Why conversion rate dropped in affiliate: a diagnostic tree

When conversion rate falls, the cause is in one of four places, and they should be checked in a fixed order. First the tracking, because a broken postback looks exactly like a traffic problem. Then the traffic itself. Then the landing page and the offer. Creative fatigue goes last, because it is the most popular explanation and the least often correct.

The order matters because each step is cheaper than the next, and each one can make the later steps pointless. There is no use rewriting creatives if the postback has been failing since Tuesday.

Define the drop first

Before you debug anything, write down three things. Which metric fell: CR from click to conversion, or something upstream such as CTR? When did it start, to the day and if possible to the hour? And which slice fell: everything, one source, one country, one device?

A hypothetical example that will run through the article. A campaign bought 1,000 clicks a day at $0.10. For the last week it converted at 2.0%, so 20 conversions a day. Today it shows 8 conversions, which is 0.8%. That is a drop of 60%, large enough that it is almost certainly one event, not slow wear.

Also check the sample. At 1,000 clicks a day and a 2% rate, normal day-to-day noise is a few conversions either way. A day with 17 is not a drop. Give small campaigns several days before drawing conclusions.

Step 1. Rule out tracking

Tracking failures are the cheapest to check and the most common reason for a sudden collapse. The visitor converted, but you were never told.

• Postback. Open the postback log, if your tracker has one, and look at the last successful postback. If the last one is from yesterday evening, the postback stopped, not the audience. The common causes and fixes are in why a postback doesn't arrive.
• Click ID. If the offer link was edited recently, check that the click ID macro is still in it. Without it the network can't send the conversion back to the right click.
• Network versus tracker. Compare conversions for the same dates in the network and in your tracker. If the network shows 20 and the tracker shows 8, the traffic is fine and the data link is broken. If both show 8, continue. For the usual size of gaps see tracker and network discrepancy.
• Time zone and window. A report filtered to a different time zone than the network's can move conversions into another day and look like a drop. Check that both use the same day boundaries. Also check you are not comparing a complete day with a half-finished one.

In our example the network shows 20 conversions and the tracker shows 8. That would settle it: the money is arriving, the report is incomplete, and nothing else needs touching. Say instead that both show 8, and we move on.

Step 2. Check the traffic itself

If tracking is fine, compare today's clicks with last week's clicks along the lines that usually shift.

Bots deserve their own look, because they lower CR without changing the click count. If 25% of yesterday's 1,000 clicks were automated, the real audience was 750 people. With 20 conversions the human CR was about 2.7%, not 2.0%. If that bot share rises while the human audience stays the same, the blended CR falls, and you pay for clicks that cannot buy. The patterns are described in signs of bot traffic.

In TDS.SO the click log shows country, city, device, browser, referrer and status for each click, with filters for bots, unique, VPN and Tor, and the overview splits clicks into humans and bots per day. Filtered clicks are logged with a reason such as vpn, datacenter or suspicious_headers, so you can see what changed. Setting up a filter level on the link is covered on the traffic filtering page.

Say in our example the click log shows 1,000 clicks, but data center clicks went from 40 to 350 a day starting yesterday. The 650 remaining human clicks would have converted around 13 at the old rate. We have 8, so there is more to find, but we found the first piece.

Step 3. Landing page and offer

Traffic can be unchanged while the destination quietly stops working. These checks take minutes.

• Open the whole path yourself on a phone, on mobile data, from the target country if you can. Click through to the offer and, if it's allowed, to the form. A broken button or a redirect loop is visible in one minute.
• Offer status. Check whether the offer is paused, capped, or limited for your geo. When a daily cap is reached, many networks keep accepting clicks and stop counting conversions or show a stub page.
• Payout and rules. A payout cut, a changed conversion event (for example from lead to approved lead) or a new traffic restriction will change what gets counted without any change in visitor behaviour.
• Page speed. A heavier page, a slow third-party script or a hosting problem can lose visitors before the page renders. Measure on a mid-range phone, not on your desktop.
• Destination routing. If you send traffic to several destinations by rules, check the rules didn't change. A rule that now catches too much sends the wrong audience to the wrong page. Rules are checked top to bottom and the first match wins, so a new rule placed above the others can swallow traffic meant for them.

Here is a way to tell traffic from destination. Look at CTR from the landing page to the offer, if you can measure it. If people arrive and read but don't click through, the page or the offer changed. If they don't arrive, or arrive and leave in a second, the problem is upstream of the page.

Step 4. Creative and audience fatigue

Fatigue is a real cause, and it has a recognisable shape. It is gradual. CTR falls first, cost per click rises as the platform has to work harder to find fresh viewers, and CR slides over days, not hours. The same creative to the same audience for weeks is the usual setup.

Check how long the creative has been running, how often the same people see it where the platform shows frequency, and whether a fresh creative on the same audience recovers the numbers. A short split test with a new angle against the old one answers this better than opinions.

A related cause is audience drift. Some sources widen the audience when they can't spend the budget, and the extra reach is colder than the audience you started with. The creative looks the same, the people don't.

Worked example, end to end

The hypothetical campaign again: 1,000 clicks a day, 2.0% CR, now 0.8%. The tree gives this sequence.

1. Tracking. Postback log has entries through today. The network shows 8 conversions as well. Tracking is fine.
2. Traffic. The click log shows clicks flat at 1,000. Data center clicks rose from 40 to 350. One placement went from 10% to 45% of clicks. Human clicks are about 650.
3. Placement. Most of the data center clicks come from that placement, and its CR is almost zero. With it paused, the remaining human traffic is at its usual rate, about 13 conversions per 650 clicks.
4. Destination. The offer is live, the cap has room, page loads in normal time. No change needed.
5. Creative. CTR was steady for the week. Fatigue is not the cause here.

Without the order you might have rewritten the ad. With it, the fix was to pause one placement and turn on a stricter filter level for data centers.

FAQ

How big a CR drop is normal?

It depends on volume. With a few hundred clicks a day, swings of a third from one day to the next can be pure noise. Judge a drop over several days and a few thousand clicks, and compare the same weekday to the same weekday.

Can bots really lower my conversion rate?

Yes. They add clicks that essentially never convert, so they lower CR and raise your cost per conversion without any change in the human audience.

My CR dropped right after I changed the creative. Is that the creative?

Possibly, but a new creative often moves the audience as well. Look at the geo, device and placement mix before and after the change, and keep the old creative running as a control if the budget allows.

The tracker shows fewer conversions than the network. Which one is true?

The network pays, so treat its number as the money figure and treat the gap as a tracking problem to fix.

When should I stop a campaign?

When you've run the four checks, found no fixable cause, and the cost per conversion stays above your payout over a sample large enough to trust. Decide the sample size before you start the test.

If you want to see clicks by country, device, referrer and bot status without exporting anything, the traffic filtering page explains how TDS.SO filters and logs them, and the analytics page covers the click log.