Search Partners: separate campaign to test, or keep them in the main Search campaign?

Search Partners tends to get turned on by default and never looked at again. The argument for leaving it on is volume: partner sites add impressions, and on a small account every extra click counts. The argument against is visibility. The search terms view gives you aggregated data, but you can't really see where the clicks landed, and quality varies a lot from one partner site to the next.

For a lead gen account where a bad lead eats real sales time, I'd split partners into their own campaign first. Run it a few weeks and compare cost per qualified lead against your main Search campaign, not cost per form fill. If it can't match the main campaign on qualified leads, you're paying for traffic you can't inspect. Has anyone seen partner traffic hold up on high-ticket services, or is it always the weaker cousin?

Start with the Network column in the main campaign (Segment > Click > Network). It puts Google Search and Search partners side by side, so you can compare cost per qualified lead without moving budget or restarting learning.

I'd keep partners in the main campaign while you look at that split. Only break them out if partner leads are clearly worse after a few weeks. A separate campaign competes with your main one for the same budget, and that makes the comparison messier. Qualified-lead lag can also be long on high-ticket services, so give it more time than you'd think.

The Network split helps, but if you're running a target CPA, the partner numbers already reflect how Smart Bidding adjusted bids for them, so a gap can look worse than the traffic really is. Which bidding strategy are you on?

I don't buy the idea that the bidder makes partner numbers look worse than the traffic really is. Smart Bidding isn't penalizing partners on purpose. It estimates conversion likelihood for each auction and bids to hit your target, so if partner CPA runs higher, that's mostly just what that traffic costs at that target. If anything, a target CPA can make partners look better than they are, because the average stays near target while the weak sites hide inside it.

That's also why I wouldn't lean on the Network column alone. It shows that partners behave differently, not which partner sites are worth paying for, and you can't see those sites anyway. The only clean read is qualified leads over the same window, with enough time for the lag Kai mentioned. If the split still looks bad after that, the gap is probably real.

Fair point on the averaging, but partner volume is usually thin, and Smart Bidding learns from conversions. With only a handful of partner leads its estimates are noisy, so a CPA gap on six conversions tells you much less than it looks like. Before reading the split, count how many qualified leads each segment produced in the window. If partners are in the single digits or low teens, treat the comparison as directional and extend the window. How many qualified leads has partner traffic sent you so far?

Six conversions tells you very little, so set a minimum before anyone reads the split. Roughly 20 to 30 qualified leads per segment is a reasonable floor, and only count leads whose form date is at least as old as your typical form-to-qualified lag. Below that, the gap is mostly noise. Above it, a difference of a third or more in cost per qualified lead is usually worth acting on, and smaller gaps can be left alone.

The lag rule matters because recent partner leads haven't had time to qualify yet, so a short window can make partners look worse than they are. That's the opposite of the bidding concern Theo raised, and with high-ticket services it's the more common problem. These numbers are rough heuristics, not a formula, so adjust them for how much a qualified lead is worth to you.

A third as the action line at 20 to 30 leads is too loose. Conversion counts are noisy, so with 20 qualified leads per segment the error on each cost per lead is roughly 1 divided by the square root of 20, about 22%. Compare two segments and both sides carry that noise, which means a one-third gap at that size is only about one and a half times the uncertainty. You could easily be reacting to luck. Hitting the count floor doesn't make the gap solid.

Before moving budget, I'd want the partner gap to show up in two separate windows, with the lag rule applied to both. That's far more convincing than one window that just got big enough. The catch is that partner volume is often thin, so two clean windows can take a long time to collect. For slow accounts, the honest answer may be to keep partners on, run a negative-keyword check on the search terms, and accept you won't have a clean answer for a while.

The uncertainty on the gap has to account for both segments, which puts it closer to 31% than 22%. A one-third gap at 20 leads each works out to roughly one standard error, not 1.5, so at that sample size it's still mostly noise.

Fair point, I only counted the error on one side. A rule of thumb that holds up: the gap should be bigger than about twice sqrt(1/leads_A + 1/leads_B). At 20 leads a side that's roughly a 60% gap, and at 50 a side it's around 40%. Anything smaller is probably still noise.

The math holds, and a 60% threshold at 20 leads a side is stricter than most of the numbers earlier in the thread. What it leaves out is the funnel. With high-ticket services, a qualified lead starts as a form fill that then has to qualify, and that second step is its own proportion with its own noise. If partners send 60 forms and 15 qualify, the qualification rate alone carries roughly 22% relative error before you even get to spend. So count forms and qualified leads for both segments, run a quick two-proportion check on the qualified rate, and only move budget if the gap clears both that and your threshold. Otherwise a gap can look solid on cost per lead while the qualification step is just noise.