Uncovering where users first heard about your app and how they came to download it is an age-old problem for app businesses. Tracking attribution is key to tapping into your best user acquisition (UA) sources, but with advertising spread across multiple channels and things like word-of-mouth hard to track, teams are often left in the dark.
‘How did you hear about us?’ (HDYHAU) questions seem to solve this problem, but are they really the solution?
Why use HDYHAU?
Many subscription apps face growing data discrepancies when it comes to tracking attribution. This is largely because iOS remains the primary source of revenue — but while these users typically have higher conversion rate (CVR) and stickiness, iOS offers limited attribution tracking. Structured frameworks for modelling attribution on iOS can help, but that’s only part of the equation.
To address these gaps, teams have opted for a simple way to measure the impact of ads: directly asking users “How did you hear about us?”. This question, commonly shown during onboarding or purchase flows, allows users to self-report their acquisition source. The answer options typically include:
- TikTok
- YouTube
- Friend
- Other
(There are some apps that randomize the order of the options to ensure answers from users are intentional, but in my experience, this doesn’t tend to impact the distribution of the answers.)
For early-stage teams running a small number of social channels (often Meta or TikTok) HDYHAU data has been a gamechanger; helping surface under-attribution in probabilistic systems and align performance with blended results. In these cases, self-reported data often appears more consistent than network or mobile measurement partner (MMP) reporting. However, as apps scale and expand into additional channels, HDYHAU responses become increasingly unreliable as an attribution source (more on this later).
Using HDYHAU for social ad attribution
The HDYHAU question fixes a common issue that all advertisers have with iOS: it gives the possibility of comparing like-to-like in a landscape where every attribution methodology works in a different way. If you use tools like Amplitude or Mixpanel, you can have graphics like this:
As you can see, this graphic is simply showing the answers for the HDYHAU question on a weekly basis, so if you take the campaign data, the MMP data, and compare them against this first-party data, you have a very simplistic way of seeing if your campaigns are underattributing or not.
According to my experience, when you use probabilistic methodologies like ADC for TikTok or AEM for Meta, you will likely track less conversions in your campaigns than have actually happened — leading you to make wrong paid UA decisions.
However, with HDYHAU, you quickly solve that problem by comparing both sources, so you can see if certain attribution channels are over- or under-attributed. That looks like this:
Since you (understandably) expect the user-reported data to be most accurate, marketers using HDYHAU as an attribution source typically place more importance on this first-party data than the MMP or ad network data. In most cases, they see greater impact on the blended results when they enable campaigns (like in the screenshot above, from a current project of mine).
Example: how HDYHAU impacts revenue
In my example above, the average difference between HDYHAU and the installs reported by Meta on a daily basis is 95. Assuming you have a 5% CVR, you’re roughly missing ~5 subscriptions per day (150/month) comparing the two attribution sources.
Note: You can also have direct subscription reports on Amplitude, if you group users by the answer they gave in the HDYHAU step, but for the sake of this example let ’s keep it simple.
Let’s consider two scenarios where we follow this first-party data:
Scenario one:
- On day 90, your app has a lifetime value (LTV) of $50; on day 60, it has an LTV of $30
- You don’t have HDYHAU in place, so you’re relying on MMP and ad network data for attribution
- You spend $10k on paid campaigns during one month, then crosscheck Meta and MMP data, reporting 200 purchases — resulting in a $50 customer acquisition cost (CAC)
- In this scenario, your LTV would cover that CAC in 90 days, so from day 90 you would start to make profits
Scenario two:
- Imagine the same scenario, but you do have HDYHAU in place:
- You see the same difference on conversions as in my example above (~150/month)
- With your first-party data in place and spending $10k on paid, you report 350 subscriptions in the same period of time, giving you a CAC of ~$29
- In this scenario, you begin making profit from day 60
In a business model where cash flow is critical, this difference is hugely impacting every decision on weekly optimization of paid campaigns, and how aggressively you can scale them.
This is why first-party data has become a highly-valued data source for app teams relying on social ads. In a world shaped by ATT, SKAN, and privacy constraints, it’s one of few reliable ways of ensuring paid campaigns get the credit they deserve — because the answer is coming directly from users, rather than depending on probabilistic attribution methodologies that rely on fingerprinting.
At a first glance, HDYHAU feels like the perfect solution to address all this attribution mess happening in iOS. No black boxes, no delayed postbacks, no large discrepancies. Just a direct answer from your user collected within your product.
But there’s two significant flaws in this:
- The easy-to-read attribution breaks down when you diversify paid ad channels
- And what’s more, HDYHAU relies on users accurately recalling and recording where they first heard about you
When HDYHAU fail as attribution for non-social ads
Most companies begin with the classic ad networks (Google, Meta, TikTok) when they start paid UA, and using first-party data with this setup is smart — it’s likely more accurate than relying on probabilistic attribution. But when you have a large budget for these channels and run alternative ad channels (e.g. a demand-side platform (DSP)), HDYHAU begins losing accuracy.
On a high level of spend, you have a large presence across social media due to paid ads and multiple channels — users will likely have come across an advert, logo, or app name long before they actually download the app. So when they click ‘Facebook’ or ‘TikTok’ as where they first heard about you, they’re actually selecting:
- What they think influenced them
- What options feel familiar to them
At this point, HDYHAU surveys aren’t capturing causal attribution; they’re identifying perceived influence, brand familiarity, and recall under pressure.
Example: how distribution of spend impacts HDYHAU
This fundamental flaw of HDYHAU at a large business level is also why the distribution of media mix and ad spend will likely affect the distribution of answers.
For example, if your media mix is 60% Meta, 20% TikTok, 20% Google, you’ll likely have a HDYHAU attribution that reflects this, even after running campaigns in an alternative channel. This interpretation becomes critical the moment you introduce these channels into your media mix.
This screenshot is real data from a recent Applovin campaign I ran for a new client. They have a rough $50–60k spend, with 90% to Meta and the remainder to TikTok. We started Applovin for the first time in November and began monitoring the answers to ‘Other’ in the HDYHAU survey.
As you can see, there’s no difference in the baseline after running the campaigns for more than 10 days and spending more than $10k (the minimum budgets in these platforms are much higher). The MMP reported +200 conversions to the campaigns based on last click, so we would have to see an uplift in this baseline.
The reason behind this is simple — platforms like Applovin operate very differently from traditional social ads: they buy inventory across thousands of apps, and show ads beyond social media platforms e.g. playing a game, scrolling, or waiting for a reward. So later, when users reach the HDYHAU question, they just select the most familiar option.
Meta, Google and TikTok benefit massively from this bias, since they’re culturally-dominant and top-of-mind. So even if users didn’t discover the app on these platforms, they’re a safe choice.
The result? HDYHAU surveys over-attribute conversions to the most recognizable platforms → you draw conclusions based on inaccurate data → you base budget allocation and channel optimization decisions on this.
TL;DR: when to use/avoid HDYHAU for attribution
When HDYHAU is treated as accurate attribution, demand-side platforms are set up to (look like they) fail. This doesn’t make HDYHAU surveys useless, but it does limit their use to specific scenarios. Here’s a summary:
| ✅ When HDYHAU surveys can be reliable: | ❌ When HDYHAU may be misleading: |
|---|---|
| – Paid acquisition is limited to one–two major social platforms – Spend is relatively concentrated (e.g. Meta or TikTok) – The goal is detecting under-attribution in probabilistic systems – Teams are early-stage and optimizing for speed over precision – You want to get indicators of shifting channel familiarity and trends – You need directional insights on brand awareness | – Multiple channels run simultaneously (social + DSPs + search) – Brand presence is already high across platforms – Users are exposed to ads outside recognizable environments |
How to measure attribution with DSPs
If you read to this point you’re probably wondering what you should rely on for attribution. HDYHAU surveys are still useful, but only on the small–medium level. Exploring how to measure attribution for larger ad budgets is another article in itself — but broadly-speaking, these are the most effective approaches to see if DSPs are really bringing impact:
- Incrementality tests with GEO holdouts
- Incrementality tests with Ads ON/OFF
- Incrementality tests with budget ramps
- Blended performance analysis
- SKAN analysis (less effective because it tends to underattribute)
In my opinion, it’s valuable to run incrementality tests where you maintain the classic channels with a stable spend for a few days, and then you substantially increase the budget on the DSP. You can then measure how the baseline of conversions changed before and after starting the campaign in the DSP — and attribute the difference in the baseline to the MMP.
With this approach, you don’t need to pause your main growth driver, since you normally start with DSPs when you’ve already reached a nice volume of spend on the classic channels.
Yes, these methods are slower, less intuitive and harder to justify but they measure causality, not memory.
The value of HDYHAU is how you use them
In the current privacy landscape, first-party data feels like a safe fallback; but data being first-party doesn’t automatically make it correct. The work of growth teams and marketers is to adapt your framework to track data to the current channel mix and then compare how it evolves when a new channel is introduced.
If you want to understand whether your spend is working, you need measurement frameworks built for causality: incrementality tests, controlled experiments, and blended performance analysis. Surveys can still play a role, but only as qualitative context — not as evidence.