Measuring Creator-Driven Installs

I measure creator-driven installs with a unique code as the floor, App Store or Play downloads against a written baseline as the cash number, and platform view-through as a ceiling I do not pay on. Views are not UA. I still use a Minimum View Guarantee on the phone ($2000 for 4 posts with 1 million views is the shape) because views are at least public. Then I measure installs separately.

Before anyone posts: screenshot 7-14 days of downloads, give each creator their own code, do not stack Product Hunt on the same morning. After: 48 hours and 7 days versus baseline, subtract known ads, label the rest estimated lift. If you cannot name the decision (repeat, Spark, stop, or a bonus), you will collect TikTok analytics and call it data.

JumpTrend will not measure this for you. Matching makes a later graph more interpretable. It does not produce a CPI. Attribution stays in your store, your ads accounts, and a spreadsheet you can explain to a skeptical friend.

Decide what the measurement is for

Pick one:

Go/no-go on a second post with this person.
Go/no-go on Spark budget.
Pay a bonus.
Report to an investor.

(1) and (2) are founder work and tolerate estimates. (3) and (4) require a written definition everyone saw in advance. Mixing them is how relationships end.

If you cannot name the decision, you will collect screenshots of TikTok analytics and call it a culture of data.

Before anything goes live

Write a baseline: daily downloads and product page views for 7-14 days, plus whether ASA or other ads were on. Screenshot App Store Connect (or Play Console) so you cannot later memory-hole the quiet week.

Give the creator an identity you control: unique offer code, unique link if the platform and Apple/Google rules allow, unique campaign if you will Spark. Put the identity in the brief.

Do not stack a launch event on the same day unless you have given up on isolation.

Note the listing conversion rate. You will need it to interpret “views went up, downloads didn’t.”

Organic posts: a humble model

You do not have a campaign ID. You have a timestamp.

Watch product page views and downloads for 48 hours and for 7 days.
Subtract known paid campaigns.
Compare to baseline noise. If your daily downloads already jump around a lot, you cannot call a small bump a win.
Count code redemptions as a floor (people who bothered).
Read comments and support mail for the handle (qualitative).
Label the whole thing estimated lift.

If estimated lift is indistinguishable from noise, you may still have bought an asset (the file) if usage rights exist. That is a production outcome, not a UA outcome. Say so.

Do not apply a “view-through rate from a blog” to their view count. That would be inventing installs.

Promo codes specific to influencers are a good idea. That way you can track who is doing well and who to cut. If you do not use codes, try to get creators to post at different times and correlate the post date with purchases.

Paid posts (Spark, whitelist, dark UGC): a less humble model

Now you have an ads object. Use it.

Optimize for the install (or current equivalent) objective, not for video views.
Read click-through installs as a floor.
Read view-through as a ceiling (see view-through attribution).
Read App Store proceeds in the same window as the register.
Put talent fee + media + JumpTrend allocation in the denominator for ROAS.
Use a payback window you can survive (30 or 60 days), net of store cut.

SKAdNetwork and privacy windows will disagree with the ads UI. When they disagree, the register still wins. You cannot deposit SKAN into a bank.

The conversion-rate fork

Creator traffic is colder than App Store search. A dip in blended conversion after a post can mean “social arrived,” not “the listing broke.” Split by source if the console lets you. If product page views rise and conversion collapses beyond what you expect for social, the video promised a product the screenshots do not show. Fix the mismatch; do not immediately fire the only creator who actually sent humans.

What to put in the contract if money depends on numbers

If you must do affiliate or a bonus:

Define the event (code redemption, net paid sub after refunds, not “platform VTA”).
Define the window.
Define who pulls the report.
Cap it.
Do not use a metric only one party can see.

Many app deals should stay flat or hybrid because iOS will not give you Shopify-grade affiliate truth. That is not cowardice. I like a Minimum View Guarantee when I am buying posts, because views are at least public. Then I still measure installs separately. Views are not UA.

A one-page log you can reuse

Date posted / handle / organic or Spark / code / baseline D7 downloads / D7 downloads after / page views after / code uses / media spend / talent fee / notes on comments / decision (stop, repeat, boost).

If this log feels too simple, good. You are a solo founder. You do not need an MMP on day one. You need to stop gaslighting yourself.

How JumpTrend fits

Matching quality is an input to measurement: a feed-qualified creator makes a lift more interpretable (those people were in the niche). It does not make a lift automatic. Scout is $1,500. Outreach is $2,500. Dealmaker is $5,000. Talent extra. No download guarantee, which is the only guarantee structure that is compatible with this pillar.

When a vendor promises a CPI on a landing page, they are volunteering to be the person this essay is warning you about.

Entity version: Measuring creator-driven installs. Neighboring: ROAS for indie apps, App Store conversion rate, Organic vs paid creator posts.

Worked example (numbers are hypothetical, labeled as such)

Suppose baseline is ~12 downloads a day with ASA off. A micro posts Tuesday. Wednesday-Thursday you see 19 and 16 downloads, then you are back to 11-13. Codes redeemed: 7. Product page views had a bump. Spark was not used.

You may write: “Estimated organic lift of a few dozen downloads in 48 hours, 7 of which showed the code. Not enough to call a channel. Enough to justify a second brief if comments looked like users. Not enough to pay a performance bonus.”

If instead you Sparked $200 and Ads Manager shows 80 click-through installs and 240 view-through, while App Store Connect shows ~90 extra downloads in the window versus baseline: believe the register (~90), treat 80 as a consistent floor, treat 240 as a story. ROAS uses net proceeds from those users over your payback window, minus $200 media minus talent minus a slice of the month’s JumpTrend invoice.

No part of that example is a JumpTrend result. It is a method. Your numbers will be different. If you cannot fill the sentence, you are not measuring yet. One quiet post is information, not a funeral.