Shumel Lais, co-founder of Day30 and previously founded Appsumer (acquired by InMobi), explains why most subscription apps feed ad platforms the wrong goal, how precision and recall reshape signal selection, and what a realistic measurement maturity ladder looks like in 2026.

Shumel walks Jacob through the five stages of measurement maturity, from apps that just compare App Store Connect revenue to ad spend, through MMP attribution and cohorted reporting, up to incrementality testing for the largest spenders. He breaks down why signal engineering only makes sense once you have the right foundation in place, shares the 10-conversions-per-campaign-per-day rule of thumb for when to go further down funnel, and unpacks the restaurant booking app mistake that first put him onto the precision/recall framework.

What you'll learn:

Why optimizing to cost-per-trial leaves money on the table for most subscription apps
How Meta's 7-day visibility window forces the signal engineering problem
Why recall, not precision, is the metric most marketers overlook
Why the restaurant booking app example was Shumel's own mistake, and what it taught him
How Meta's event-day reporting can hide renewals inside new purchase counts
Why server-side events struggle more with matching than client-side events
How to decide between revenue-value signals and binary convert/no-convert signals
Why subscription apps are years behind gaming on analytics maturity
The 10 conversions per campaign per day floor before attempting signal engineering
When LTV curves become reliable enough to extend payback from 30 days to 6+ months

Key Takeaways:

Signal engineering is closing the gap between what the platform can see and what you actually care about. Meta sees 7 days. You care about month 3 revenue.
Recall is the metric most teams forget to measure. Precision tells you if the users firing your signal convert. Recall tells you what share of your actual converters it captures. A signal with 90% precision and 40% recall tells the algorithm that 60% of your good users are bad.
There are five levels of measurement maturity, and most apps skip steps. ASC comparison → platform attribution → MMP → cohorted reporting → incrementality. Signal engineering is a level 3 or 4 exercise. Attempting it earlier wastes the effort.
The 10-conversions-per-campaign-per-day rule. Below that, Meta cannot learn from a more selective signal. Above 30 to 40 per day, you are leaving performance on the table by not going further down funnel.
Meta reports on event day, not install day. Renewals fire as purchase events, so Meta can claim credit for users who were already paying. Without install-cohorted MMP visibility, you are paying to acquire users you already had.
Speed of signal affects matching quality and algorithm learning. Events sent within 24 hours have more matching parameters, and they let Meta decide if a user is good without waiting 7 days for the purchase to come through.
The restaurant booking app was Shumel's own mistake. Before Day30, he optimized toward behaviors that correlated with bookings but were not causal. Performance did not move. The fix was cohorts, observation windows, and a binary prediction statement.
Measurement problems are not an excuse anymore. In 2026, the tools exist and the playbooks exist. Hiding behind attribution gaps is a choice, as is hiding behind blended CAC when direct CAC is uncomfortable.

Links & Resources

Day30: https://day30.ai
Shumel Lais on LinkedIn: https://www.linkedin.com/in/shumellais/

Timestamps
00:00 Shumel's background and early mobile agency days
00:56 The signal engineering framing and how Day30 landed on it
03:30 A basic example: trials vs trials plus behavior
05:56 Why signal engineering exists (attribution gap, not just subscriptions)
08:45 Signal volume as the second dimension after precision
09:30 Defining recall and the photo storage app example
15:58 When to send revenue values vs binary convert/not-convert
16:41 The restaurant booking app mistake and causation vs correlation
19:33 Experiments are still the only real proof
20:00 Measurement maturity level 1: no MMP, just ASC
22:37 Do you actually need an MMP to start?
23:39 Level 3: why MMP matters (Meta's event-day reporting trap)
25:37 Level 4: cohorted metrics and aligning on day-30 ROAS
26:30 Level 5: incrementality and MMM for the largest spenders
27:35 The 10 conversions per campaign per day threshold
29:30 Why the MMP matters for signal engineering (measurement, not the signal itself)
31:03 MMP vs Conversions API for sending signals
33:04 SDK vs server-side: matching and speed
36:43 Payback periods and when to extend them
40:32 Simple inputs for a basic predictive LTV model
42:52 If you're running Meta to CPT today, what do you change first
44:41 The quantity vs quality of signal tradeoff
46:48 Hot takes: no more hiding behind attribution
48:02 Favorite pricing and packaging tactics seen recently
50:08 Day30's free signal audit offer