Why do most traffic source tests waste money?

Because the rules get written after the spend. A new network goes live with a vague budget, tracking that was “mostly working” and no definition of success, and three weeks later the team argues over whether it worked. The money was spent learning nothing.

A useful test answers one question, “Can this source deliver conversions at or above our floor, at a volume worth the effort?”, and it fixes the budget, the tracking, the quality checks and the decision rules before the first impression. Everything below is about doing that in order.

How do you size the test budget?

Derive it from your unit economics, then cap it. This guide gives no numbers because the right ones depend on your payout or margin, your conversion rate and how long value takes to arrive. The method:

  1. Set the floor. The most you can pay per conversion and still make money once cohort value matures. If you sell subscriptions, see CAC payback and cohort ROAS.
  2. Set the read size. How many conversions you need before a result is more than noise. Fewer conversions means a less certain read; decide in advance how much uncertainty you will accept.
  3. Multiply. Read size times floor cost per conversion gives the spend needed to read the source at break-even.
  4. Add the learning cost. New sources usually run worse while you prune placements and bids. Allow for it explicitly.
  5. Cap it. The total is the most you are willing to lose. Write it down as a hard cap.

If that cap is more than you can afford to lose, narrow the test (one geo, one format, one offer) rather than running an underpowered test across everything.

What tracking has to work before launch?

Click IDs and server-to-server (S2S) postbacks, tested end to end. Every click the network sends should carry its own click ID in the landing URL. Store it with the session, carry it through signup or checkout, and when a conversion you accept occurs, fire a postback from your server to the network’s endpoint with that click ID and the event.

Why this matters:

  • It doesn’t depend on the browser. Blocked scripts and cleared cookies don’t break it.
  • You decide what counts. Send only validated conversions (approved lead, first payment), so the network optimizes toward value, not form fills.
  • You can reconcile. Your records and the network’s should match by click ID. If they don’t, you can see which clicks diverge.

Before spending, run test clicks through each creative and placement type and confirm the click ID arrives, survives redirects, and returns in the postback. Also tag traffic with source, campaign and placement parameters so your analytics can break results down independently of the network’s reporting. Your own system is the source of truth; the network’s dashboard is a second opinion. This mirrors the reconciliation we describe in pay-for-performance media buying.

How do you check for bots and invalid traffic?

Use the industry’s definitions, then look for the patterns in your own data. The Media Rating Council’s IVT standards addendum splits invalid traffic into two categories:

Category How it is detected Examples from the MRC
General invalid traffic (GIVT) Lists and standardized parameter checks Known invalid data-center traffic, declared bots and crawlers, non-browser user agents, pre-fetch traffic, 0x0 or 1x1 placements
Sophisticated invalid traffic (SIVT) Advanced analytics, multi-point corroboration, human review Hijacked or infected devices, emulators, click farms, forced clicks and redirects, domain and app misrepresentation, hidden or stacked ads, ad injection, cookie stuffing

Google’s own invalid clicks definition covers the same ground in plainer terms: accidental clicks from poor placement, automated clicks from bots, and manual clicks meant to inflate costs or publisher profits.

In your own data, look for signals that don’t fit real people: click-through rates far out of line with the format, sessions with no scrolling or engagement, conversions that arrive in bursts or seconds after the click, many conversions sharing a device, IP range or data-center origin, and signups that never verify an email or phone. For programmatic inventory, check the publisher’s ads.txt file. Per IAB Tech Lab, ads.txt lets publishers publicly declare which companies are authorized to sell their inventory, which makes counterfeit inventory harder to sell.

Why does placement-level reporting matter?

Because a source is really a collection of sites, apps or placements, and they perform very differently. A network that only reports a blended total hides the few placements that drive the losses and the invalid traffic. Before testing, confirm the network passes a site, app or placement ID you can capture in your tracking.

Then manage it actively:

  • Read by placement on your conversion and quality data, not on clicks.
  • Blocklist placements that fail quality checks or burn spend without converting. Where a platform supports it, keep the list at the account level so it applies to every campaign. Google Ads, for example, supports account-level placement exclusion lists and scheduled lists updated from a file.
  • Allowlist proven placements when scaling, so new spend goes where results are known.
  • Cap frequency if the network allows it. High frequency on a small set of placements inflates impressions and accidental clicks without adding reach.

What should the brand safety and policy review cover?

What the network’s inventory looks like, and what your ads and landing pages say. Review sample placements before launch for content you don’t want next to your brand. Check the network’s own ad policies against your offer, plus any age-gating or regional rules your product needs.

Never accept a network’s suggestion to cloak or vary pages by visitor to pass review. Google’s circumventing systems policy defines cloaking as showing different content to different people or to Google, and says suspension occurs on detection without prior warning. Other networks may treat it the same way, and cloaked traffic poisons any read you take from it.

What does the step-by-step test plan look like?

  1. Write the hypothesis, floor, budget cap and decision date.
  2. Write the kill and scale rules (table below) and get them agreed.
  3. Confirm the network’s capabilities: placement IDs, click IDs, S2S postbacks, blocklists, frequency caps.
  4. Build and test tracking end to end with test clicks.
  5. Review creative, pages and sample inventory for policy and brand safety.
  6. Launch narrow: one geo, one offer, a small set of formats.
  7. Check daily for tracking breaks and invalid traffic patterns; prune placements weekly.
  8. Read by cohort at the decision date, on the outcomes that pay, against your existing sources.
  9. Decide: keep, change, scale or stop, and record why.

Which kill and scale rules should you write before launch?

Rules that name a metric, a threshold you set from your own economics and an action. Fill in your own values:

Condition at a checkpoint Action
Tracking discrepancy between your records and the network’s above your tolerance Pause and fix tracking before reading anything
Invalid traffic signals confirmed on placements Blocklist them, request credits, and stop if the pattern is source-wide
Spend cap reached, cohort value below the floor Stop
Below floor overall, a subset of placements above it Change: allowlist the subset and retest with a fresh cap
At or above floor with clean quality checks Scale in steps, re-reading at each step
Not enough conversions to read by the decision date Decide in advance: extend once with a new cap, or stop

A source that wins on front-end cost but loses on approved or paying conversions is a loss. If you need to know whether the source adds customers you wouldn’t have reached anyway, follow up with a holdout; see incrementality testing.

How MSG approaches this

We buy specialist networks and custom media as well as Google and Meta, and we test new sources the same way we run everything: a written hypothesis, a budget and a decision date, with tracking and conversion definitions settled before spend. For partners who pay on performance, budgets are approved per traffic source with a ROAS target and a floor, reporting is reconciled weekly to the partner’s tracking, and we optimize on cohort revenue rather than front-end CPA. We don’t run bots, incentivized clicks or cloaking. See media buying and planning and measurement.

Sources

  1. Media Rating Council: Invalid Traffic Detection and Filtration Standards Addendum (June 2020 update)
  2. Google Ads Help: Invalid clicks, definition
  3. IAB Tech Lab: ads.txt
  4. Google Ads Help: Exclude specific webpages and videos
  5. Google Ads Policy Help: Circumventing systems

Platform features and policies change. This guide reflects the sources above as of October 28, 2026.