The Five Questions to Ask Any Traffic Provider

We measure our own delivery, session by session, so we know which questions actually separate a good provider from a bad one. These five are the ones that matter - in this order, and with the numbers you should expect as an answer.

The evidence behind them is in our study of 57 million delivered traffic sessions, which includes the failures we published alongside the good numbers.

1. "How many sessions landed, and how many were only requested?"#

Look for: two numbers, not one. Sessions delivered and sessions attempted.

A provider who counts requests as visits will report a higher number than one that confirms delivery - and will not tell you that this is what they are doing. In our own 30-day window, 56,964,896 sessions were attempted and 56,888,455 were delivered: a 99.866% delivery rate, and 76,441 attempts that we refused to count because we could not confirm them.

The follow-up question that matters: what were the failures? If the top failure reason is "the measurement system said no" - a GA4 timeout, a proxy limit, a closed quota window - that is a provider being careful with your money. If the top failure reason is "visitor bounced", they are counting exits as visits and quoting you a bigger number than they delivered.

Why it matters: you are billed on one side of that gap and you get the other side in your analytics.

2. "Show me the duration histogram, not the average."#

Look for: a distribution, in buckets. Any average can be manufactured by a handful of very long sessions - and an average tells you nothing about the worst half of your traffic.

Ask for something like this shape, and compare it with ours from 5.93 million delivered sessions:

The tell: a healthy traffic mix is heavily concentrated in a single band (for us: 5 minutes and longer, 83.73% of sessions) with a small, sharp spike at the very front - those are the visitors who leave immediately.

The red flag: a provider whose sessions pile up in the first 30 seconds has sold you a pile of instant exits. Their average session duration may still look respectable. The average is what they will quote you; the histogram is what actually happened.

3. "What is your bounce rate, and what does a bounce look like in your data?"#

Look for: a rate and the definition behind it.

Our bounce rate is 8.92%, and we can tell you exactly what it means: every session flagged as a bounce ended within the first 30 seconds (we checked - 529,515 of 529,515), and only 320 of the 529,835 sub-30-second sessions were not flagged as a bounce. Bounce and "left within 30 seconds" are the same population in our data.

Why it matters: a bounce rate without a definition is decoration. A provider counting "any session under 60 seconds" as a bounce and another counting "no interaction event" as a bounce will report numbers that are not comparable, and neither number is wrong - which is exactly the problem.

4. "What is the device split, and the country split?"#

Look for: both, and the quality metrics behind them - not just the volume.

Our 3-day delivery was 81.2% mobile, 17.1% desktop, 1.7% tablet. Here is the part that surprised us, and the part worth asking about:

Desktop traffic bounced almost four times as often as mobile traffic in our own delivery (22.39% against 5.85%). If you are buying traffic for a desktop audience and your provider is delivering mostly mobile, your campaign will not look like what you bought - and you will blame the wrong thing.

For countries, the same logic applies: our 30-day window covered 205 countries, and the distribution is not the usual "US, UK, Canada, Germany" shape. South Africa, Poland and Turkey each deliver more sessions than Australia or Germany in a 7-day window. If you sell outside the obvious five, that is where the volume is.

5. "What failed, and why?"#

Look for: a breakdown you could verify.

Ours, over 30 days, 76,441 failures in total: GA4 timeout 55,259 (72.3%), closed quota-day dispatches 17,256 (22.6%), proxy HTTP 466 limit reached 2,235 (2.9%), other GA4 connection errors 1,432 (1.9%), everything else 259 (0.3%).

The test: if the failures are infrastructure - measurement timeouts, provider limits, closed quota windows - you are dealing with a provider that verifies before it counts. If the failures are the visitors themselves, the provider is counting exits as visits and you are paying for a number that never reached your analytics.

The one-line version#

Asked, answered and published: 5.93M sessions, median 7:00, 8.92% bounce, 83.73% over five minutes, 81% mobile, 205 countries, 99.866% delivered - and the failures.

That is the level of detail you should be able to get from any provider you pay. A provider that answers with one average and refuses the histogram is telling you the one number you cannot check.

Questions we did not cover, or the ones we answered badly - write to us.