Case study • Two-sided marketplace

Refusing to scale a marketplace spend that couldn't measure new customers.

Golfnest was optimising paid Meta against a proxy metric — pixel purchases — that conflated first-time bookers with returning ones. Arctis Labs stopped the "scale or kill" debate and reframed the real question: what's the minimum data foundation a marketplace needs before paid can be trusted?

Weekly Meta spend (recent): ~1,700 THB
Pixel purchases this week vs T4W avg: 6 vs 17
Core marketplace metrics measurable: 0 of 4
Platform integrations live at audit: Meta only

Context

A marketplace where retention is the whole business.

Golfnest connects players with simulator time across partner venues. The business model isn't "acquire a customer once" — it's "build a booking habit." In categories like this, the 30-day repeat rate is the single most important number. Everything paid media does either helps that rate or quietly works against it.

Meta Ads was live at roughly 1,700 THB per week, pushing measurable traffic to the site and pixel-reported purchases back to the business. On the surface, it looked like a standard small-scale paid acquisition account. Underneath, the measurement stack was missing the pieces that make marketplace paid media defensible.

The challenge

Optimising against a proxy metric isn't optimising.

The audit found three compounding gaps that made the paid media number unreadable — and one creative signal that argued against the obvious conclusion.

No booking-platform integration.

The Meta pixel fires a purchase event on any booking, new or repeat. Without booking-platform data, there's no way to split first-time-booker CAC from returning-customer noise. For a marketplace, that's the wrong side of the entire cost equation.
No payments / Stripe integration.

Without transaction data, 90-day LTV per new customer is unmeasurable — which means CAC has no denominator. Every paid spend decision becomes a leap of faith.
No supply-side visibility.

Simulator utilisation and partner-venue active count — the supply-weighted scoreboard a marketplace actually needs — were both TBD. Paid spend was running flat regardless of where idle capacity actually sat.
But the top of funnel wasn't broken.

582 link clicks and 364 landing page views held steady week over week against T4W ratios. Pixel purchases dropped from 17/wk (T4W average) to 6/wk on a week when spend actually ticked up — arguing against creative fatigue and toward either a conversion-rate issue, a booking-availability issue, or a mix shift to returning customers the pixel wouldn't see as a fresh purchase.

The approach

Refuse the scale decision. Force the integration.

The strategic bet: don't scale, don't kill — integrate. Every agency instinct says "do something" with a dropping metric. Arctis Labs did the opposite, because acting on a proxy metric in a marketplace is how you buy a booking drop and call it marketing.

01

Hold spend flat, document the blockers

No budget increase, no budget cut. Label the week explicitly as directional-only reporting until booking-platform data lands. Resist the reflex to manufacture a decision.

02

Rebuild the scoreboard around real metrics

Replace pixel-purchase reporting with a scoreboard that tracks first-time-booker CAC, 90-day LTV per new customer, simulator utilisation %, and partner-venue active count. Every row marked TBD — deliberately visible, not hidden.

03

Make integration the unblocker

Scope booking platform + payments integration as the #1 priority ahead of creative work, ahead of scale testing, ahead of channel expansion. Nothing else moves until this moves.

04

Pre-stage the creative read

Queue Meta by-ad CPA analysis — tied to first-time bookers rather than blended purchases — so the moment booking data is available, the creative read is already structured and ready to run.

Results to date

Deliberately empty boxes as a forcing function.

A scoreboard full of TBD entries looks like failure in a recap deck. It looks like clarity in an operating review. The Arctis weekly report for Golfnest ships with empty boxes on purpose — each one a visible reminder of the integration that needs to happen before paid media can be judged honestly.

4 Core marketplace metrics formally scoped (CAC, LTV, utilisation, supply)

1 Weekly operating review cadence installed, explicitly directional-only

0 Scale-up decisions made on pixel-purchase data alone

#1 Priority ranking given to booking-platform integration vs creative/budget work

Metrics drawn from the week of 2026-04-13. Same-period-last-year paid Meta was effectively dormant (~43 THB spend, 5 clicks) — year-on-year is not a meaningful comparison.

What's next

The 90-day read.

Booking platform integrated — first-time-booker CAC and 30-day repeat rate reporting end-to-end.
Payments integrated — 90-day LTV per new customer computed; CAC finally has a denominator.
Supply-weighted spend — budget reweighted toward off-peak simulator utilisation, so paid dollars aren't displacing organic at peak.
Creative round one — Meta by-ad analysis run against first-time bookers, not pixel purchases; winners identified, losers retired.

Why this matters

The agency reflex is to recommend a scale decision. The operator reflex is to refuse it.

Marketplace paid media lives and dies on a single number: the 30-day repeat rate. If acquisition spend is buying one-off transactions, it's dilutive. If it's building a booking habit, it compounds. The Meta pixel can't tell you which.

Golfnest is the case study that proves the Arctis Labs stance is not just messaging: when the right answer is "don't scale yet," we say so, on the record, in writing, in the weekly report.

Your turn

Running a marketplace without the data to defend paid spend?

If your Meta account is reporting pixel purchases but can't tell you which ones are first-time bookers, you're optimising against a proxy metric. That's the engagement we're built for.

Get in touch