Traffic loss between a tracker and an affiliate network, between a source and a tracker, or farther down a redirect chain is unpleasant. Even a 1% loss can be expensive.
Sometimes finding the cause is hard because the ping-pong starts. Tracker vendors say everything is okay—and often that is true. With self-hosted trackers, you have to investigate it yourself and wrestle with the hosting and its configuration. Different domains, geos, and offer types add complexity. The affiliate network says (in Russian) everything is fine. They do not want to dig into it; you have to go to the platform vendor or the internal tech people. Maybe the affiliate screwed up, maybe it is the platform, maybe the internets broke.
A traffic-loss case
Some time ago I helped investigate a case: traffic loss between a tracker and an affiliate network. I went fairly deep into it. At first, as usual, everything looked chaotic. Different campaigns, offers, and geos showed losses piling up quickly. The farther we went, the bigger they got: 2%, 3%, 7%, 10%, 17%.
We managed to untangle it partly because the stars aligned:
- The team used Everflow to consolidate offers. So the
affiliate networkwas not literally an affiliate network, though it used the corresponding platform. Everflow is a decent platform, but dealing with them is tedious. I will not say the support is terrible, but everyone is in different time zones and they do not immerse themselves in complicated cases. - The tracker (in Russian) was Zeustrack—not the best-known tracker, but a good one, especially if you work with Facebook or Google. We had good contact with them, so we could look closely at what was happening: how a visit traveled, what technical guarantees existed, what could break, and what could not by design.
The stars do not always align.
How to fix traffic loss
Localization:
- traffic parameters: geo, OS, ISP, browsers (in Russian), and so on
- source
- buyer
- landing page (in Russian)
- campaign
- campaign domain
- campaign server
- offer domain (in Russian)
- campaign, redirect, and offer IPs
- sometimes indirect factors matter: offer type, advert, redirect chain
Usually nothing except a visit ID gets passed to the affiliate network. You either have to build aggregate statistics or bolt on extra parameters.
Causes of traffic loss
You need localization so you can formulate hypotheses, find the cause, and fix it.
A common cause is the server region not matching the traffic region. Everyone knows about it and everyone still steps on it sometimes.
What else:
- A long or slow redirect chain; a network resells a reseller; an advert uses a couple of tracking solutions. Slow is bad.
- An expired certificate somewhere in the chain.
- Domains on blacklists: Google Safe Browsing, antivirus products, DNS filters, blocks, ISP equipment filters.
- Ad blockers.
- In-app browsers: Facebook loves doing weird shit and interfering with the redirect chain.
- Servers under load.
- Cloudflare breaks the chain: WAF, BFM, and so on.
Sometimes it is a reporting discrepancy instead:
- time zones
- how visits are counted—not all uniques are equally unique
- removing the tracking part of a URL
- silent filters on the platform side
So what was wrong in the original case?
I will not describe the whole detective story. The short version: we ruled out simple things like time zones and regions. We ruled out the tracker with confidence. In the end we found that the issue was in some Everflow IP addresses. Their integrations use EF servers to bind domains. IPs can be shared, collect crap, and end up on different blacklists. Yes, the cause was from the known list, but difficult to diagnose.
There are many blacklists now, and not all of them are obvious. The nasty ones are blacklists in provider equipment—vendor or third-party. It is a zoo because different geos, states, and cities all have their own. From the outside it looks as though loss on a domain or IP is gradually increasing, and it is not always tied to one ISP. Monitoring is hard: a domain or IP can collect flags and keep working; it can look green in public checkers and still lose 20–30%.
We solved it like this:
- move to self-managed domains with our own IPs in the required regions
- routine reconciliation of loss across different slices
Watch discrepancies, learn how to find losses, and squeeze the whole chain, including adverts.
By the way, do not confuse this with shave (in Russian).