing

These are event-level results, and they depended on the partner's co-funding and on the product and positioning work owned by teammates, not on execution alone. My defensible claim is the execution that converted foot traffic into sign-ups, the venue, vendor, merch, staffing and flow that turned a plan into a booth that ran for three days.

The method behind the numbers

01

Matched control

Every result compared against lookalike clients, not just before against after.

02

Difference-in-differences

Each group measured against its own baseline first, then against each other.

03

Whale-proof

Headline results survive removing the single largest account on each side.

Figures are rounded and generalised to protect commercially sensitive data. No client personal information is shown. Company and partner names withheld, available on request.