Colette Nataf (CEO & Co-Founder at Lightning AI - helping advertisers to find their target audiences) talks with Shamanth Rao about how to think about automating some tasks of user acquisition, potential quick wins and finding new audiences.
For automation, start with mapping your own process: 1. Which steps are you taking each day as a marketer? 2. Which ones are repetitive? 3. Do they need actual human intelligence?
Automating bid changes requires low effort. The hard part is getting the access required and setting up the connection between your engineer's workflow and Facebook. heads up: Facebook documentation is hard to figure out and sometimes outdated.
Really explain what the problem is, not just how you think it should be solved. Engineers might prefer having guidelines and adapt things in the middle.
One of the things that can make the biggest difference is LTV predictions and sending back predictions of lifetime values to Facebook. It makes a huge different in optimization and allows you to work with the Facebook algorithm.
Count the first purchase as just a purchase. As soon as you can, send back to Facebook whatever event allows you to predict LTV as purchase and with the value. Facebook will figure it.
The way LightningAI creates audiences: 1. Look for something the audience likes 2. Search different social media sites to find people talking about that. 3. Make correlations to find new audiences, put them in a dataset 4. Rapidly test these new audiences
The impact really shows when people apply these learnings on correlations to their creatives or even their product.
Something that can make an impact for people is to have a way to scale up ads and upload them to Facebook automatically.
For automation, start with mapping your own process: 1. Which steps are you taking each day as a marketer? 2. Which ones are repetitive? 3. Do they need actual human intelligence?
Automating bid changes requires low effort. The hard part is getting the access required and setting up the connection between your engineer's workflow and Facebook. heads up: Facebook documentation is hard to figure out and sometimes outdated.
Really explain what the problem is, not just how you think it should be solved. Engineers might prefer having guidelines and adapt things in the middle.
One of the things that can make the biggest difference is LTV predictions and sending back predictions of lifetime values to Facebook. It makes a huge different in optimization and allows you to work with the Facebook algorithm.
Count the first purchase as just a purchase. As soon as you can, send back to Facebook whatever event allows you to predict LTV as purchase and with the value. Facebook will figure it.
The way LightningAI creates audiences: 1. Look for something the audience likes 2. Search different social media sites to find people talking about that. 3. Make correlations to find new audiences, put them in a dataset 4. Rapidly test these new audiences
The impact really shows when people apply these learnings on correlations to their creatives or even their product.
Something that can make an impact for people is to have a way to scale up ads and upload them to Facebook automatically.
For automation, start with mapping your own process: 1. Which steps are you taking each day as a marketer? 2. Which ones are repetitive? 3. Do they need actual human intelligence?
Automating bid changes requires low effort. The hard part is getting the access required and setting up the connection between your engineer's workflow and Facebook. heads up: Facebook documentation is hard to figure out and sometimes outdated.
Really explain what the problem is, not just how you think it should be solved. Engineers might prefer having guidelines and adapt things in the middle.
One of the things that can make the biggest difference is LTV predictions and sending back predictions of lifetime values to Facebook. It makes a huge different in optimization and allows you to work with the Facebook algorithm.
Count the first purchase as just a purchase. As soon as you can, send back to Facebook whatever event allows you to predict LTV as purchase and with the value. Facebook will figure it.
The way LightningAI creates audiences: 1. Look for something the audience likes 2. Search different social media sites to find people talking about that. 3. Make correlations to find new audiences, put them in a dataset 4. Rapidly test these new audiences
The impact really shows when people apply these learnings on correlations to their creatives or even their product.
Something that can make an impact for people is to have a way to scale up ads and upload them to Facebook automatically.
Notes for this resource are currently being transferred and will be available soon.
When they started thinking about automation it's because they were doing the same things again and again. Example: bid adjustments based on conversion rate.
[💎@04:57]
Map the process of what you're doing multiple times and explore what
To automate you typically need a backend engineer (for about 90% of your tasks).
Finding such an engineer can be hard (if not on staff you might need to get buy in from a project manager) but you can hire consultants (from Upwork or word of mouth).
[💎@10:03] Automating bid changes requires low effort. The hard part is getting the access required and setting up the connection between your engineer's workflow and Facebook. heads up: Facebook documentation is hard to figure out and sometimes outdated.
[💎@10:58] Really explain what the problem is, not just how you think it should be solved. Engineers might prefer having guidelines and adapt things in the middle.
A lot of people are optimizing towards the wrong thing on Facebook: add to cart instead of purchase, etc.
A lot of gaming companies are not optimizing towards LTV which can lead to churn or lower conversion rate than users coming from organic.
It's not enough to just send back through the mobile SDK the events that are happening, you also need to send back predictions of future value.
[💎@13:10] One of the things that can make the biggest difference is LTV predictions and sending back predictions of lifetime values to Facebook. It makes a huge different in optimization and allows you to work with the Facebook algorithm.
Most complex part: LTV prediction and leading indicators of LTV.
In mobile gaming you typically can't predict LTV based on the first purchase.
[💎@15:20] Count the first purchase as just a purchase. As soon as you can, send back to Facebook whatever event allows you to predict LTV as purchase and with the value. Facebook will figure it.
If CPAs are high you might not have enough conversions for the algorithm. If you can predict potential conversion rates based on some early signals and then send that value back to Facebook you'll be able to "fake it" with more events.
Example for SaaS business: predict lead score based on the email address and the info they can get from the company name. It allows them to not have to wait for a purchase.
Take a data analytics class: data science, stat 101, etc. so you can understand how you and engineers think about data/numbers and these processes.
Thinking about finding rules to make decisions without using your own judgment is really important when you start thinking about how to automate. It's about how you frame what you're doing.
As a marketer it's best to come up with how something should be done and the result and hand it off to an engineer.
Facebook constructs audiences under the hood.
[💎@24:13] The way LightningAI creates audiences:
They now have a big data set, and they keep testing new audiences within that system.
[💎@27:09] The impact really shows when people apply these learnings on correlations to their creatives or even their product.
It's really exciting when people and computers can work together
A lot of people don't think they have a set template when producing creatives, but they probably do.
Working off a template with different texts for copy where you can use different images to create a lot of ads can be done by a computer (even if finding the images is a human's job), as well as uploading and testing them.
[💎@30:01] Something that can make an impact for people is to have a way to scale up ads and upload them to Facebook automatically.
When they started thinking about automation it's because they were doing the same things again and again. Example: bid adjustments based on conversion rate.
[💎@04:57]
Map the process of what you're doing multiple times and explore what
To automate you typically need a backend engineer (for about 90% of your tasks).
Finding such an engineer can be hard (if not on staff you might need to get buy in from a project manager) but you can hire consultants (from Upwork or word of mouth).
[💎@10:03] Automating bid changes requires low effort. The hard part is getting the access required and setting up the connection between your engineer's workflow and Facebook. heads up: Facebook documentation is hard to figure out and sometimes outdated.
[💎@10:58] Really explain what the problem is, not just how you think it should be solved. Engineers might prefer having guidelines and adapt things in the middle.
A lot of people are optimizing towards the wrong thing on Facebook: add to cart instead of purchase, etc.
A lot of gaming companies are not optimizing towards LTV which can lead to churn or lower conversion rate than users coming from organic.
It's not enough to just send back through the mobile SDK the events that are happening, you also need to send back predictions of future value.
[💎@13:10] One of the things that can make the biggest difference is LTV predictions and sending back predictions of lifetime values to Facebook. It makes a huge different in optimization and allows you to work with the Facebook algorithm.
Most complex part: LTV prediction and leading indicators of LTV.
In mobile gaming you typically can't predict LTV based on the first purchase.
[💎@15:20] Count the first purchase as just a purchase. As soon as you can, send back to Facebook whatever event allows you to predict LTV as purchase and with the value. Facebook will figure it.
If CPAs are high you might not have enough conversions for the algorithm. If you can predict potential conversion rates based on some early signals and then send that value back to Facebook you'll be able to "fake it" with more events.
Example for SaaS business: predict lead score based on the email address and the info they can get from the company name. It allows them to not have to wait for a purchase.
Take a data analytics class: data science, stat 101, etc. so you can understand how you and engineers think about data/numbers and these processes.
Thinking about finding rules to make decisions without using your own judgment is really important when you start thinking about how to automate. It's about how you frame what you're doing.
As a marketer it's best to come up with how something should be done and the result and hand it off to an engineer.
Facebook constructs audiences under the hood.
[💎@24:13] The way LightningAI creates audiences:
They now have a big data set, and they keep testing new audiences within that system.
[💎@27:09] The impact really shows when people apply these learnings on correlations to their creatives or even their product.
It's really exciting when people and computers can work together
A lot of people don't think they have a set template when producing creatives, but they probably do.
Working off a template with different texts for copy where you can use different images to create a lot of ads can be done by a computer (even if finding the images is a human's job), as well as uploading and testing them.
[💎@30:01] Something that can make an impact for people is to have a way to scale up ads and upload them to Facebook automatically.
When they started thinking about automation it's because they were doing the same things again and again. Example: bid adjustments based on conversion rate.
[💎@04:57]
Map the process of what you're doing multiple times and explore what
To automate you typically need a backend engineer (for about 90% of your tasks).
Finding such an engineer can be hard (if not on staff you might need to get buy in from a project manager) but you can hire consultants (from Upwork or word of mouth).
[💎@10:03] Automating bid changes requires low effort. The hard part is getting the access required and setting up the connection between your engineer's workflow and Facebook. heads up: Facebook documentation is hard to figure out and sometimes outdated.
[💎@10:58] Really explain what the problem is, not just how you think it should be solved. Engineers might prefer having guidelines and adapt things in the middle.
A lot of people are optimizing towards the wrong thing on Facebook: add to cart instead of purchase, etc.
A lot of gaming companies are not optimizing towards LTV which can lead to churn or lower conversion rate than users coming from organic.
It's not enough to just send back through the mobile SDK the events that are happening, you also need to send back predictions of future value.
[💎@13:10] One of the things that can make the biggest difference is LTV predictions and sending back predictions of lifetime values to Facebook. It makes a huge different in optimization and allows you to work with the Facebook algorithm.
Most complex part: LTV prediction and leading indicators of LTV.
In mobile gaming you typically can't predict LTV based on the first purchase.
[💎@15:20] Count the first purchase as just a purchase. As soon as you can, send back to Facebook whatever event allows you to predict LTV as purchase and with the value. Facebook will figure it.
If CPAs are high you might not have enough conversions for the algorithm. If you can predict potential conversion rates based on some early signals and then send that value back to Facebook you'll be able to "fake it" with more events.
Example for SaaS business: predict lead score based on the email address and the info they can get from the company name. It allows them to not have to wait for a purchase.
Take a data analytics class: data science, stat 101, etc. so you can understand how you and engineers think about data/numbers and these processes.
Thinking about finding rules to make decisions without using your own judgment is really important when you start thinking about how to automate. It's about how you frame what you're doing.
As a marketer it's best to come up with how something should be done and the result and hand it off to an engineer.
Facebook constructs audiences under the hood.
[💎@24:13] The way LightningAI creates audiences:
They now have a big data set, and they keep testing new audiences within that system.
[💎@27:09] The impact really shows when people apply these learnings on correlations to their creatives or even their product.
It's really exciting when people and computers can work together
A lot of people don't think they have a set template when producing creatives, but they probably do.
Working off a template with different texts for copy where you can use different images to create a lot of ads can be done by a computer (even if finding the images is a human's job), as well as uploading and testing them.
[💎@30:01] Something that can make an impact for people is to have a way to scale up ads and upload them to Facebook automatically.