Vivek Girotra
VP of Marketing at Times Internet
Where to hear and read about Vivek Girotra1 episode or test
Vivek Girotra is VP of Marketing at Times Internet and runs Facebook and Google app install campaigns, including for the Abound cash back and rewards app. The lesson most worth taking is to start at the install, walk down the funnel, and steer the algorithm with broad targeting, selective creative and high-quality feedback signals.
you always want to be slowly walking down the funnel
The Simplest Facebook Ads Strategy for App Installs
Vivek Girotra: the story and the lessons
1How Vivek Girotra steers the Facebook algorithm
Girotra nudges the algorithm from the side and never controls it directly. He compares it to sweeping a curling stone: you guide it but never touch it. The three levers are targeting, creative and feedback signals.
Each lever corrects the algorithm's course after the campaign has started.
He starts every campaign on the install, then walks down the funnel. He does not optimize for something like level 10 of a game on day one, because that starves the campaign of data. The path runs from install to a level 2 or trial start, then to subscription.
Along the way he gathers data and fixes the product or onboarding before asking for rarer events.
He judges campaign cost against category norms before calling it good or bad. Hyper casual games get a CPI of roughly $3 to $6, midcore games $20 to $25, and a 4X builder game almost $50 per install.
Product, category and audience all shift these numbers.
2Girotra's method for feedback signals
He grades every feedback signal on recency, frequency and quality. Recency is how soon after install the event fires. Frequency is volume: Facebook needs about 50 events in a rolling seven days, and Google about 10 per campaign per day.
Google's 10 a day is why he budgets about 10 times the target CPA. Quality means the event must correlate with revenue.
He has engineering build synthetic events for users judged high quality. Examples are an "HQ sign up" and an "HQ bank link", sent to Facebook and Google as the optimization event so the algorithm finds more users like them.
He used this for a cash back and rewards app at Times Internet. The event has to track the real goal, usually revenue.
At Elevate Labs he delayed the trial start event by 12 hours. Charting cancellations showed 30-odd percent came in the first six hours of a 7-day trial on brain training games. He fired the adjusted event 12 hours in and skipped anyone who had already cancelled.
User quality improved by almost a third. In his words: "I introduced a mild DeLay So my recency is still taken care of".
3Girotra on targeting and creative
Girotra goes broad on targeting when the goal is scale. Interest targeting can work but never gave him scale. Even without an audience list, he keeps targeting broad because the machine learning finds the best audience from a small sample.
For a venture-backed company at scale he runs experiments to find the audience: "the only way out is Broad targeting".
He uses creative as a screen so the right people respond. Abound launched as a cash back and rewards app, and shopping is so big a category that it drew the wrong people. Adding Indian faces and a send-money-to-India feature made viewers self-select.
The result was an 80% match rate with the target audience, where only 1% of the demographic is Indian. B-level but recognizable celebrity cameos have also had some success.
He seeds lookalikes from sister apps and tests 1%, 5% and 10%. Times Internet used audience lists from other apps in its portfolio as seeds for each locale.
He keeps running experiments across the three lookalike sizes to see which finds the audience best.
He caps audience expansion at 50% above his estimated audience. If there are 4 million Indians in the US and the audience grows to 15 million, it has expanded too far.
When Advantage Plus stretches a 10% lookalike too much, he dials back to a 5% seed list and runs Advantage Plus on that.
He segments first-party data with RFM to build lookalikes from power users. Recency, frequency and monetization analysis splits users into power, casual and churned. Messaging those segments through custom audiences is retargeting.
A custom audience of power users becomes the seed for a lookalike used in prospecting.
For a B2B app aimed at salespeople, he would start on LinkedIn. On a limited budget, LinkedIn targets by company, designation and years of experience, so context matters as well as audience.
If using interests, he says the more granular the better. The host cut CPI for a sports betting client by moving from the NFL interest to people who liked all 32 teams.
4Girotra's advice for people starting out
He tells beginners to copy what works and not be a hero. Bring in someone experienced with Facebook, budget for learning costs, and copy winning ads from the Facebook or TikTok ad libraries. Skip expensive famous influencers early.
Track retention and where the curve flattens to see how sticky the product is. His line: "don't be a hero".
He warns that early small-budget wins do not prove product market fit. At a few hundred dollars a day the algorithm easily finds the low-hanging fruit, so an MVP can look like it has found fit. Companies then raise money and scale.
At $10,000 a day costs often go out of control. He treats early cheap results as a weak signal.
He does not count on iOS retargeting being measurable. With no device ID there is no suppression list, so ads reach people who already installed the app. Android still has more avenues, and SKAN can now tell redownloads apart.
He uses lookalikes for prospecting and custom audiences for retargeting.
He now leaves placements to the algorithm with dynamic placements. The algorithm chooses between Instagram, Messenger and Facebook. He once manually excluded Audience Network but no longer does.
He finds Instagram ads effective for app downloads.
The Simplest Facebook Ads Strategy for App Installs
Teaches how to run Facebook app install campaigns by walking down the funnel, going broad on targeting, and steering the algorithm with creative and feedback signals. The result worth hearing is an 80% audience match rate on a niche app and a trial-event trick that lifted user quality by almost a third.
Girotra nudges the algorithm from the side and never controls it directly. He compares it to sweeping a curling stone: you guide it but never touch it. The three levers are targeting, creative and feedback signals.
Each lever corrects the algorithm's course after the campaign has started.
He starts every campaign on the install, then walks down the funnel. He does not optimize for something like level 10 of a game on day one, because that starves the campaign of data. The path runs from install to a level 2 or trial start, then to subscription.
Along the way he gathers data and fixes the product or onboarding before asking for rarer events.
He judges campaign cost against category norms before calling it good or bad. Hyper casual games get a CPI of roughly $3 to $6, midcore games $20 to $25, and a 4X builder game almost $50 per install.
Product, category and audience all shift these numbers.
He grades every feedback signal on recency, frequency and quality. Recency is how soon after install the event fires. Frequency is volume: Facebook needs about 50 events in a rolling seven days, and Google about 10 per campaign per day.
Google's 10 a day is why he budgets about 10 times the target CPA. Quality means the event must correlate with revenue.
He has engineering build synthetic events for users judged high quality. Examples are an "HQ sign up" and an "HQ bank link", sent to Facebook and Google as the optimization event so the algorithm finds more users like them.
He used this for a cash back and rewards app at Times Internet. The event has to track the real goal, usually revenue.
At Elevate Labs he delayed the trial start event by 12 hours. Charting cancellations showed 30-odd percent came in the first six hours of a 7-day trial on brain training games. He fired the adjusted event 12 hours in and skipped anyone who had already cancelled.
User quality improved by almost a third. In his words: "I introduced a mild DeLay So my recency is still taken care of".
Girotra goes broad on targeting when the goal is scale. Interest targeting can work but never gave him scale. Even without an audience list, he keeps targeting broad because the machine learning finds the best audience from a small sample.
For a venture-backed company at scale he runs experiments to find the audience: "the only way out is Broad targeting".
He uses creative as a screen so the right people respond. Abound launched as a cash back and rewards app, and shopping is so big a category that it drew the wrong people. Adding Indian faces and a send-money-to-India feature made viewers self-select.
The result was an 80% match rate with the target audience, where only 1% of the demographic is Indian. B-level but recognizable celebrity cameos have also had some success.
He seeds lookalikes from sister apps and tests 1%, 5% and 10%. Times Internet used audience lists from other apps in its portfolio as seeds for each locale.
He keeps running experiments across the three lookalike sizes to see which finds the audience best.
He caps audience expansion at 50% above his estimated audience. If there are 4 million Indians in the US and the audience grows to 15 million, it has expanded too far.
When Advantage Plus stretches a 10% lookalike too much, he dials back to a 5% seed list and runs Advantage Plus on that.
He segments first-party data with RFM to build lookalikes from power users. Recency, frequency and monetization analysis splits users into power, casual and churned. Messaging those segments through custom audiences is retargeting.
A custom audience of power users becomes the seed for a lookalike used in prospecting.
For a B2B app aimed at salespeople, he would start on LinkedIn. On a limited budget, LinkedIn targets by company, designation and years of experience, so context matters as well as audience.
If using interests, he says the more granular the better. The host cut CPI for a sports betting client by moving from the NFL interest to people who liked all 32 teams.
He tells beginners to copy what works and not be a hero. Bring in someone experienced with Facebook, budget for learning costs, and copy winning ads from the Facebook or TikTok ad libraries. Skip expensive famous influencers early.
Track retention and where the curve flattens to see how sticky the product is. His line: "don't be a hero".
He warns that early small-budget wins do not prove product market fit. At a few hundred dollars a day the algorithm easily finds the low-hanging fruit, so an MVP can look like it has found fit. Companies then raise money and scale.
At $10,000 a day costs often go out of control. He treats early cheap results as a weak signal.
He does not count on iOS retargeting being measurable. With no device ID there is no suppression list, so ads reach people who already installed the app. Android still has more avenues, and SKAN can now tell redownloads apart.
He uses lookalikes for prospecting and custom audiences for retargeting.
He now leaves placements to the algorithm with dynamic placements. The algorithm chooses between Instagram, Messenger and Facebook. He once manually excluded Audience Network but no longer does.
He finds Instagram ads effective for app downloads.