Relaunching the Same Ad Didn't Reset Anything.

Scientific Logo, founded by Subhash Bothare, sells a ₹1,999 logo-analysis call to business owners. The call is the entry point to a higher-ticket branding programme, so a front-end return on ad spend below 1 is not automatically a loss.

Subhash is not a beginner. By his own account he has run his own Meta ads for about four years and managed ₹3–4 crore of spend. His team runs the account. He came to me in October 2026 because results had slipped and he wanted an outside view of why.

My role was a consultation: an audit of the ad account, the landing page and the funnel, and a 90-day roadmap for his team to run. I did not run the account, and nothing here should read as if I did. The audit covered ten live weeks, 3 August to 10 October 2026, in which the account spent roughly ₹22.5 lakh on 776 front-end purchases.

The account had grown to 127 campaigns. Almost every one was the same shape: one campaign, one ad set, one ad, aimed at the same broad India audience with the same purchase goal. 146 of 158 ad sets held a single ad, and 88 of the 125 campaigns that spent ended with five purchases or fewer, too few for Meta to learn from.

When an ad worked, the habit was to pause it and relaunch it as a new campaign. In late September a "Curious" series added 36 more campaigns at ₹1,000 a day, one hook each, which by October was taking almost all of the spend.

The logic is common, and I hear it often: Meta's newer delivery system, Andromeda, is said to reward fresh campaigns, so relaunching a winner should give it a new start. It was worth testing that belief against the account's own data before arguing about it.

Thirteen ads had been relaunched as new campaigns with identical targeting, which made a clean comparison possible: each ad's first run against its relaunch.

The relaunches were slightly better at one thing, getting people to click Book. They were worse at almost everything that costs money: they paid more per thousand impressions in 10 of 13 cases, converted fewer Book clicks into payments in 11 of 13, and had a worse return in 10 of 13. Combined, spend went up about 76%, purchases went from 190 to 187, and cost per purchase went from roughly ₹1,675 to ₹3,001.

One honest caveat: the relaunches ran after 31 August, when buyers fell across the whole account for a different reason. So this test cannot prove that relaunching caused the decline. What it does show is that relaunching gave no benefit, and the same sales cost almost twice as much. The copies also mostly ran one after another, not side by side, so they weren't bidding against each other. Each relaunch simply took the same ad back to the same people and started learning again.

The weekly numbers looked like one long slide. Splitting them by date showed two different problems, with two different causes, starting two weeks apart.

31 August, after the click. With the same kind of traffic, buyers per 100 landing-page visits fell from about 2.1 to 1.6. In the same week the checkout moved from a WordPress page to a GoHighLevel two-step payment page, and the landing page was rebuilt: a new headline, a shorter video, no price on the Book button, and a "report on WhatsApp" promise in place of a 30-minute call. The Book-click event also started firing more often per visit, which made the drop after Book look bigger than it really was. That is why we judged it on buyers per visit, the one number the event change couldn't distort. Both the iPhone and Android numbers moved on the same date, which points to the site rather than a device bug.

16 September, before the click. Clicks suddenly became very cheap and buyers almost disappeared. The activity log showed no change to targeting. What changed was one new ad in an existing ad set: a brand-gossip hook about a telecom company's logo mistake. With broad targeting, Meta shows an ad to whoever responds to it, and this one drew young men who follow phones and brands rather than business owners, at roughly a tenth of the usual cost per thousand views. It took about ₹58,000 for 65,000 clicks and 7 purchases. A second local-interest ad did the same about ten days later. Together the two took about 7% of spend and brought about two thirds of all clicks.

The tempting fix for the second leak was to restrict age. The data said no: the other ad in the same ad set sold well to 18–24 year-olds. The problem was the creative brief, not the audience.

A secondary finding sat under both leaks. The landing page scored 39 out of 100 on a Lighthouse mobile test, took about 9.6 seconds to show its main content and downloaded around 6.9 MB per visit, most of it testimonial videos that loaded before anyone scrolled to them. The Meta pixel only started after about 15 seconds in the test.

The cost showed up in the click data: on Android, about a third of paid clicks never registered as a landing-page visit, against only 2–3% on iPhone. Some of those people left before the page loaded, and some were never counted at all. A slow page doesn't just lose visitors; it hides them from the system that is supposed to optimise for them.

Andromeda first decides which ads are even considered for a person, and it does that better when it has genuinely different ads to choose from. Copies of one video count as one ad. Budget also learns faster in one place than across fifty small campaigns. So the recommendation was the opposite of the relaunch habit: keep one campaign that works and keep adding new, different creatives to it.

We ranked every creative in the account by cost per Book click, because the payment drop flattened differences in purchases, and 16 cleared the bar to carry forward. Testimonials and logo analyses made up most of them. Everything else, including the cheap-click ads, was retired with a rule the team could apply without me.

This is a diagnosis case study, published before the results. The audit and the 90-day roadmap were delivered in mid-October 2026, and the first two weeks of the roadmap belong to Scientific Logo's team: pulling payment data from mid-August, placing real test orders and fixing whatever breaks before ads go back on.

The cause of the 31 August drop is inferred from timing. The checkout and the landing page changed in the same week, and only the payment data can say which one mattered more, or whether it was something else. I will add a "what happened next" section once there is enough post-fix data to say something honest.