For a while now, I’ve been coming across discussions from notorious VCs about CAC, and how they distrust companies that promise to optimize CAC as they scale. The main reason for this is that they say audiences shrink as a company grows, and for that reason CAC increases. 

‍

I disagree.

‍

After helping more than 3,000 companies with data analytics  projects, I’ve seen that  it is possible to reduce CAC as you scale.  The main driver for its optimization is the rise of technologies like ML that are still not very widely used across organizations. A recent Gartner research found that 85% of big data projects fail and part of that is because companies are failing to put ML into production. Saying that reducing CAC is not possible is like saying 100 years ago that going to the moon would not have been possible because there were no means to get there. Machine Learning is starting to help companies at later stages and at hyper growth reduce CAC significantly as they scale because now we are starting to understand how to use data and technology to get better.

‍

A couple of years ago the only way to optimize ads was to do AB testing. Then, Facebook and Google and other vendors launched ML capabilities. At some point that just stopped working, because certainly, audiences don’t grow exponentially, and although ML helped, it was just exclusively tied to the ad network audiences.. The shift happened in how companies are storing data, processing it and analyzing it nowadays. Here is an example of a billion dollar startup that we work with. The traditional process was to just use an optimization software that would just generate AB tests to try to understand what audiences and creatives were more effective. Of course CAC at the beginning lowered a bit but then, as they scaled their ad budget by 50% MoM, CAC started to get uncontrollable. 

‍

Then, they started to step up their game by working with all of their customer and behavioral data, and started analyzing things like: what are the first products their clients were buying, what products they should recommend for first time or second time buyers. Also, variables like at what time of day are they selling more, what cities , among many other things available in their different data sets. Finally, they runned algorithms to try to understand what was causing cart abandonment, which consequently was causing CAC to be high. 

‍

But not only that, experiments went from being AB experiments, to advanced ML experiments with ML systems that would allocate budget based on cost and objectives.


Thanks to these initiatives this company was able to reduce CAC by 52%. This did not happen overnight of course. It was an ongoing effort that lasted about 4 months. Certainly, when CAC reduction is approached it can be seen as a problem that is exclusive to ad implementation. CAC reduction is a business problem where different areas need to work in collaboration and where deep technology needs to come in place to analyze data patterns and find new ways of doing things.


To sum it all up, companies can reduce CAC as they scale if they follow these recommendations:


1. Don’t just focus on optimizing ads. Facebook and Google are getting pretty good at that. Stop thinking that you can manually outwin their ML.


2. Analyze the data of your product with advanced ML. There’s always things you can optimize that will reduce your CAC. 


3. Leverage your customer data to understand patterns. Run ML algorithms to try to predict segments. Use those predictions to run different ads.


4. Use ML technology to allocate budget between platforms and bidding mechanisms that can allocate your budget wherever is best, taking into consideration real time costs, budget and goals.


5. Build creatives based on data, not on gut filling.


6. Collaborate based on data with different areas of the company and vendors. For example, share data points with growth, BI, Agencies, etc.


7. Be patient. I’ve seen how growth hackers think that by making manual changes every day everything will just work better. ML takes time to learn. Give it some time, at least 10 days to start to get conclusions.


8. THe role of the Growth Hacker is not just to run multiple AB testing. The role of the growth hacker is to also be able to dig deep into the data. And for that they need to have the tools and the team members that can help them accomplish their goal. 


Putting Machine Learning pipelines into production is quickly becoming one of the top priorities at companies, and it is the job of data and growth team to work in collaboration to be able to make data actionable.