REBUILT SEARCH AS APP CAMPAIGNS
The legacy search campaigns became app campaigns, with far more inventory to buy from and no gender setting. The conversion had to carry the targeting.
FIELD REPORT 07 · APP ACQUISITION · MATCH GROUP
The female
profile event
App campaigns have no gender targeting, so I built it out of a conversion event. It fired only when a woman created a profile. Pointed at that one event, the campaigns learned the exact funnel that mattered.
45%lower
CPA
CPA
8xthe
scale
scale
01 THE BRIEF
The goal was more women. App campaigns cannot target by gender.
At Match Group I owned app and web acquisition across Google, Meta and Apple Search Ads, roughly $2,000,000 a month in spend. For the female focused dating app portfolio, the goal was simple to say: acquire more women.
The legacy targeting was underdelivering, and subscription CPAs were rising across the industry. App campaigns could reach far more people than our search campaigns did, and app campaigns offer no gender targeting.
Women are the scarce side of a dating marketplace. How many are on an app shapes what everyone else finds there, and how many people stay long enough to pay.
COMPANYMatch Group
(S&P 500)
(S&P 500)
ROLESenior Digital
Marketing Manager
Marketing Manager
SPEND~$2,000,000
a month
a month
WHENNov 2019 to
Mar 2024
Mar 2024
The bet: if the only conversion the campaigns ever heard about was a woman creating a profile, they would learn to find women without a targeting setting.
02 THE MECHANISM
The algorithm finds more of whoever fires the event you count.
Pick the event the campaign optimizes toward and watch who it finds in 120 installs. The female profile end carries the real result. Everything else is a model.
● WHAT THE ALGORITHM LEARNSA MODEL · 120 INSTALLS PER MODE
OPTIMIZE ON
- A WOMAN WHO CREATED A PROFILE
- EVERYONE ELSE
- COUNTED AS A CONVERSION
THE MATHCost per install 1.00 × 120 installs ÷ 12 female profiles = 10.00 per female profile. This is the model's baseline, index 100.
COUNTS AS A WINAny installEvery install is a conversion, from anyone. The campaign learns to find cheap installs.
CPA · COST PER FEMALE PROFILE100BASELINEIndex. Optimizing on installs is the model's baseline.
SCALE1xBASELINEThe model's starting point.
A model of the mechanism, not campaign data. The female profile end matches the real result: 45% lower CPA at 8x the scale, measured against the search campaigns the app campaigns replaced. The install and registration modes, the people and the math are illustrative.
03 THE EVENT
Two checks, one conversion.
Part one, a profile is created. Part two, the profile's gender is woman. When both are true the event fires once for that user and travels through the MMP to the ad networks.
●EVENT SPECAPP · MMP · AD NETWORKS
- eventprofile_created_woman
- part_1profile_created// profile saved and complete
- part_2profile.gender == "woman"// read from the saved profile
- firesonce per user// the first time both parts are true
- dedupeuser_id// repeats, edits and replays dropped
- routeapp > MMP > ad networks
- goalthe only optimization event// on every campaign that moved over
ONCE PER USER
A woman who edits her profile is still one customer. Counting her twice would teach the campaign to overpay for her.
THE PROFILE STEP
A registration says little about who signed up. The saved profile says who she is.
ONE GOAL
Install and registration goals stayed off these campaigns, so nothing diluted what they learned.
04 WHAT I DID
I proved it on a slice of budget, then moved the campaigns.
BUILT THE TWO PART EVENT
Profile created, and the profile's gender is woman. The app fires it once per user, the first time both are true, and it is the conversion the campaigns optimize toward.
CHECKED THE PLUMBING
Before budget moved, the event count in the MMP was reconciled against the app's own profile data. Duplicates were dropped before the postback, so the ad networks learned from the real number.
PROVED IT ON A SLICE
A slice of budget ran on the new event first, read against the legacy setup on cost per female profile.
MOVED THE CAMPAIGNS OVER
Once the slice held, campaigns moved across and budgets stepped up while the cost held. The read went past the event: whether those women went on to subscribe, and whether any traffic fired the event without behaving like a real user.
PROVEA slice of budget runs on the event.
MOVECampaigns cross over as the cost holds.
SCALE8x the scale at 45% lower CPA.
APP CAMPAIGNSCUSTOM APP EVENTMMPPOSTBACKSCOHORT READS
05 THE RESULT
Cheaper and bigger at the same time.
Scaling usually pushes CPA up. This went the other way: 45% lower CPA at 8x the scale, and the portfolio recorded its first directly positive ROI.
45%lower CPA
8xthe scale
$5to$18CPA across the female app portfolio, depending on scale. Women cost more to acquire.
1stdirectly positive ROI the portfolio had recorded
● BEFORE AND AFTERCPA INDEX AGAINST SCALE · OLD SETUP = 100 AT 1x
- SEARCH CAMPAIGNS, THE OLD SETUP
- APP CAMPAIGNS ON THE FEMALE PROFILE EVENT
- OLD SETUP CPA LEVEL
Both points are real, shown as indexes. Scaling a setup usually costs more per customer as it reaches past its best audience. The event let the campaigns reach further and pay less per woman.
Figures are from my own reporting at the time. The 45% and 8x compare the app campaigns on the new event with the search setup they replaced. The $5 to $18 range covers the female app portfolio at different levels of scale. Raw volumes stayed with Match Group.
06 WHAT CARRIES OVER
If the platform will not target your customer, make your customer the conversion.
01
Write the customer into the event. Here it was gender. Elsewhere it is plan type, company size or a second order: whatever field in your own data marks the customer.
02
Deeper events fire less often. Check that each campaign will see enough of them to learn before budget moves onto one.
03
Treat the event like revenue. Dedupe it, reconcile it against your own data, and watch for traffic that fires it without acting like a customer.
04
Grade past the event. A cheap conversion that never pays is a cost, so the read runs all the way to subscription.