Selling an Indian brand overseas. The budget stayed fixed. The ROAS did not.

The instinct with a 1.89x account is to rebuild it on day one. We spent the first two months making it readable instead: a naming convention carrying product and geography, one measurement window, and enough spend held steady to separate signal from noise. February moved a rounding error, which was the point. By the end of it we knew clicks were cheap, conversion was thin at 2.49%, and the basket was small at $65. The leak had an address, and it was not the auction.

March ran the highest spend of the window, roughly 24% above January, and the return went up rather than down. That settled whether efficiency was a small-budget illusion. It was not. March also separated out the products that would carry the second half, Mud Mask, Milk Powder, SPF 50 and the anniversary offer, as readable campaigns rather than buried inside a catch-all, which is what made the later merchandising decisions possible.

Spend came down 37% and concentrated on what March had proven. Conversion hit 3.07% and the average basket peaked at $79.92, the highest of the seven months. Worth stating plainly: cost per click rose to $1.04 in April, the most expensive traffic we bought all year. We bought it anyway, because the funnel was now converting well enough that expensive clicks still paid. That trade is the whole engagement in one sentence.

Three campaigns launched that month took 31.6% of the budget and returned 2.02x, while the campaigns carried over from April held 2.19x. So May was the visible cost of a launch wave running into a normal seasonal softening, not a performance failure. We did not restructure in response. The launches got a fixed read window and a floor, and by June the two that had not earned their place were off the account.

June and July ran the tightest structure of the window: eight campaigns between roughly $2,000 and $4,300 each, none dominant, all proven. Face Brightening Kit and All Kit joined at 2.88x and 2.87x. July closed at 2.94x and produced $110,648, the highest revenue month of the seven, on 28% less spend than March. Cost per purchase finished at $26.12 against January's $34.54.

A second channel does not fix a first channel, it multiplies whatever the first one is already doing. So Meta had to be compounding before anything else got a dollar, and we took it from 1.89x to 2.50x first. Only once the funnel behind the ad was converting above 3% did TikTok make sense, because by then anything we pointed at the site converted better than it would have in January.

We built the TikTok account from zero: structure, campaign architecture, creative format, market prioritisation and offer logic. Creator-led video at the top to reach diaspora audiences Meta was saturating, with catalogue and remarketing underneath for efficient conversion while the top of funnel learned. In our hands it ran roughly $12.7K and 138 orders at a $92 blended cost per order, with ROAS climbing from about 0.5x to 0.8x. Singapore produced the efficient orders. The US audience converted at zero and was cut rather than nursed. Once the account was built, structured and growing, we handed day-to-day running to Vilvah's internal team; the channel now sits in-house by design.

Reading campaign names back into product families shows the account's real economics. Milk Drops is the engine and always will be, but it is not the most profitable dollar in the account, and the two best-returning lines hold only 6.3% of budget between them.

Face care has the lowest basket in the account at $57 and the lowest return at 1.73x. It does not need better targeting. It needs to stop being sold as a single item.