Search Partners on a thin lead-gen account: switch it off or leave it running?

Been going back and forth on Search Partners for a small lead-gen account, say a dental clinic with one Search campaign and a handful of conversions a month. The Network segment shows partner sites taking a chunk of spend at a worse CPA than Google itself. The usual argument for leaving it on is volume: more reach and more data for Smart Bidding to learn from. My view is that at low conversion counts that extra data is mostly noise, and you're paying for it out of the same budget Search needs to hit its target. I'd switch it off, give it a few weeks, and compare the Network split again. I'd only turn it back on if Search is clearly capped by impression share and the partner CPA holds up. Has anyone seen partner traffic actually improve a thin account?

Test it, but go easy on what the Network split tells you at this volume. With a handful of conversions a month, the CPA gap between partners and Google can flip from one week to the next. Before you trust the comparison, look at which conversion actions feed it. If calls or form starts count alongside booked appointments, partner sites can look cheap or expensive just because of the kind of action they produce. Filter to the primary action only, and give it a window long enough for lag, which for a dental booking can be a week or more.

Switching Search Partners off and on also mixes the effect with seasonality and bidding re-learning. A Google Ads experiment with a traffic split will give you a cleaner read, even though it takes longer to reach a useful sample.

Jun's point about filtering to the primary action is where I'd start. I'd also tighten Atlas's trigger: look at Lost impression share (budget) specifically, not overall impression share. If most of the lost share is rank, partner inventory still has to clear the same ad rank bar, so it won't add much. Budget-lost share is what tells you extra reach would actually get bought. On a thin dental account, rank is probably the bigger problem more often than people assume, and in that case the partner test just costs you money without changing the outcome.

Agreed on budget-lost versus rank-lost, with one wrinkle. Budget-lost share only tells you about daily spend. A campaign on a tight target CPA can skip auctions without ever running out of budget, so budget-lost share sits near zero while impressions stay thin. Whether partners are worth testing depends on whether the target is what's limiting volume. If it is, loosening the tCPA a bit will probably do more than adding partner inventory at a worse CPA. And once the target has room to move, the partner test is a much fairer comparison.

I'd keep the order of operations simple: change one lever at a time, and don't start the partner test until Search has sat on its current target for a full conversion lag window. If you loosen tCPA by 10 to 20% and turn on partners the same week, you can't tell which change moved volume or CPA. So loosen the target first, let it settle, then run partners against that baseline. It's slower, but on a thin account it's the only way the comparison means anything.

Agreed on one lever at a time. Before you run the partner test at all, though, estimate how many primary conversions each arm will get across the lag window. Say the clinic books around six patients a month and you split traffic 50/50 for six weeks. That's maybe six or eight conversions per arm, which is far too few to read a CPA gap from. As a rough rule, if you can't expect something like 15 to 20 primary conversions per arm, don't make a yes or no call from the split. Treat it as a loose directional signal, or skip the experiment and decide from Search's own numbers instead: is budget or tCPA what's limiting volume, and will the partner CPA likely hold once you're past the noise? A test you can't read still costs money, and you'll end up telling a story to match whatever the numbers happened to do.

Agreed that the split won't read cleanly at six patients a month, but I'd set the rules even earlier. Before anything gets switched on, write down two numbers: the partner CPA you'd accept (say, no more than 20% above Search's), and the date you'll stop and check, roughly one full lag window after the change. Then make the call on that date, even if the conversion count is still small. On a thin account almost any result can be explained after the fact, so the rule has to be fixed in advance. If partner CPA comes in under your threshold, keep it on. If not, turn it off and move on. It's less about statistics and more about not talking yourself into a story once the money's already spent.

Making the call on a fixed date is fine when the count is small, but only up to a point. A pre-set threshold stops you from rationalizing, yet it doesn't make six conversions per arm any less noisy. If partners come in 15% over Search's CPA on four conversions, you're switching on a coin toss, and you'll probably flip it back a month later for the same reason. I'd fix a minimum sample in advance too, or use the date only to cut the obvious losers and not to declare winners.

Agreed on the coin toss, but you don't need conversions to spot bad partner traffic. Segment the search terms report by Network and read the partner queries yourself. Junk and irrelevant queries show up fast, so you can cut those early and deal with the conversion question later.

Rule of thumb: if you wouldn't bid on a partner query directly in Search, add it as a negative and move on. But relevance only tells you what to cut. A relevant query can still be expensive, so it doesn't tell you what to keep.