Last verified: August 22, 2026
What You Can Now Do
Context Memo now supports oCPC as a selectable campaign type in the ad campaigns interface, alongside manual CPC and CPM. Rather than setting a fixed bid per click, you set a target cost per conversion (tCPA) and name the conversion event you care about; the system predicts the conversion probability of each impression opportunity and bids against that prediction. What changes for you is the unit of work. You are no longer adjusting bids by placement or audience. You are defining one CPA target and one clean conversion event, then judging the campaign on blended CPA instead of cost per click.
Where It Is in Context Memo
It lives in the ad campaigns interface, in the campaign type selector at setup. Choosing oCPC replaces the bid field with a tCPA field and a conversion event picker. The same screen now shows a recommendations panel: suggested tCPA ranges derived from your account's historical conversion data, the minimum conversion volume needed to exit the learning phase, and a warning if the conversion event you selected fires too infrequently for the model to train against.
How to Use It
- Confirm the conversion event fires cleanly. Open the conversion event picker and check the recent firing history for the event you plan to optimize toward. If the event is double-firing, firing on page load, or attached to a step most visitors never reach, fix that before anything else. The model will optimize toward whatever you tell it, including a broken signal.
- Select oCPC in the campaign type selector. The manual bid field disappears and the tCPA and conversion event fields take its place. Your targeting, creative, and budget settings carry over unchanged.
- Set tCPA inside the suggested range. The panel calculates the range from your account's own conversion history. Setting a target well below the floor of that range will starve delivery; the campaign will simply not bid on much. If you want a lower CPA than the range allows, start at the range and step down after the campaign stabilizes.
- Read the volume threshold before you launch. The panel states the minimum conversion volume required to exit the learning phase. If your selected event does not clear it, stay on manual CPC and either pick an upstream event that fires more often or accumulate more data first.
- Leave the campaign alone through the learning window. Expect roughly 7 to 14 days during which CPA can swing wide. Editing targets, creative, or budget mid-window restarts learning. Check delivery and spend pacing, not daily CPA.
- Evaluate on blended CPA after the window closes. Clicks will usually be fewer and more expensive than on manual CPC. That is the intended trade. The question is whether cost per conversion came down, not whether cost per click went up.
Why We Built It
Manual CPC treats every click the same. A repeat visitor clicking through on a high-intent product page and a first-touch click from a cold audience get bid at the same rate, even when their conversion probability differs by an order of magnitude. Advertisers told us the same thing in different words: they were paying for traffic and reporting on outcomes, with no mechanism connecting the two. oCPC moves spend toward the impressions the model believes will convert, so the bid reflects predicted value rather than a flat rate you set once and forgot.
What It Does Not Do Yet
You lose granular bid control. There is no override to raise or lower the bid on a specific placement or audience segment while oCPC is running; the target and the event are the only levers.
The learning phase is real and unavoidable. CPA will be unstable for the first week or two, and campaigns with thin conversion volume may never stabilize. If the event fires rarely, the algorithm trains on noise and produces results that look random because they effectively are.
Misconfigured conversion events remain your responsibility to catch. The warnings in the recommendations panel flag low firing frequency, but they cannot tell you that an event is firing on the wrong page or counting the same visitor twice. Verify the event in your own analytics before you hand bidding over to it.
Accounts without conversion history will see a suggested tCPA range that is either absent or very wide. In that case the recommendations panel is telling you something useful: run manual CPC until you have enough conversion data for the model to learn from, then switch.