The Uptrix UA Glossary.
A working reference for growth managers. Not a marketing piece — the level of depth a PM running a paid funnel actually needs.
Why attribution matters
Attribution decides which channel, campaign, or creative gets credit for a conversion. Get this wrong and the budget follows the loudest dashboard, not the actual contribution.
Mobile attribution evolved from desktop's last-click model into something denser. Multiple ad networks, multi-touch funnels, and platform-level privacy changes mean the question is not just “who got the click” — it's “what actually moved the LTV.”
Multi-touch vs last-click
Last-click hands all the credit to the final touchpoint before conversion. It's simple, auditable, and almost always wrong. Multi-touch attribution distributes credit across the path — linear, time-decayed, U-shaped, or model-based — and reflects how users actually decide.
View-Through Attribution (VTA)
VTA credits an impression that the user saw but didn't click, if the user later converts. The window matters: a 24-hour VTA window is conservative; 7 days is generous; 30 days is generally a fiction.
- Click-through (CTA): requires a click. High intent, harder to game.
- View-through (VTA): impression only. Easy to over-attribute.
- Self-reported attribution (SRA): from SRNs like Meta/Google — opaque, but real.
SRNs vs DSPs
Self-Reporting Networks (Meta, Google, TikTok) attribute inside their own walled garden. DSPs run open programmatic supply and report through an MMP. The two views never quite match — and reconciling them is part of the job.
Uptrix doesn't treat attribution as a reporting problem. It's the input to decision-making. We weight signal by source, model the gaps between SRN and MMP views, and surface the channels actually compounding LTV — not the ones loudest in the dashboard.
What is Real-Time Bidding?
RTB is the auction layer of programmatic advertising. When a user opens an app, an impression is offered to multiple buyers simultaneously. Each buyer submits a bid; the auction resolves in under 100 milliseconds; the winning ad is rendered.
OpenRTB is the protocol behind most of this — a standardised request/response format that lets exchanges, SSPs, and DSPs talk to each other without bespoke integrations.
How an auction works
- Bid request: SSP sends device, app, and audience signals.
- Bid response: DSPs return a price and a creative.
- Auction resolution: highest bid wins, typically at second-price.
- Creative render: winning ad served, impression logged.
Floor prices & supply paths
Floor prices set a minimum bid an exchange will accept. Set them too low and revenue leaks; too high and demand walks. Dynamic floors — adjusted per impression and audience — are where modern yield work lives.
Uptrix runs pCVR-weighted bid logic — every bid is sized to the predicted probability of conversion, not to a flat budget. The auction works the same way; we just stop bidding on impressions the model already knows won't convert.
Performance metrics — the working set
A short reference of the metrics a working growth team is reading every day. Each one answers a specific question; together they form a picture, not a verdict.
We optimise to ROAS-on-day-7 and LTV-on-day-30. CPI is informational only — it tells you if your media buy is sane. It doesn't tell you if your business is working.
Mobile and web ad fraud — the working taxonomy
Fraud in mobile and web UA is engineered to look like a user. Bot traffic, hijacked attribution, fake installs from device farms. The cost is doubled: you pay for the fraud, and the fraudulent installs corrupt the optimisation signal downstream.
- Click spam: firing fake clicks to last-touch attribution.
- Click injection: malware claiming the install retroactively.
- Install hijacking: stealing credit from a legitimate network.
- Device-ID fraud: reset farms generating “new” devices on loop.
- CTIT anomalies: too-fast or too-slow click-to-install windows.
How Uptrix's fraud shield works
Publisher ID, IP, hosting provider, device fingerprint, and CTIT distribution are scored against historical baselines before the bid is even placed. Suspect traffic is filtered pre-auction, not flagged post-conversion — which means the budget doesn't spend on it and the model isn't trained on it.
Catching fraud after the fact is a refund problem. Filtering it before the auction is a performance problem solved. We treat fraud detection as part of bid logic, not as a post-mortem.
Want this depth on your account?
A working reference is one thing. Running it inside your campaigns is the work.