AI for Meta Ads: spotting what wears out before it costs you

On Google Ads, waste shows row by row: one query, one cost, zero conversions. On Meta it is diffuse. Nobody types a search: the algorithm chooses who sees the ad, and it keeps showing it to the same people as long as they respond. A creative performs, then its audience has seen it too often, and the same campaign gets gradually more expensive without a single setting having moved.

This is what is called ad fatigue. It never triggers an alert: cost per result climbs a few pence a day, click-through rate slips a tenth of a point a week. Taken one by one, every figure looks normal. The slope is what is not — precisely what someone opening a dashboard once a week never sees coming.

AI has no extra intuition about your market. Its advantage is more modest and more useful: it rereads every ad every day, without forgetting one, and compares each with the others rather than with a memory.

No single signal is enough: one bad day is not a trend. Combinations are what matter, and they show at ad level, not across the whole campaign.

• Rising cost per result — at a constant budget, each purchase or lead costs a little more, week after week.
• Falling click-through rate — the ad is still delivered, but it stops the scrolling thumb less and less often.
• Spend concentrating — a single ad absorbs most of its ad set’s budget while its results no longer follow.
• Results plateauing — spend goes up, the number of results stays flat: you are now paying for repeated impressions.

The connection uses a Meta authorisation you grant and can withdraw. One authorisation covers every ad account it exposes, and each sync rereads the last thirty days.

• Campaigns — status, budget, spend, results and cost per result.
• Ads — spend, impressions, clicks, click-through rate, results and the image of each ad — the level where fatigue shows.
• The daily series — the day-by-day movement of each campaign, which makes a slow slope visible.
• Spend per platform — each campaign’s spend on Facebook, Instagram, Messenger, WhatsApp and Audience Network.
• Account issues — the problems Meta reports that prevent normal delivery.

The Meta API does not return “conversions”. It returns a list of actions in which purchases sit next to link clicks, three-second video views and Page likes. Adding them all up gives an absurdly low cost per acquisition — the kind of figure someone raises a budget on before realising they were paying for views.

SaveAds therefore only counts as a result what an advertiser would call one: a purchase, a lead, a sign-up, an application, a booked appointment, a contact or a conversation started on Messenger or WhatsApp. Clicks and views stay visible, but they never inflate the counter. If your campaign optimises for a custom event, the figure shown can be lower than in Ads Manager: that is deliberate.

A Meta ad account delivers on several platforms at once, and cost per result is not remotely the same on each. The Meta logo alone suggests a single channel, and therefore an average cost that matches none of the platforms it covers.

SaveAds reads each campaign’s spend platform by platform and labels it after the one receiving at least 60% of the budget — or “mixed” when none dominates. You can see that a campaign designed for Instagram actually spends mostly on Facebook, or that part of the budget goes to Audience Network, outside the two networks you had in mind.

Take a fictional account spending £1,500 a month across four ads. For three weeks all is well. Then one of them, a carousel launched at the start of the season, sees its cost per result rise from £20 to £38 while its click-through rate falls. Over the month it has absorbed £300 for a handful of results.

Nothing in Ads Manager flagged it: the campaign is active, delivery is normal. Compared with the three other ads in the same ad set, however, it stands out from the second week. Refreshing that creative puts the month’s £1,500 back on ads that still convert.

Today the permissions Meta grants at connection time cover reading. A Meta recommendation therefore appears as an action for you to take, with the exact place to take it in Ads Manager: which campaign, which ad, and why.

Write permission is being obtained from Meta. Once it is granted, SaveAds will create and optimise your Meta campaigns directly. Until then, no “apply” button that would fail on click.

A Meta account read on its own teaches little more than Ads Manager does. Read next to Google Ads, LinkedIn, TikTok and Snapchat, it enables the only question that matters at budget level: where does the next pound go.

The platforms appear in the same view, each with its currency. A Meta campaign at £25 per lead and a Google Ads campaign at £40 become comparable without a spreadsheet — while remembering that Meta often creates the demand Google Ads then harvests, when the same person searches for your brand after seeing your ad.

The SaveAds assistant reads the same data as the dashboard and answers with figures. A few questions that work well on a Meta account:

• “Which Meta ad saw its cost per result rise the most this month?”
• “Is my Instagram campaign really spending on Instagram?”
• “Compare Meta and Google Ads over the last thirty days, result by result.”
• “Which campaigns have spent without a single result for a week?”