Skip to content
AppDues
Log in
Contact salesSign up

HDYHAU surveys for subscription apps: a decision guide

Asking new subscribers where they first encountered your app feels like the simplest attribution method available. It is also one of the easiest to over-trust. This guide explains what a 'How did you hear about us?' (HDYHAU) survey can legitimately tell you, where it stops being a causal signal, and how to replace it with measurement that survives a broader media mix.

Why teams reach for HDYHAU in the first place

Subscription apps often earn most of their revenue on iOS, where conversion and retention tend to be stronger but deterministic attribution is thin. Probabilistic methods fill some of that gap, yet they still leave teams comparing numbers produced by incompatible methodologies. A single in-product question sidesteps the comparison problem: every respondent answers the same prompt inside your own app.

The question usually appears during onboarding or at the point of purchase. A typical option set covers Facebook, Instagram, TikTok, YouTube, Google, Friend, and Other. Some teams shuffle the order to discourage reflexive tapping; in the original author's experience, that shuffle did not meaningfully change the distribution of answers.

For a young app running one or two social channels, this data can be genuinely useful. It surfaces conversions that probabilistic systems miss and often lines up more closely with blended revenue than network or mobile measurement partner (MMP) reporting does. The trouble starts when the channel mix widens.

Using HDYHAU alongside social ad reporting

The survey's real value early on is as a like-for-like comparator. If you chart weekly HDYHAU answers in a product analytics tool such as Amplitude or Mixpanel and place campaign and MMP numbers beside them, you get a quick read on whether your paid channels are under-reporting.

That gap is common. Probabilistic methods such as TikTok's ADC or Meta's AEM typically record fewer conversions than actually occurred, which can push teams toward the wrong paid UA decisions. Comparing those figures against first-party survey answers gives you a directional check on over- or under-attribution by channel.

Marketers who use this comparison usually weight the first-party answers more heavily than network or MMP data, on the reasonable assumption that users know their own behavior. When campaigns are switched on, the blended result often moves in the direction the survey data predicted.

Worked example: how an under-counted conversion base inflates CAC

Suppose your daily HDYHAU count runs about 95 installs above what Meta reports. At a 5% conversion rate, that is roughly five subscriptions per day, or about 150 per month, missing from the network view.

Scenario A — no survey in place. You spend $10,000 in a month. Meta and MMP data together report 200 purchases, so CAC looks like $50. With LTV of $30 at day 60 and $50 at day 90, payback arrives around day 90.

Scenario B — survey in place. The same $10,000 now maps to 350 subscriptions once the first-party count is included, so CAC is about $29. Payback arrives around day 60.

The difference matters most for teams whose growth depends on cash flow. A CAC that is overstated by roughly 40% changes how aggressively you can scale and how you prioritize weekly campaign optimization. Note that you can also group subscription events by HDYHAU answer directly in your analytics tool; the example above keeps the arithmetic simple on purpose.

Two failure modes that appear as you scale

The first flaw is structural: the clean comparison breaks down once you diversify paid channels. The second is cognitive: the survey depends on users accurately recalling and reporting where they first heard about you.

At high spend levels, users encounter your brand repeatedly across many surfaces before they install. When they tap 'Facebook' or 'TikTok,' they are often reporting what they believe influenced them or which option feels most familiar, not a traceable causal path. At that point the survey is measuring perceived influence and brand recall under pressure, not attribution.

This is why the distribution of your media spend tends to shape the distribution of answers. A mix of 60% Meta, 20% TikTok, and 20% Google will produce a survey result that mirrors those proportions even after you add a new channel. The new channel's effect hides inside 'Other' or disappears entirely.

Example: a DSP campaign that left no trace in the survey

Imagine an app spending roughly $50,000–$60,000 per month, with 90% on Meta and the remainder on TikTok. The team launches AppLovin for the first time in November and watches the 'Other' bucket in the HDYHAU survey. This scenario mirrors a campaign the original author reported running for a new client, so the figures are illustrative rather than a fresh test.

After more than ten days and more than $10,000 in DSP spend, the 'Other' baseline had not moved. The MMP, using last-click, credited the DSP with over 200 conversions, so an uplift should have been visible. It was not.

The explanation is mechanical. DSPs like AppLovin buy inventory across thousands of apps and show ads in contexts far from social feeds — mid-game, mid-scroll, mid-reward. When those users later reach the survey, they pick the option that feels most familiar. Meta, Google, and TikTok benefit from that bias because they are culturally dominant and top-of-mind, so they remain the safe choice even when they were not the discovery channel.

The downstream effect is a budget decision made on distorted data: the recognizable platforms look stronger than they are, and the DSP looks like it failed when it may not have.

When to trust the survey and when to treat it as context

The survey is not useless; it is narrow. Use the following split as a starting checklist for your own account.

Reliable enough to act on:

  • Paid acquisition is concentrated in one or two major social platforms.
  • Spend is relatively concentrated, for example mostly Meta or mostly TikTok.
  • Your goal is detecting under-attribution in probabilistic systems.
  • You are early-stage and optimizing for speed over precision.
  • You want directional signals on shifting channel familiarity or brand awareness.

Likely to mislead:

  • Multiple channels run at once, including social, DSPs, and search.
  • Brand presence is already high across platforms.
  • Users are exposed to ads outside recognizable environments.

A second check is timing. If you add a new channel and the survey distribution does not move within a few weeks while MMP last-click credits that channel with meaningful volume, treat the survey as context rather than proof. The table below reflects the conditions described in the original article published in December 2025; channel behavior and platform measurement change, so check current platform documentation before applying it to a live account.

Measuring DSP impact without relying on recall

Once DSPs are in the mix, the question shifts from 'where did users say they came from' to 'what changed because we spent money.' The methods that answer the second question are slower and less intuitive, but they measure causality rather than memory.

Approaches worth running:

  • Incrementality tests with geographic holdouts.
  • Incrementality tests with ads on versus off.
  • Incrementality tests using budget ramps.
  • Blended performance analysis across all channels.
  • SKAN analysis, with the caveat that it tends to under-attribute.

A practical pattern is to hold spend steady on your established channels for a few days, then raise DSP budget substantially. Compare the conversion baseline before and after the DSP launch and credit the difference through your MMP. Because DSPs usually enter after the classic channels already carry meaningful volume, you do not need to pause your main growth driver to run this test.

First-party does not automatically mean correct

In a privacy-constrained landscape, first-party data feels like safe ground. It is not automatically accurate. The useful discipline is to fit your measurement framework to the current channel mix, then watch how it behaves when a new channel enters.

Surveys can keep a role as qualitative context: they show what users believe, which channels feel familiar, and how brand awareness shifts. They should not be treated as evidence of incremental revenue. If the decision is whether spend is working, use incrementality tests, controlled experiments, and blended performance analysis.

Survey reliability by account situation
✅ 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

Conditions are summarized from the original article published in December 2025; verify against current platform documentation before applying them to a live account.

Frequently asked questions

Is a 'How did you hear about us?' survey a form of first-party attribution?

Yes, in the sense that the answer is collected inside your product from the user directly. That does not make it causal. It records what the user recalls or believes, which can diverge from the channel that actually drove the install.

Why does the survey over-credit Meta, Google, and TikTok?

Those platforms are culturally dominant and top-of-mind. When a user is asked to pick from a list under time pressure, the familiar brand is the safe choice, even if the ad they saw was on a DSP or another surface.

Should I remove the survey once I start running DSPs?

Not necessarily. It can still serve as qualitative context on brand awareness and channel familiarity. What should change is the weight you give it in budget decisions; causal questions need incrementality tests or blended analysis.

Does randomizing the answer order change the results?

In the experience described in the original article, randomizing the option order did not meaningfully change the distribution of answers. It may still be worth testing in your own app, since results can vary by audience and option set.

What is the simplest incrementality test for a new DSP?

Hold spend steady on your established channels for several days, then raise DSP budget substantially. Compare the conversion baseline before and after the DSP launch and attribute the difference through your MMP.