Microsoft Ads audiences on Search: bid-only, targeting or observation for a small list?

Microsoft Ads gives you a few ways to attach remarketing audiences to Search campaigns: bid-only (you adjust bids for people on the list, everyone else still sees the ads), targeting (only people on the list see the ads), or observation (you just collect data). The debate usually comes down to which one suits a small list.

Bid-only feels safe because you don't lose reach, but with a list of a few hundred people the bid adjustments barely move anything, so you mostly end up with reporting noise. Targeting on a small list cuts volume hard. That's fine if those people are your high-intent buyers, and a problem if the list is just anyone who hit the homepage last month.

My view is to run observation for a few weeks and compare how those visitors convert against everyone else. Only switch to bid-only or targeting once the list is big enough that the numbers mean something. Has anyone settled on a minimum list size they actually trust?

I'd split this by what the list is, not how big it is. For people who viewed pricing or started a checkout, go bid-only straight away even if the list is small. It doesn't cut your volume, so the worst case is some noisy bid adjustments. Observation fits better for broad lists like all site visitors, where you don't know yet whether they're buyers. Targeting is the one to hold back until the list is big enough to keep impressions steady, because it's the only option that can quietly starve a campaign.

Mira's split by list intent makes sense to me, but I'd check the measurement before trusting bid-only on a small list. With a few hundred people, the conversion counts behind each bid adjustment are tiny, so the report can look like it's proving something when it's really just noise. Are you judging those adjustments on conversions per audience, or on the audience's share of total conversions?

Mira, I don't think "noisy bid adjustments" is the worst case. With a pricing-page list, the worst case is that you raise bids on people who were going to convert anyway, often through branded searches they'd type regardless. The report then shows a strong return on those bids, but you're paying a premium for conversions you'd have gotten at the normal price. Jun's point about tiny counts holds, but the bigger problem is that the list is selected on the outcome, so its conversion rate tells you very little about what a bid increase actually adds. I'd split out brand-term conversions before judging any adjustment.

Splitting out brand terms isn't the main fix. It removes the most obvious freeloaders, but a pricing-page visitor who came in through a generic search and was already ready to buy has the same problem, and the report can't tell them apart. Even with brand excluded, the list's conversion rate is still inflated by who ended up on it in the first place.

The only clean read I know of is a holdout. Keep a random slice of the list at normal bids for a few weeks and compare it with the bid-adjusted group.