What is CPE?

A campaign spends A$2,400 and the platform reports 30,000 engagements, for a cost per engagement of A$0.08. Breaking the denominator apart: 24,000 are reactions, 4,500 are image expansions, 1,200 are comments and 300 are shares. Priced on shares alone, the same campaign cost A$8.00 an engagement. The blended figure is dominated by the cheapest and least committal event in the set.

Cost per engagement divides what a campaign spent by the number of interactions the platform counted. The arithmetic is one division, and the interesting question is entirely about the set of events being counted.

That set is wide. Depending on the platform and the placement it can include reactions, likes, comments, replies, shares, reposts, saves, follows, taps that expand an image, taps on the page or profile name, poll votes, sticker taps, and some video interactions. Several of these require a deliberate decision. Several are indistinguishable from a mis-tap while scrolling.

Because the set is wide and inconsistently defined, a blended cost per engagement is dominated by whichever event in it is the most common — which is reliably also the cheapest and the least committal. An ad earning tens of thousands of reactions and a few hundred shares will report a cost per engagement that is effectively a cost per reaction. The shares, which are the event a marketer would actually pay for, are invisible in the average.

The first useful action on this metric is therefore not to improve it. It is to take it apart.

Ranking the set by the commitment it takes is a more useful exercise than tracking the average.

A reaction is a thumb moving a few millimetres. It costs nothing, it is often reflexive, and on some placements it happens by accident. An expansion of an image means somebody wanted to see more, which is weak but real. A comment means somebody had a reaction strong enough to type, which cuts both ways — arguments engage extremely well. A save means somebody intends to come back, which is among the strongest signals available in a feed. A share means somebody was willing to attach their own name to the message, which is the strongest of all and the rarest by a wide margin.

Priced individually, these events differ by orders of magnitude in the same campaign. Priced together, they tell you very little. Any account using this metric seriously reports the mix, not the blend.

Three mechanisms make cost per engagement look better than the campaign is.

The first is denominator width. Anything that adds events to the counted set lowers the price without changing behaviour. A placement that counts profile taps will report cheaper engagement than one that does not.

The second is inventory. Placements with high accidental-interaction rates — dense feeds, small tap targets, autoplaying carousels — produce cheap engagement volume that no one intended.

The third is creative that provokes rather than persuades. Outrage, disagreement and bait all engage efficiently. They also attach a comment thread to the ad, which is a durable asset when the ad runs for months and a durable liability when the thread is hostile.

An account optimising toward the lowest cost per engagement will find all three of these without being asked to.

There is a legitimate use for buying engagement, and it is narrower than the metric's prominence suggests: building social proof on an ad that will carry traffic for a long time.

An ad with substantial comment and share counts converts better than an identical ad with none, because the counts are visible and act as evidence. Where an account runs a small number of evergreen ads over months, deliberately seeding engagement early and then switching the campaign to an outcome objective is a sound and well-worn tactic. The same logic applies to a post being promoted ahead of a launch.

Outside those cases, engagement is a diagnostic rather than a purchase. It says whether the creative earned a response from the audience it reached, which is genuinely useful information about the creative and almost no information about whether the campaign is working. A campaign meant to produce leads, sales or installs should be bought on leads, sales or installs, and its engagement figures should be read the way a mechanic reads a gauge.

The most informative pairing is cost per engagement beside the outbound-click figure, because the two together separate resonance from traffic.

High engagement and low outbound clicks is a common and specific outcome: the ad is entertaining and the offer is not compelling, or the ad has satisfied the curiosity it created so there is nothing left to click for. This is fixable, and the fix is in the copy rather than the media.

Low engagement and healthy clicks is the opposite pattern and usually fine — direct-response creative that states an offer plainly and sends people away is doing its job whether or not anybody reacts to it.

Both high is a strong ad. Both low is a targeting question before it is a creative one.

Whether anybody remembered the brand. Interaction is not attention and neither is memory.

Whether the response was positive. The metric counts an angry comment and an enthusiastic one identically, and the angry one usually generates more of its own kind.

Whether it reached the right people. Engagement is cheapest among audiences that engage with everything, and those audiences are not usually the ones with a budget.

Whether it did anything at all. Engagement rarely appears in an incrementality test as the mechanism behind a lift. It correlates with creative that works; it is not evidence that the creative worked.

Make something worth responding to, which is the whole of the honest answer. Ask a question people have an opinion about, show a result rather than describe one, or say something specific enough to disagree with.

Then match the format to the interaction being sought. Carousels earn expansions, polls earn votes, and a static image with a plain offer earns neither and should not be judged on them.

Then split the report by event type and by placement, and set the target on the event that matters rather than on the blend.

Then decide whether the campaign should be bought this way at all. For most accounts the answer is no, and the figure goes back to being a gauge.

• Break the denominator apart before you use the figure. A blend of reactions and shares is dominated by reactions, and the two events say almost opposite things about the creative.
• Price the events you actually want. Saves and shares carry intent that a reaction does not, and a campaign optimised toward the blended count will buy the cheap end of the set every time.
• Read it beside the outbound-click figure. Ads that earn engagement and no traffic are a real and common outcome, and the pair of numbers is what exposes it.
• Ask what the engagement is for before buying it. Social proof on an ad that will run for months has a use; a comment count on a two-week promotion generally does not.
• Moderate the comments you are paying for. An engagement-optimised campaign will happily buy an argument, and the argument is attached to the ad for as long as it runs.