The instinct is understandable. You want meetings, the platform offers an objective called conversions, so you select it. Then delivery becomes erratic, costs climb, and the campaign never seems to settle. The usual response is to blame the creative, and the usual result is three weeks of iteration on a problem that is structural.
The volume requirement
Conversion optimisation works by learning which people in your audience resemble the ones who converted. That requires a supply of conversion events to learn from — enough of them, recently enough, to distinguish signal from chance.
The commonly cited floor across major ad platforms is somewhere around fifty conversions per campaign per week for stable optimisation. Now do the arithmetic on a typical named-account programme: 400 accounts, six committee members each, a matched audience of a few thousand people, a realistic monthly conversion rate on a considered B2B offer. You are generating a handful of events a week, not fifty.
The algorithm is not broken in that situation. It simply has nothing to learn from, so it falls back on weak proxies, and delivery becomes unstable in ways that look like bad luck.
What starvation looks like
Four symptoms, and they are consistent enough to be diagnostic:
- Delivery lurches. Two thousand impressions on Tuesday, four hundred on Wednesday, no change in budget or bid.
- Cost per result swings wildly week to week, with no corresponding change in the auction or your creative.
- Coverage collapses toward a small subset. The algorithm finds a handful of people who resemble a past converter and serves them repeatedly, which is precisely the opposite of what an account-based programme needs.
- The campaign never exits learning, or exits and re-enters after every edit.
The last one is worth watching for specifically, because the natural response to poor performance — editing the campaign — resets learning, which makes performance worse, which prompts another edit.
Three better objectives for named-account work
Reach, with a frequency cap
For the coverage stage of a programme this is almost always correct. You are explicitly telling the platform to spread delivery rather than concentrate it, which is exactly the job. Pair it with a cap, or the "cheapest available impressions" behaviour reasserts itself.
Measure it on accounts reached and cost per account reached. Do not measure it on CTR.
Engagement or video views
For the depth stage, where you want people to actually consume something. These objectives optimise toward events that occur in far greater volume than conversions, so the algorithm has enough to learn from while still selecting for interest rather than mere presence.
Traffic, with the landing page doing the qualifying
Underrated for bottom-funnel ABM. Clicks are plentiful enough to optimise on, and the qualification happens on your page rather than in the ad platform's model. It gives up some efficiency in exchange for stability, which on a small audience is a good trade.
When conversion optimisation is the right call
Three situations where it works:
Your audience is large enough. If your list is several thousand accounts and the matched audience runs into six figures of people, conversion volume can clear the threshold. This is programmatic ABM territory, and different rules apply.
You define a shallower conversion. Optimising for a document download or a page view rather than a demo request can generate enough volume. You are then optimising for something correlated with what you want rather than the thing itself, which is a real trade-off — but a working shallow proxy beats a starved deep one.
You are running a bottom-funnel campaign against an already-warm audience. If the audience is accounts that have crossed a scoring threshold, the base rate is high enough that conversion volume can be viable even on a small list. This is the one case where the classic setup fits an ABM programme cleanly.
Outside those three, choosing conversions on a named-account campaign is choosing an objective the campaign cannot feed, and the resulting instability will be misdiagnosed as a creative problem for as long as you let it.