What we walked into
This client is a freemium, subscription-based AI video generator app on iOS. Users turn photos and text prompts into short videos, and the business makes its money when free users start a trial and convert to a paid plan. Apple Search Ads was its main paid channel, because people searching the App Store for an AI video tool are about as close to installing as a prospect gets.
When we audited the account in early September, it had spent about $1,700 over three weeks and was producing installs at roughly $2.61 each. The structure was accidental rather than designed. Most of the budget ran through one broad-match campaign and a single exact-match keyword, "ai video generator", which was the most expensive source of installs in the account at $3.30 each. A separate campaign aimed at non-US markets was paying $3.67 per install, and only 33 in every 100 people who tapped its ads went on to install.
Just as telling was what the account wasn't doing. There was no brand keyword at all, so competitors could bid on the app's own name unopposed. Competitor app names were being picked up only by accident through broad match: some converted well, others burned spend. Proven search terms that clearly deserved their own exact-match keywords were sitting inside the broad campaign, where the account had little control over what it paid for them.
What we changed
- We rebuilt the account into four purpose-built US campaigns (Brand, Generic, Competitor and Discovery), each with its own daily budget and bid level, so every type of search was bought on purpose instead of by accident.
- We moved the search terms that were already producing installs out of broad match into exact-match keywords, so the account paid for exactly the searches it wanted and could bid on each one individually.
- We built a dedicated competitor campaign around rival AI video app names, taking what broad match had been doing by accident and turning it into a deliberate, separately budgeted channel.
- We added a brand campaign so the app's own name was protected, and cut the non-US campaign and the keywords that had been costing more than $4 per install.
- We kept a smaller broad-match Discovery campaign to find new search terms, with the Generic and Competitor terms added as exact negatives. In week three, we raised the exact-match bids above the broad ones so Apple would serve the cheaper, more precise keyword first.
What happened next
Exact-match installs got cheaper week after week. Across the Brand, Generic and Competitor campaigns, cost per install fell from $3.30 in the old exact campaign to $1.85, a 44% reduction, and the weekly figure dropped from $3.26 in the last full week before the rebuild to $1.72 two weeks after it. More of the people who tapped went on to install as well: 66 in every 100, up from 60.
The competitor campaign became the cheapest non-brand source of installs in the account at $1.70 each. The brand campaign produced 12 installs for $1.84 in total. Only about 0.3% of spend went to keywords costing more than $4 per install or producing nothing, down from 4.7% before, and spend outside the US went to zero.
Not everything moved. Cost per install across the whole account held flat ($2.61 before, $2.65 after), because the broad Discovery campaign still took about three-quarters of the budget at $3.18 per install. Its bids were sitting above the exact-match keywords on the same words, so Apple kept serving the pricier broad ad, and the exact campaigns spent less than their budgets allowed. That also meant fewer installs overall. We fixed the bid order in week three, and that day came in at $1.93 per install, the cheapest day in a month. The client then paused paid acquisition, so the fix only had a day to prove itself.
The takeaway
On Apple Search Ads, structure decides what you pay. Exact-match keywords, a dedicated competitor campaign and a protected brand term turned this app's most expensive installs into its cheapest. The lesson from week three matters just as much: when broad and exact keywords overlap, the exact keyword has to carry the higher bid. Otherwise Apple serves the broad match, and the precision you built never gets a chance to run.