About

My name is Priya. Five years of push data, then I left.

I ran Mobidea's data science team from 2019 to 2024 — push-format attribution, audience-fatigue modelling, publisher quality scoring. I write here because the conversations I had with media buyers in private were ones the industry couldn't have in public while everyone was on someone's payroll. Let me show you the numbers.

Email meMonetize traffic
My name is Priya. Five years of push data, then I left.
Illustrated portrait of Priya Anand

Author

Priya Anand

Independent push-ad consultant (ex-Mobidea data science lead)

Lisbon, Mobidea, and a dashboard that didn't add up

I joined Mobidea in March 2019 as a junior analyst on the affiliate-side dashboard team. The push-format business was scaling fast — Mobidea was buying push inventory from a dozen networks and reselling segmented audiences to performance advertisers — and the data was a mess. Conversion windows were inconsistent across sources. Publisher IDs were aggregated differently per network. The attribution stack had been built incrementally over four years and no one person understood it. I spent my first eighteen months rebuilding the conversion-latency model from scratch, then got promoted twice in nine months when the rebuilt model surfaced $1.4M of misattributed revenue in a single quarter.

What five years inside the push stack actually taught me

Not what's in Mobidea's case studies. It's in the gap between panel CTR (which any network can show in real-time) and the seven-day-stabilised CR (which most advertisers never wait to see). It's in the publisher-quality distribution where the top 8% of sub-sources deliver 60% of human conversions and the bottom 15–20% of sub-sources deliver 40–60% of clicks but convert at near-zero. The distribution is bimodal, not normal — and any network that aggregates it into a 'premium / standard' bucket is hiding the diagnostic. The marketing deck calls that 'premium traffic.' The publisher-quality histogram calls it bimodal.

Why I left in October 2024

There was a $200K push campaign that failed in week three because the audience-fatigue curve collapsed at frequency cap 5/day. The data was unambiguous (n=12.4M, p less than 0.01). I was asked to remove the fatigue chart from the quarterly report and attribute the failure to creative quality. I pushed the chart back in, the report shipped unedited, and I resigned two weeks later. The writing started as cleaned-up internal memos shared with three former colleagues. It became a private Notion. It's now this site. I still run A/B tests for a small roster of clients across Tier-1 and LATAM push traffic. Owns three Python environments and refuses to consolidate them.

What I do here, on pushadsnetwork.com

Push-format analysis with sample sizes, GEOs, verticals, and date ranges attached. No 'premium traffic' label without a defined source. No A/B test write-up without n and p. No 'Top 10' piece where every network sounds the same — I name which ones win on which metric. When push is the wrong format for an offer, I say so. When the panel CTR is celebrated and the day-7 CR is hidden, I show both. Push CTR is a lagging indicator of nothing useful. CR at day 7 is the real number.

Professional track

  1. 2019–2024 Data science lead — push-format attribution + fraud Mobidea, Lisbon
  2. 2022–present Contributor — A/B testing methodology, attribution Affiliate-industry forums + conferences
  3. October 2024–present Independent consultant — push campaigns + A/B testing Push Ads Network + direct clients

How I evaluate a push network before I recommend it

A push network is not 'good' or 'bad' in isolation. It's fit or unfit for a vertical, in a GEO, against a defined attribution window. These are the four diagnostics I run before I put a client's budget on it.

  1. 1. Subscriber-list freshness and opt-in age distribution

    Push CR collapses as the subscriber base ages. I want opt-in age distribution by GEO — the share of subscribers under 30 days old, 30–90, 90–180, and 180+. Networks where 60%+ of subscribers are over 180 days old are running a decay curve, not a live audience. Day-7 CR on a 180+ day cohort runs 0.3–0.5x a sub-30-day cohort on the same GEO and vertical.

  2. 2. CTR decay curve from day 1 to day 14

    Healthy push subscribers show CTR decay of roughly 30–45% from day 1 to day 7, then stabilise. A network whose panel shows flat or rising CTR through day 14 is almost certainly running bot inventory — real human attention has variance and decay. I pull a 14-day CTR-by-day curve from any network before I commit budget.

  3. 3. Sub-source ID granularity and bimodality disclosure

    Sub_id1 through sub_id5 must be exposed in the panel and in postbacks. If a network buckets sub-sources behind 'premium' / 'standard' labels, I assume they're hiding the bottom of the bimodal distribution. The top decile delivers 50–65% of human conversions; the bottom 15–20% delivers 40–60% of clicks but converts at near-zero. Aggregating those is a diagnostic failure.

  4. 4. Smart-bidding threshold and rule-based fallback

    Smart CPA / Smart bidding products on push need a documented conversion floor below which the optimization model overfits on noise. The threshold I look for is 200 conversions in a single campaign — the same number Google Ads documents publicly for Smart Bidding and which generalises across platforms. Networks that auto-enroll campaigns into smart bidding from impression one are charging you for their model's learning curve out of your first month's budget.

Editorial line

Sample size, GEO, vertical, date — or it didn't happen

Every numeric claim attaches n, the GEO, the vertical, and a date range. 'We saw a meaningful lift' is not a finding. 'p=0.04, n=18,200, Tier-1 push iGaming, Q3 2024' is a finding.

Networks named, never [Network A]

PropellerAds, Adsterra, RichAds, Adcash, Monetag, AdPushup, Mondiad. Readers can name the field. So can I.

Bimodal distributions, not panel averages

The mean CR on a push campaign hides the diagnostic. The histogram tells the story — usually a tall left tail of low-CR bot sub-sources and a smaller right tail of high-CR human sub-sources.

Anti-hype when push is the wrong format

Push doesn't work for slow-consideration B2B SaaS, 30-day attribution windows, or creative that needs more than five seconds of evaluation. When your offer lives there, I'll say it.

Start today

A specific A/B-test or attribution question? Email me.

I work with a small roster of clients on push campaigns across Tier-1 and LATAM. For independent consultation, email me directly.

[email protected]Monetize traffic

Privacy

Your privacy choices

We use cookies to operate the site and, with your consent, to measure usage and personalize content. You can change your choices anytime.

Accessibility

Accessibility settings

Customize how the site looks and moves. Saved to this browser only.