Flexible Mapping: Smadex’s new solution for SKAd campaign optimization

Flexible Mapping: Smadex’s new solution for SKAd campaign optimization

We’ve mentioned why utilizing SKAdNetwork knowledge has such restricted utility, however we do have a number of different vital knowledge units accessible to us to assist mannequin predicted marketing campaign income. SKAdNetwork has put in movement a brand new cell programmatic framework marked by knowledge receipt delayslack of user-level knowledge, and lowered visibility of post-install occasions.

Table of Content


With the most recent adjustments to its privateness coverage, Apple has modified the way in which measurement and attribution work on iOS. Advertisers face additional limitations on this new actuality that SKAd introduces. With much less knowledge to work with, there’s an rising want for transparencyAt Smadex, we constructed the answer to excel on this context: our new Flexible Mapping software.

Knowledge accessible pre and put up full implementation of SKAdNetwork

If probabilistic attribution isn’t allowed by Apple, SKAd would be the solely supply of data to trace installs of customers that don’t opt-in. It’s already the case for some MMPs.

Do you might want to know which artistic, trade or artistic sort is performing higher? Properly, one factor is for certain: SKAdNetwork is not going to present this data by default. That being stated, utilizing the campaign_ID well with Smadex Versatile Mapping, you may obtain this objective.

Smadex’s Versatile Mapping

Smadex’s Versatile Mapping is the superpower advertisers have been ready for. With this new software, advertisers are in a position to arrange the 100 fields for campaign_IDs accessible on SKAd to map creatives, connection sort, advert measurement… or some other variable most related for his or her campaigns.  

With Smadex Versatile Mapping, advertisers can establish which mixtures of variables get one of the best efficiency with out uniquely figuring out the person and complying with Apple privateness guidelines. “It’s like constructing your personal algorithm primarily based on what you might want to know to optimize for SKAd. You are able to do it your self, straight from the Smadex platform”, highlights Jordi de los Pinos, CEO at Smadex.

And it will get higher. This new Versatile Mapping resolution for SKAdNetwork is totally built-in with our present reporting instruments. Within the under instance, one shopper used the Versatile Mapping to see which trade supplied extra SKAd installs, and the outcomes confirmed within the studies instantly.

Reporting – Versatile Mapping On Smadex

Advertisers face the problem on how decoding the SKAd Campaign_IDs to have the ability to map every variable. Smadex’s Versatile Mapping is the reply. “We let every advertiser select what variables they need to map for that campaign_ID subject. So, if you wish to map the advert measurement, you may. If you wish to map the connection sort, you may.”

Why Solely Utilizing SKAdNetwork for Marketing campaign Optimization Is Inadequate

As readability begins to type round how MMPs and advert networks plan to work with SKAdNetwork, we’d prefer to spend a while discussing in additional element how we will use Apple’s SKAdNetwork for efficiency advertising and marketing and why it’s essential to contemplate extra than simply this set of information for marketing campaign optimization.

What does it imply to solely use SKAdNetwork?

The important thing piece of information inside SKAdNetwork to assist with marketing campaign optimization is “ConversionValue” (conversion worth). This piece of information provides some indication of post-install efficiency and is distributed to the advert community by the app and reported on the marketing campaign degree. This conversion worth will be outlined utilizing early income, engagement, and retention occasions, or—most optimally—predicted LTV (pLTV).

An advertiser can determine to optimize their campaigns solely through the use of the conversion worth, making bid and finances choices primarily based on the variety of conversion values in any single marketing campaign or channel, successfully normalizing the efficiency throughout channels and campaigns. In follow, this might imply that one conversion worth from Fb could be valued the identical as one on, say, Unity Adverts.

For the rest of this text, we’ll assume that the ultimate conversion worth is distributed to the advert community inside the first 24 hours after set up, as per Fb’s definition. Subsequently, utilizing solely SKAdNetwork knowledge to optimize your cell advertising and marketing campaigns means you:

  • Solely optimize campaigns towards D0 ROAS or different D0 KPIs
  • Can’t replace marketing campaign ROAS primarily based on up to date cohort knowledge

When utilizing SKAdNetwork to optimize campaigns, it’s attainable to make use of solely the conversion worth to find out the allocation of bids and budgets. On this case, in the event you wished to contemplate the long-term efficiency of your promoting campaigns, you wouldn’t mannequin the anticipated income that marketing campaign drove, however solely mannequin the long-term ROAS of customers/cohorts as a operate of conversion worth by constructing a mannequin that maps conversion worth to LTV (i.e. if we all know that).

The above chart exhibits how an advertiser would try and map conversion worth to LTV, first by mapping conversion worth to D0 income, after which by extrapolating D0 income to D365 LTV (if that is your goal).

There are two vital inefficiencies when extrapolating conversion worth to LTV:

  1. Approximating D0 income from ConversionValue is difficult, particularly if the person can’t generate income inside the first 24 hours (e.g. an app that monetizes via subscriptions with a free trial). One of the best approximation for D0 income is a variety or cumulative view—e.g. $0-$5 or $10+.
  2. Utilizing your D0 income approximation to venture D365 income can be difficult because it doesn’t adequately mirror correct variations in income. It assumes all campaigns and channels have precisely the identical conduct. It will penalize excessive ROAS campaigns/channels and profit low ROAS campaigns/channels to the detriment of the portfolio returns.

The conversion worth, whether or not primarily based on early income, engagement, or predicted LTV (pLTV) acts as an early sign for advert networks to optimize towards. Ideally, we’d prefer to replace the anticipated ROAS (pROAS) for campaigns or channels primarily based on up to date person conduct, which might give us a extra knowledgeable view of the historic efficiency of campaigns. For instance, if we see that person LTV has modified after the primary 24 hours, it’s prudent to replace our understanding of the marketing campaign pROAS, despite the fact that that could be a number of days previously.

When utilizing solely conversion worth, nevertheless, to optimize SKAdNetwork campaigns, there’s no consideration for the underlying customers that make up these campaigns. Which means that as soon as the conversion worth is acquired from the advert community (by way of an MMP) and the D0 KPI is calculated, there’s no method to then replace this prediction primarily based on up to date information of every marketing campaign’s underlying person conduct.

The core problem with solely utilizing SKAdNetwork knowledge to optimize campaigns is that we don’t know what the underlying person conduct of a marketing campaign appears like. If we don’t have this knowledge, we will’t decide if customers or campaigns with the identical conversion worth really behave the identical. We should always then attempt to perceive what the underlying conduct of customers is within the marketing campaign to resolve this downside. When engaged on this downside, we’re not attempting to create a one-to-one mapping of install-to-campaign, that will be unimaginable primarily based on how SKAdNetwork works and towards the spirit of Apple’s privateness initiatives. It’s helpful, nevertheless, to make use of statistical strategies to create possibilities for which marketing campaign every set up may need come from.

What different knowledge units can be found?

We’ve mentioned why utilizing SKAdNetwork knowledge has such restricted utility, however we do have a number of different vital knowledge units accessible to us to assist mannequin predicted marketing campaign income.

These knowledge units are:

1. Anonymized (user-level) behavioral knowledge: It is a person’s in-app engagement and income knowledge all reported towards an nameless person ID. As a person engages with an app, this can proceed to be an considerable dataset encompassing all customers’ conduct all through their lifetime of utilizing that app.

It is a essential knowledge set to allow us to:

  1. Predict LTV for longer horizons than D1
  2. Replace historic marketing campaign predictions primarily based on up to date person conduct

2. Advert network-reported metrics: Primarily marketing campaign spend, clicks, and impressions.

3. Deterministic attribution from MMPs: The place we all know the attribution of a person, this knowledge reduces the quantity of statistical modeling we have to do to foretell which customers are coming from which campaigns. We will’t use this knowledge as the premise for a mannequin as this knowledge isn’t a consultant pattern however helpful nonetheless.

Once we incorporate the SKAdNetwork knowledge, the Anonymized (user-level) behavioral knowledge and the Advert network-reported metrics, we will mannequin the underlying person conduct of a marketing campaign. That is key to fixing the 2 core issues raised by solely utilizing SKAdNetwork postbacks that we highlighted earlier. Leveraging the nameless user-level behavioral knowledge, we will probabilistically attribute installs again to campaigns as a result of we all know the user-level conduct and might outline the conversion worth primarily based on this person in-app engagement knowledge. We will then map again to conversion worth knowledge from the SKAdNetwork postbacks.

Proven within the diagram above – the probabilistic attribution resolution (we should always word right here that this isn’t a so known as “fingerprinting” method) is the one use case for the info which leverages all three datasets to their full potential. Measuring efficiency utilizing solely SKAdNetwork conversion values, as defined earlier, ignores the richer set of person knowledge, leading to a stale somewhat than a dynamic image of marketing campaign efficiency.

One other potential resolution is to make use of media combine fashions (MMMs), which mannequin high-level KPIs (e.g. complete day by day installs, complete cohorted income) as a operate of extra granular inputs, corresponding to marketing campaign or channel finances allocation. In follow, MMMs have a excessive diploma of uncertainty, because it’s exhausting to get sufficient indicators to really measure the incrementality of particular person channels or campaigns, particularly when these results are altering over time and knowledge rapidly turns into outdated. This method would additionally ignore the bottom-up measurement functionality that SKAdNetwork offers.

Probabilistic attribution bridges the hole and offers advertisers with probably the most holistic measurement resolution given the info that’s accessible.

Leave a Reply