How to Choose a CPI Ad Network
Every CPI network's landing page says the same things: quality traffic, transparent reporting, global reach. The pages are interchangeable; the operations behind them are not. Here is how to evaluate a CPI network on what actually determines your results.
First question: whose inventory is it?
A CPI network is either selling you inventory it controls or inventory it buys at auction and resells. Both models can work, but they differ in what the network can actually see and enforce. Owners of inventory can answer placement-level questions: which apps your ads run in, what those placements earn, which ones get cut and why. Resellers see exchange averages, and averages are where bad placements hide.
Ask directly: what share of volume comes from owned or directly-managed inventory? Can I get placement-level reporting? If the answer to the second is no, the answer to your next quality dispute will also be no — there's nothing to show you. This single question filters the field faster than any other.
Measurement: your MMP, your rules
The referee of a CPI campaign must be your measurement stack, not the network's dashboard. Serious networks integrate with AppsFlyer, Adjust, Branch, Kochava and Singular, accept your attribution verdicts, and write the commercial rules for rejected installs into the agreement before spend starts: which fraud rules are active, what happens to rejected conversions, which attribution windows apply.
Red flags are consistent across bad actors: pressure to optimize on the network's internal numbers, vague answers about rejection handling, resistance to documented windows. A network that trusts its traffic has no reason to fear your tracker. We detailed what 'good' looks like on our traffic quality page — use it as a checklist against anyone you evaluate, us included.
Fraud policy is economics, not PR
Mobile fraud is not an edge case; it's a permanent feature of the ecosystem that reprices your CPI silently. The question is never 'do you have fraud' — everyone touches it — but 'what happens commercially when it's found'. Who absorbs rejected installs? How fast are bad placements removed? Is filtering applied before your tracker sees traffic, or after invoicing?
Ask for the filter categories by name: emulators, click injection, click flooding, SDK spoofing, install farms, incentivized abuse. A network with real filtering lists them without hesitation, because the list is the product. A network without it says 'industry-leading quality' and changes the subject.
Make them justify the pricing model
CPI is one model among several — CPA, CPM, CPL, hybrid, revenue share — and each distributes risk differently. CPI is efficient when your monetization event is frequent and close to the install; it's a trap when value arrives weeks later, because the network gets paid at a point where quality is invisible. A trustworthy partner asks about your funnel before quoting a model and will sometimes talk you out of CPI entirely.
The deeper signal: does the network ask what happens after the install? Buyers who plan to deliver value ask about retention, payers and payback windows, because they'll be judged on them. Buyers who plan to deliver installs ask about volume and budget. The questions in your first call predict the results in your first quarter.
Run the pilot like an audit
Never commit scale to an untested network; never judge one on an unstructured test either. The right pilot has a defined budget, a defined measurement window, source-level reporting access from day one, and pre-agreed success criteria on your billable event — not on install count. Include your incumbent channel in the same window if you can: identical measurement makes the comparison honest.
Watch behavior during the pilot as much as results: does the network proactively show you what it cut and why? Does reporting arrive at the promised granularity? Pilots reveal operations, and operations are what you're actually buying. When you're ready to run one with us, the advertiser registration starts the audit — and read our companion guide on choosing a mobile ad network for the full framework.
Contract terms that protect you
The evaluation doesn't end at the dashboard; it ends in the agreement. Four clauses decide how the relationship survives its first dispute. Rejection handling: which MMP verdicts are final, and how rejected installs are credited — future invoice or never billed are very different answers. Payment timing against reconciliation: net-30 on unreconciled volume quietly makes you the fraud insurer. Makegoods: when a source underdelivers quality mid-flight, what replaces it and when. Termination: the right to pause within days, because the campaigns that need pausing never announce themselves a month ahead.
A network confident in its traffic negotiates these clauses quickly — they've had the conversation a hundred times and their answers are pre-built. Hesitation here is information too. The contract is the fraud policy with teeth; everything on the landing page is the fraud policy without them.