CaddieOpen in CaddieDownload for Mac

What is Marketing Mix Modeling (MMM)

Shared by Caddie Library · Updated
App Masters14:51Open on YouTube
Chapters show their main points

The bottom line

Marketing mix modeling estimates each marketing channel's impact on results from aggregated spend and outcome data, so it needs no user-level tracking. That makes it a practical way for app marketers to regain the visibility lost to iOS 14.5 and ATT.

Key takeaways

  • Last touch attribution and privacy changes have left app marketers unable to see what really works. 1:38
    • Last touch gives 100% of the credit to the final touch point and ignores every other step in the user journey.
    • Opt-in rates for the advertising ID have dropped to around 20%, so conversions can no longer be matched to touch points.
  • Marketing mix modeling links daily spend and outside factors to results with statistics. 3:38
    • Inputs are daily spend per channel plus external variables such as holidays and market conditions. Outcomes are installs or revenue per day.
    • The models use regression or Bayesian methods to estimate the effect of each variable on performance.
  • Because it uses aggregated data, it works without advertising IDs and covers offline media. 6:01
    • Podcasts and linear TV are hard to track at user level, but their daily spend can be fed into the model.
    • Digital ads that look expensive per install may still lift other channels, such as Apple Search Ads or organic installs, and the model can show it.
  • The model accounts for ad carryover and saturation, so it shows diminishing returns. 10:20
    • Carryover means an ad has its strongest effect on day one and then fades over time.
    • Saturation means each extra ad adds less value than the one before.
  • It can simulate next steps, and it gives small apps the confidence to scale spend. 11:04
    • The predictions are a good advisor to double check plans against, but they cannot be relied on 100%.
    • Without clear unit economics, such as a CPI under $1 to get 5,000 users for $5,000, it is hard to spend more on a campaign with confidence.

If you eat one ice cream you're happy, if you had 100 ice creams you're unhappy.

Roy Nam, CEO of Airbridge

Chapter summaries

0:00 Why last touch and privacy rules call for a new method

Last touch credit and privacy rules both hide what works, so measurement has to work without user-level data.

  • Last touch misjudges what advertising is worth. 1:40
    • It cannot measure incrementality.
    • Credit should reach the touch points that influenced a conversion.
    • Roy Nam calls last touch "a pretty simplistic methodology" because it gives one touch point all the credit.
    • The fix is to look across the whole user journey, not at one conversion.
  • Privacy rules distort the numbers and will only tighten. 2:30
    • iOS 14.5 and ATT skewed the figures on MMP and ad platform dashboards.
    • Browsers are phasing out third-party cookies.
    • Safari uses Intelligent Tracking Prevention, and Chrome and Firefox have similar policies.
    • GDPR and CCPA add pressure, with more regulation expected.

3:34 How the model is fed

The model sets each day's spend and outside conditions beside that day's results and learns which variables move them.

  • Spend and outside factors are listed day by day, then matched to results. 4:00
    • Example: on day one, $1,000 goes to channel A, $2,000 to B and $3,000 to C.
    • Outside variables include holidays, FX moves and a stock price index.
    • A holiday is a simple yes or no, such as Christmas yes, any other day no.
    • The model learns how performance rises and falls as each variable changes, then estimates each one's impact.
    • Nam says you need not understand the math, only that it is machine learning or statistics.

5:48 Why app marketers need it

Without the advertising ID, conversions and touch points can no longer be matched, and MMM does not need that match.

  • Losing the advertising ID breaks the link between touch points and conversions. 6:14
    • MMPs used the ID as the matching key.
    • With it, you knew who saw an ad and who subscribed.
    • Without it the user journey is incomplete, so attribution cannot run.
    • Organic install numbers can look skewed as a result.
  • MMM offers four benefits. 7:03
    • It uses aggregated data and covers digital and non-digital media.
    • It accounts for carryover and saturation.
    • Its fourth benefit is predictive and prescriptive insight, not only a look back.
    • Nam presents aggregated data as the first and most important benefit.

7:26 Why the whole mix matters

A channel that looks unprofitable on its own can still help the rest of the mix.

  • Judging each channel alone hides its value to the rest of the mix. 7:31
    • One developer knew a channel looked unprofitable, yet believed it helped the overall mix.
    • At Walmart.com, TV and podcast ads were hard to track.
    • Listeners often just search for the app afterwards, so the ad never gets credit.
    • Roy calls this one of the most classic marketer questions. CTV, linear TV and video campaigns all feed the final conversion.

9:00 Built for the privacy era

Using only aggregated data makes MMM friendly to ATT.

  • The biggest benefit is running the analysis without advertising IDs or user-level data. 9:07
    • Opt-in rates have fallen, so conversions rarely match to touch points.
    • Roy calls MMM an "ATT friendly" solution.
    • Skipping individual touch points is the second benefit. It covers podcasts and linear TV.
    • Most measurement tools look backward. MMM can also simulate what to do next.

12:46 Feeding the model your channels

The model only knows about channels whose data you share.

  • Marketers must supply daily spend for each channel. 13:06
    • Airbridge cannot know about a podcast campaign until told.
    • Sharing a spend that moves, say $200 a day rising to $300 or $400, lets the model run.
    • Machine learning compares podcasts and TV with Facebook impressions. It looks for effects on organic or Apple Search Ads downloads.
    • Roy's advice is to view all spend holistically, using MMM or a tool that does the calculation.

In the growth library

Roy Nam, CEO of Airbridge, a mobile measurement partner (MMP). Marketing mix modeling (MMM) measures how each channel drives installs and revenue using only aggregated data, which makes it a workable replacement for user-level attribution lost to iOS 14.5 and ATT. It explains what inputs the model needs and what it adds over last touch attribution, including offline channels, carryover, saturation and forecasting.

Lessons that draw on this:

More from App Masters · The whole library, and ask it anything

Transcript391 lines
0:00

what is up app Nation it is Steve p young founder of appmasters.com and in this video we're going to explain mmm marketing mix modeling and why you really need it given all the privacy issues that we have especially when it comes to app marketing and to explain it all he prepared a fancy little deck for us is Roy Nam he is the CEO of airbridge go check them out it is airbridge dot Io if you need an MMP that's mobile measurement partner Roy welcome to the stream hi Steve thanks for inviting me today all right I'll follow your lead so you prepared a document for us let's go through this presentation explain to us mmm yeah yeah for sure so marketing mixed modeling as you said it's definitely one of the rising technologies that you should keep in mind and especially this technology is needed in the age of privacy so we all know that iOS 14.5 and ATT is causing a lot of visibility issues into terms of

1:00

digital marketing and in order for you to actually recover this visibility to better understand what's happening to your M marketing you would definitely need to consider mmm as one of your useful tools in the future so I want to actually just briefly talk about what mmm is and why is it important and how as a smaller applications you can utilize this technology to have a better visibility into your marketing performance so can we go to the next slide so just a little bit of background why Triple M mmm came into the marketing picture so definitely because of two reasons so first of all limitations of the last touch Point model so we all know that many ad marketer has been relying on this attribution model called last touch Point model which gives 100 credit to the last touch point on the user Journey however there's some limitations to this last touch Point model like for example it's enabled to

1:54

accurately measure the true effectiveness of advertising this means uh you cannot measure decrentality and also it does not take multiple touch points in a user journey into account and lastly uh it's a pretty simplistic methodology which gives 100 attribution credit to the single last touch Point rather than Distributing to the multiple touch point on the user Journey so what we should do about it is we need to recognize the touch points that have actually impacted the conversion not just a single conversion but across multiple touch points on the user journey and second uh reason why we would need mmm is because of the Privacy change so we all know that is 14.5 and ATT had a huge impact across the industry and many of us have felt this strongly because it actually kind of distorted the numbers that we are seeing on the on the dashboard of the MMP or on the ad platforms many of these numbers

2:52

has been impacted and knowledge is that we also have a lot of these browsers implementing new policies to Decay third-party cookies for example we do you have uh intelligence Tracking prevention and Safari we do have a similar one in Chrome or in Firefox and there are also a lot of these regulations coming out to the market like gdpr CCPA and even more so in the future so we need to measure the contribution of advertisements without using individual level without using user level granular data set so then it comes mmm as a as a very reliable alternative to overcome some of these limitations that I have just shared with you there are actually four characteristics of mmm and before that I want to just briefly explain to you what mmm is marketing mix modeling is a statistical analysis method that evaluates the impact of individual marketing variables and performance by analyzing the correlation between these

3:49

variables and uh performances so it's pretty complicated but if you see the picture uh it clearly explains what it is so basically you you have all kinds of Mark marketing variables for example you're spending your budget on channel a channel B and channel C so each day each one of these channels will have the ad spend for example like on day one you're spending 1 000 USD to channel a 2000 USD to channel B and 3000 USD to channel C and also we will have some external variables like seasonality or stock price index some of these external Market variables that you want to take into account so you actually list up all of these variables like it's today holiday yes no if it's Christmas yes if it's not Christmas no simple and maybe um up and downs of these you know FX up and downs of the Marquette conditions and also you would have to put in some of the marketing related inputs like S Pen for each channel on a daily basis

4:51

and if you have all these marketing variables on one hand if you do have some of the outcome for example the number of installs per day the total revenue per day if you do have this two inputs two types of inputs marketing variables and performances actually mmm models are usually based on machine learning such as regression or some kind of statistical methodology like Bayesian models so you don't have to understand all about this but just make sure that making sure that this is based on machine learning or statistical methodology so by leveraging whichever ml or statistical methodology they will actually find out the correlation between these variables and the performances and by doing that eventually they learn the increase and decrease in performance caused by changes in the marketing variables and estimate and predict the actual impact of each individual variable on performance so this is a very rough

5:45

picture of how it actually works why does it matter uh why it doesn't matter for applications and the end marketers because of four reasons so I won't actually dive in to one of these reasons um each one of them so if the first reason why you have to take this into account is because it utilizes aggregated data not user level data so this is really important because when we think of the current status quo many of these mmps or many of these measurement Solutions actually leverage advertising ID as a major matching key in terms of connecting the conversions and the touch points so basically if you if you do have Ed ID you know who watched the ad you know who bought the product at the end of the day inside your application who actually subscribed to your app but without this ad ID you kind of lost the matching key between the conversions and the touch points which kind of misleads um you into a very maybe a little bit

6:44

inaccurate um user Journey what you might have observed is as a clear decrease of organic and still the clear increase of organic insults because if you cannot actually build the complete user Journey you would not be able to run the attribution and that is causing a lot of issues utilizing aggregated data not user level data that way the first benefit second evaluates digital and non-digital media holistically third accounts for ad carryover and saturation effects and the fourth one would be it provides predictive and prescriptive insights not just the retrospective uh insight into the marketing performance Roy if I can take a break real quick it all it's almost like like hey we we're taking a holistic view on the entire marketing channel and the marketing mix because you know I had one developer stayed me look I know I'm spending money unprofitably or this channel is super high in terms of conversion but I know

7:41

that it also helps out the overall marketing mix right and so it's not just being like and for the longest time I used to work for a bigger company walmart.com we used to run these tv ads we used to run podcast ads and it's hard to track these these ads because people are probably just searching for better help or any other app you name it but mmm is starting to take that into consideration and be like yes you know if we're just doing these digital ad media placements this may be a super high cost per install but it actually does impact our ASA campaigns our Facebook ad campaigns and all that jazz it's a nutshell feels like that's what mmm's trying to do yes you're right Steve it's actually one of the most classic questions and things on any marketer would ask it's actually taking a multi-touch uh perspective basically you mentioned as you mentioned there's some a lot of these brand advertisements

8:32

like CT be linear TVs or maybe you would run a lot of these video campaigns eventually all of these middle touch points contribute to the end conversion however under last touch Point model they will give the credit to the you know the single last touch point which doesn't make sense for all of these metal touch points to also play their part how to really accurately measure all these middle touch points is one of the major missions that we always want to overcome and I can tell you that mmm can certainly do that as well yeah just uh each one of these slides kind of highlights what I just what I just summarized it's a well suited for post privacy era because it only uses aggregated data not individual user level info especially this is a good news for iOS and marketers we all know that opt-in rate for advertising ID is actually dropped to around 20 that range and because of that the matching rate

9:25

between the conversion and the touch Point has dropped dramatically as well which is causing the issues in terms of visibility into the marketing performance so basically because it only uses aggregated data I would say this very ATT friendly solution so even without having to collect all of these advertising IDs or user level raw data you would be able to run the analysis and that would be the biggest benefit that you will get from using mmm and the second one because it uses aggregated data you don't have to worry about collecting all of these individual touch points from non-digital media like podcast or lunar TVC TVs that would be I would say the second benefit of using mmm the third one is actually um can we actually move on to the second the next slide piece yeah so that was just one that I just mentioned about digital and non-digital media the coverage the next slide please and also

10:20

there's two kinds of effects that you might want to consider one is carryover effect and saturation effect the other one so carry over effect in other words is a long-term influence of the advertisements over the time being it's basically day one it would have the strongest effect and they it would have to decrease effect and we want to actually take this into account and saturation effect is if you eat one ice cream you're happy if you had 100 ice creams you're unhappy so basically the marginal utility you would get uh from each ad uh would decrease over time so we won't actually also take this effect into account mmm actually take these two effects into account so much naturally because it's based on statistical methodology and that's another benefit you will get from Mmm so maybe the last uh site about mmm is it's about prediction uh so many of these measurement Solutions have been focused

11:11

on bringing you to the retrospective uh insights rather than something that would happen in the future because it's based on machine learning and system Methods at some part it will it can actually run simulations and tell you what to do next as is next step of course you cannot really rely upon this Insight 100 but certainly actually act as a good advisor and you double check checking what you will do next that you have in mind these are the four benefits of mmm and maybe the last slide so why do you have to care again as a small application or as an app builder the biggest challenges you may face under iOS 14.5 and ATT is building a unit economics that allows you to scale your marketing operations effectively so you've been measuring CPI you've been measuring CPA if you want to acquire 5000 users uh with um with 5000 USD you have to make sure that CPI is always below one USD but if you don't know this

12:08

you can't just be so much confidence in spending more money on each campaign or creative and that is causing a lot of issues so why marketing mixed modeling at the end of the day would be useful for you is because it can be available tool for small apps or Publishers looking to overcome this visibility loss under is 14.5 and ATT so arabic's a little bit about us we are a unified nursing stack a mobile Mission partner uh helping Brands and marketers optimize their media mix and media spend if you want to reach out to Nam you can do that his email is pretty easy Nam at airbridge dot IO hey now one of the things I wanted to also talk to you about is it looks like mmn like with open Ai and chat gbt come into play it sort of like takes into consideration all the different marketing channels like hey are you running these channels like for example Facebook or podcast advertising do we have to tell airbridge

13:04

that this is like how do you know that I'm running podcast advertising for example we don't know until you tell us the easiest way you can tell is it's actually giving us the data so for example to run the mmm model you can actually let us know daily spend for that podcast so maybe you're spending like 200 each day or maybe you increase it to 300 or 400 yesterday to actually share with us this data set and we will know and we can run the model based on that and it's just using some sophisticated machine learning type of thing that says hey we know podcasts and the effectiveness of podcasts and TV versus like maybe a Facebook impression you guys have the modeling and the predictions underneath that like hey maybe because of all the data that you're running within airbridge you know that maybe podcasts or Facebook ad impressions are impacting and let's just for example organic download or an apple

13:54

search ads download hey look if you want to look cool when you're in an app event just throw out mmm it's going to be the trendy term in 2023 I believe and that's why I wanted to talk to Roy and bring him on to really explain it for us because I've been hearing a lot about it and for me this is a learning platform as well Roy anything I miss that you want to make sure we cover about mmm no I think we have covered most of the topics Steve it seems like the trend and it seems the way to go and again look at your ad spend all your marketing channels at a holistic level and mmm or use a tool like air bridge where they're going to do all that fancy calculation and machine learning so once again it is airbridge dot IO and I'm all up their website one more time so every Bridge dot IO go check them out and it is linked up on your favorite podcast app or in the YouTube description yeah I mean right thank you so much what do you

14:42

prefer Roy or now Roy you could call me numb but yeah Roy sweater for me thank you guys for watching I'll see you on the next video