View-through attribution

Definition

View-through attribution credits an install or conversion to an ad impression the user saw but never clicked, provided the conversion happens inside a view window that is usually much shorter than the click window.

It is the most contested setting in mobile measurement, because it is the one that most easily manufactures results. A click is an action the user took; an impression is something that happened near them. Both can be causal, but they are not equally good evidence, and a report that adds them into a single install number has quietly changed what the word install means.

A view is a weaker causal claim than a click

Click-through attribution requires the user to have done something. View-through requires only that an impression was recorded — and impression records are generated at enormous volume, cheaply, by parties who benefit from the credit. That asymmetry is the whole subject.

Because a click can carry a parameter through to your servers, click-through attribution can be deterministic. An impression cannot carry anything to a device the user never navigated from, so view-through is almost always resolved by matching device attributes or by a network's own logged-in graph — which is why self-attributing networks are where most VTA volume comes from.

The arithmetic of inflation

Take an app with 10,000 installs in a month, of which 6,000 are genuinely organic. Turn on a one-day view window across a large display campaign and a share of those 6,000 organic users will have been served an impression somewhere in the previous 24 hours. Those installs move columns. Nothing about the business changed.

This is also why VTA and organic move in opposite directions in almost every dashboard. When someone reports that organic installs collapsed the month a display campaign started, the first thing to check is not the campaign — it is whether the view window began reclassifying users who would have installed anyway. See effective CPI for the metric that survives this reclassification.

View-through on iOS is Apple's, not the network's

SKAdNetwork does support impression-driven attribution, but on Apple's terms. A signed ad impression rendered by the device can produce a postback even without a tap, and Apple distinguishes the two cases with fidelity-type: 1 for a StoreKit-rendered ad the user interacted with, 0 for a view-through impression. Where both exist, Apple prioritises the higher fidelity.

{
  "version": "4.0",
  "ad-network-id": "example123.skadnetwork",
  "source-identifier": "5239",
  "app-id": 525463029,
  "transaction-id": "6aafb7a5-0170-41b5-bbe4-fe71dedf1e28",
  "attribution-signature": "MEUCIQ...",
  "redownload": false,
  "fidelity-type": 0,
  "conversion-value": 12,
  "postback-sequence-index": 0
}

So on iOS you cannot set a view window, and you cannot mix the two claims by accident — the postback tells you which one it is. Split every SKAN report by fidelity-type before comparing it to anything. A campaign whose postbacks are overwhelmingly 0 is buying impressions and reporting installs, which may be fine, but should be a decision rather than a discovery. The SKAdNetwork postback page covers the rest of the payload.

Reporting view-through without lying to yourself

1. Keep VTA in its own column. Never a single "installs" figure that silently contains both.
2. Set the view window from evidence, not from the network's default. One day is already generous for most categories.
3. Run a holdout at least once a quarter on any channel where VTA is more than a fifth of claimed installs.
4. Watch the organic line. A campaign that only ever moves the split between paid and organic is not growing anything.
5. Deduplicate before you sum. The same install can qualify for a click credit on one network and a view credit on another.

-- One row per install, with the strongest touch that qualified.
-- Clicks outrank views; views are only credited when no click exists.
SELECT
  campaign,
  COUNTIF(touch_type = 'click') AS click_through_installs,
  COUNTIF(touch_type = 'view')  AS view_through_installs,
  SAFE_DIVIDE(
    COUNTIF(touch_type = 'view'),
    COUNT(*)
  ) AS view_through_share
FROM (
  SELECT
    i.install_id,
    t.campaign,
    t.touch_type,
    ROW_NUMBER() OVER (
      PARTITION BY i.install_id
      -- click beats view first, then recency
      ORDER BY (t.touch_type = 'click') DESC, t.touched_at DESC
    ) AS rn
  FROM installs i
  JOIN touches t
    ON t.device_match = i.device_match
   AND t.touched_at <= i.installed_at
   AND t.touched_at > i.installed_at - (
         CASE t.touch_type WHEN 'click' THEN INTERVAL 7 DAY
                           ELSE INTERVAL 1 DAY END)
)
WHERE rn = 1
GROUP BY campaign
ORDER BY view_through_share DESC;

The last column of that query is the one worth putting on a wall. A channel whose view-through share climbs month over month while total installs stay flat is absorbing credit, not creating demand — and the attribution window settings behind it are usually where the change happened.