Three case files from one engagement

Building a paid acquisition engine: ~6× paid volume at a held cost per user

A GCC fintech app had strong organic traction and thin paid acquisition: roughly 8 in 10 installs arrived with no paid media behind them, from store search, word of mouth and referral. The mandate was to build a genuine paid engine, in a small market, without letting the cost of a real user run away.

There was a harder problem underneath the brief: the measurement layer could not support the mandate. iOS conversion mapping was not aligned to the funnel, ad networks were feeding the platforms noisy or irrelevant events, live campaigns showed zero ad spend inside the MMP, and influencer and offline activity was estimated from reach rather than measured.

They scale spend first and fix measurement "later." Budget goes into channels whose true cost per real user nobody can compute, the platforms optimise toward whatever miswired event they are being fed, and six months later the blended dashboard cannot answer the only question that matters: which channel brings users worth keeping.

The other classic: judging the new paid operation against the old organic-heavy blend. A low blended cost built on free volume is not a benchmark for a scaled paid engine. It is a different product.

Measurement first, roughly two to three months of foundational work in parallel with early spend:

• Rebuilt the iOS SKAN conversion architecture around the funnel that matters (registration, onboarding, the bottom-of-funnel value event), restoring usable iPhone measurement under Apple's privacy rules.
• Cleaned and re-mapped every ad network's event signal. One major platform was being fed a large volume of irrelevant technical events that actively degraded its optimisation; each partner was rebuilt to optimise toward real business outcomes.
• Reconnected the cost pipelines so spend, CPA and ROI became computable cross-channel inside the MMP, and set explicit source-of-truth rules per channel to end unlike-for-like comparisons.
• Built the funnel-stage audience framework and unique tracking links for every influencer and offline activation, converting reach-estimated activity into measured install-to-activation funnels.

Then the engine: a multi-channel build across Meta, Google and TikTok with objective discipline (buy the event, not the install), language-split creative (Arabic and English read separately), a weekly kill-or-keep cadence with pre-committed criteria, and channel expansion once the core was measurable.

The objective switch: same audience, ~40% cheaper real users

The same GCC fintech needed more registered, onboarded users, not more downloads. Yet significant budget sat in install-optimised campaigns across channels, because installs are the metric everyone can see and the volume the platforms most happily deliver.

Two things. They optimise to installs long after they have the event volume to optimise to what actually matters. And when they do switch, they keep judging the new campaigns on cost per install, see a "worse" number, and switch back. The metric has to change with the objective or the test is rigged against itself.

• Named the true KPI per channel: a completed registration or onboard, not an install.
• Re-pointed campaigns at the event wherever weekly event volume could feed the algorithm (roughly 50 per week as the working threshold), keeping install-optimised campaigns only where they belong: new channels and signal-starved surfaces, as seed capital.
• Used the bridge campaign type where it exists (starts on installs, hands optimisation to the in-app event once signal accumulates).
• Enforced one reporting rule everywhere: judge each campaign only on the cost-per-result column that matches its objective.
• Re-tested creative per objective rather than assuming winners travel.

Finding the ceiling: scaling 20% a week until the market said stop

A winning conversion campaign for the same GCC fintech justified aggressive scaling, and got it: budget increased roughly 20% per week, compounding, for three consecutive weeks. The market it runs in is small, with a finite pool of high-intent users.

They treat a scaling rule as a perpetual lever. "It worked three times" becomes "it always works." And because campaign dashboards do not alarm on audience saturation, the first warning most teams get is a cost chart that has already doubled.

• Recognised the signature in the data: ad frequency had climbed to around 50 while cost per onboarded user roughly doubled. Past saturation, incremental budget buys repeat impressions and self-inflated auction prices, not new users.
• Pulled budget back roughly 30% to the last efficient level, then held it deliberately still for two to three weeks. No structural rebuild, no learning resets.
• Refreshed creative to reset attention, and moved incremental growth to surfaces with frequency headroom instead.
• Made the frequency check a standing pre-condition for every scale decision, and paired it with marginal-cost reads: on another channel, the bid simulator priced the next users at roughly 3× the account average. Averages describe the past; margins price the next dollar.