Gina Kwong
Mobile Growth lead at Electronic Arts (EA)
Where to hear and read about Gina Kwong1 episode or test
Gina Kwong is the Mobile Growth lead at Electronic Arts (EA) and a self-described very conservative user acquisition tester who has helped several startups set up measurement. The thing most worth learning from her is to test a new app broadly with small budgets and trustworthy tracking, and scale only once the return is confirmed.
I never advocate testing, like huge budgets in the beginning because you really just don't know.
How to Get Started with Facebook Automated App Ads
Gina Kwong: the story and the lessons
1How Gina Kwong tests a new app on Facebook Automated App Ads
Kwong treats Automated App Ads as the cheapest way to test a new product. She calls it the easiest, fastest and likely cheapest way to test. It lets her check creatives and geos quickly and see early whether users can hit a target ROI, cost per purchase or CAC.
"I think it's, it's the easiest and fastest and likely the cheapest way to test a new product."
She settles geos, platforms and the launch goal before spending. Kwong decides which geos should succeed, whether to run iOS, Android or both, and whether the goal is enough installs to check retention and engagement or users to monetize.
She clusters countries using market research the team already has, since that information shapes the whole strategy.
Kwong tests broadly, then relaunches on the audience clusters that worked. Automated App Ads and Google UAC are built for broad tests that show which audience clusters hit your KPIs. She then relaunches campaigns focused only on those clusters.
This approach suits mid-size marketers, and a new product is approached differently from an older one.
She uses it for broad strokes because it hides placement and demographics. Automated App Ads no longer shows placement and shows very little demographic data, so geo and creative are the main levers left.
Kwong takes those learnings to UAC, other ad networks or DSPs.
2Kwong's rules for budgets, creatives and optimization events
Kwong starts each campaign at a couple hundred dollars a day. She sets that budget per campaign and per platform, across one or two geos or a US versus worldwide test, and raises it only after seeing good ROAS.
Her reasoning is that you can always spend more later, so a slow start costs little. "I never advocate testing, like huge budgets in the beginning because you really just don't know."
She launches with 10 to 20 creatives and holds the rest back. The tool accepts up to 50 assets, but Kwong recommends about 10, 15 or 20 at launch. Fast, cheap installs burn through creatives, so she keeps a reserve.
Plan to refresh at least every two weeks, ideally every week.
Kwong has seen static images outperform video on Facebook. A small team can start with simple static images that clearly show what you want customers to do. Playable ads cannot be tested on Automated App Ads.
She notes that Facebook users behave differently from users on other platforms, and that static suits some products more.
She begins with install optimization and moves to purchase once there is data. A new campaign usually optimizes for installs first, or purchases if the system learns fast enough. There is no target CPA to set.
Facebook serves the ad to anyone likely to convert and moves on if they do not convert after a few impressions.
Kwong holds off on a minimum ROAS target until data exists. Value optimization takes a minimum ROAS target, such as a 10% return in the Facebook UI. A brand new campaign lacks the data to hit it.
"Now, when you're just starting a AAA campaign, you don't have enough data to do that."
3Why Kwong requires an MMP and data pipeline before spending
Kwong insists a new app has an MMP before any user acquisition spend. Events pass in real time from the internal system to the MMP and then to Facebook. With a dedicated engineer this goes faster, and without one it can take a few weeks.
She has helped several startups move to an MMP and connect their data, and notes MMPs are fairly affordable for a smaller product.
She does not trust Facebook's own dashboard on its own. Facebook is a self-attributing network, so its UI data always looks a little better than MMP data. Kwong will not name the size of the gap, only that it always exists.
"Facebook UI data will always look a little bit better than MMP data."
Kwong ties every user back to a campaign with a visualization tool. Looker, Tableau or internal queries should show new users, revenue and retention per campaign, at every level after users enter the app.
She runs a small test campaign to confirm the MMP and Facebook connect. Before real spend, Kwong launches a small campaign to check that the two systems pass data to each other.
Some teams use only the Facebook UI. She warns them of the consequences but works within the limits they set.
How to Get Started with Facebook Automated App Ads
Explains how to launch a new app on Facebook Automated App Ads (AAA): how much to spend, how many creatives to use, which optimization events to pick, and what tracking must be in place first. The payoff is a cheap, fast way to find the geos and creatives that work before committing a larger budget.
Kwong treats Automated App Ads as the cheapest way to test a new product. She calls it the easiest, fastest and likely cheapest way to test. It lets her check creatives and geos quickly and see early whether users can hit a target ROI, cost per purchase or CAC.
"I think it's, it's the easiest and fastest and likely the cheapest way to test a new product."
She settles geos, platforms and the launch goal before spending. Kwong decides which geos should succeed, whether to run iOS, Android or both, and whether the goal is enough installs to check retention and engagement or users to monetize.
She clusters countries using market research the team already has, since that information shapes the whole strategy.
Kwong tests broadly, then relaunches on the audience clusters that worked. Automated App Ads and Google UAC are built for broad tests that show which audience clusters hit your KPIs. She then relaunches campaigns focused only on those clusters.
This approach suits mid-size marketers, and a new product is approached differently from an older one.
She uses it for broad strokes because it hides placement and demographics. Automated App Ads no longer shows placement and shows very little demographic data, so geo and creative are the main levers left.
Kwong takes those learnings to UAC, other ad networks or DSPs.
Kwong starts each campaign at a couple hundred dollars a day. She sets that budget per campaign and per platform, across one or two geos or a US versus worldwide test, and raises it only after seeing good ROAS.
Her reasoning is that you can always spend more later, so a slow start costs little. "I never advocate testing, like huge budgets in the beginning because you really just don't know."
She launches with 10 to 20 creatives and holds the rest back. The tool accepts up to 50 assets, but Kwong recommends about 10, 15 or 20 at launch. Fast, cheap installs burn through creatives, so she keeps a reserve.
Plan to refresh at least every two weeks, ideally every week.
Kwong has seen static images outperform video on Facebook. A small team can start with simple static images that clearly show what you want customers to do. Playable ads cannot be tested on Automated App Ads.
She notes that Facebook users behave differently from users on other platforms, and that static suits some products more.
She begins with install optimization and moves to purchase once there is data. A new campaign usually optimizes for installs first, or purchases if the system learns fast enough. There is no target CPA to set.
Facebook serves the ad to anyone likely to convert and moves on if they do not convert after a few impressions.
Kwong holds off on a minimum ROAS target until data exists. Value optimization takes a minimum ROAS target, such as a 10% return in the Facebook UI. A brand new campaign lacks the data to hit it.
"Now, when you're just starting a AAA campaign, you don't have enough data to do that."
Kwong insists a new app has an MMP before any user acquisition spend. Events pass in real time from the internal system to the MMP and then to Facebook. With a dedicated engineer this goes faster, and without one it can take a few weeks.
She has helped several startups move to an MMP and connect their data, and notes MMPs are fairly affordable for a smaller product.
She does not trust Facebook's own dashboard on its own. Facebook is a self-attributing network, so its UI data always looks a little better than MMP data. Kwong will not name the size of the gap, only that it always exists.
"Facebook UI data will always look a little bit better than MMP data."
Kwong ties every user back to a campaign with a visualization tool. Looker, Tableau or internal queries should show new users, revenue and retention per campaign, at every level after users enter the app.
She runs a small test campaign to confirm the MMP and Facebook connect. Before real spend, Kwong launches a small campaign to check that the two systems pass data to each other.
Some teams use only the Facebook UI. She warns them of the consequences but works within the limits they set.