· 4 min read · Jason Youngjin Koh
Scale what works, test the rest
When sales stall, the instinct is to change everything. The brands that grow keep what works, fund it harder, and run a small, disciplined test budget beside it.
Put about 80% of the ad budget behind campaigns, products and creative that are already proven, and 20% into deliberate tests. Winners from the 20% graduate into the 80%. Losers get cut on a rule set in advance. An AI agent watches both sides every day and flags what should move.
Why not rebuild everything when growth stalls?
A rebuild throws away the one thing you know for certain: what already sells. New branding, new campaigns and new audiences all at once means that when results move, nobody can say why. It also resets Meta's learning on every campaign you touch, so the first weeks of the new plan are paid for twice.
A stalled account is usually under fed. A handful of products and ads carry most of the revenue, capped by budget or worn out creative, while spend leaks into things that never worked. Finding that split is the first job.
How do you find what is actually working?
Start with six to twelve months of sales and ad data, broken down by channel, product, offer and audience. Three questions: which products earn the margin, which ads bring customers who come back, and where the money goes that returns nothing. That diagnosis takes the first weeks of any engagement we run, and it decides everything after.
Rank products by margin and return on ad spend together. A best seller on a thin margin can lose money on every ad dollar. A mid seller with strong margin and good imagery is often the product to push hardest.
Why build distribution before promotions?
A sale is only as big as the channels that carry it. Before planning a promotion, check that the ad account, the retargeting audiences, the email and Zalo lists and the creator content are in shape to deliver it. A 30% discount pushed through a thin audience sells to the same few hundred people who would have bought anyway.
Build the channels in the quiet months and run promotions through them in the loud ones. In Vietnam that means being ready before 9.9, 10.10, 11.11 and 12.12. In Australia, before Click Frenzy, Black Friday and Christmas.
What goes in the 80% and what goes in the 20%?
| Share | What it funds | How it is judged |
|---|---|---|
| 80%, maintain and scale | Proven campaigns, best margin products, ads with a track record, retargeting | Holding cost per purchase while spend rises |
| 20%, test | New creative angles, new products, new audiences, new formats such as creator content or live selling | One question per test, a fixed spend, a pass mark set before it starts |
When do you scale a test and when do you kill it?
Set the pass mark before spending a dollar. For example: a new ad has to reach purchases at or under the account's target cost within a set spend. If it does, it moves into the 80% with a modest budget increase, and keeps climbing while the cost holds. If it does not, it stops, and the lesson goes into the research file so nobody tests the same idea twice.
Change one thing per test. An ad with a new hook, a new product and a new audience that wins has taught you nothing you can repeat.
Where does AI fit in the loop?
In the watching. An agent reads every campaign daily and reports three lists: what is working and can take more budget, what is fading and needs fresh creative, and what should stop. It catches what people miss at scale: a winning ad whose frequency has doubled, a product whose cost per purchase crept up over a fortnight, a test that quietly beat the control.
People decide the moves. The agent makes sure no decision waits for someone to notice.
What happens to the budget when the brand changes?
Rewrite the proven campaigns into the new voice gradually, and test each rewrite like anything else. Switching off every winning ad on the day a new brand launches is the most expensive way to adopt it. Keep the mechanism that sells and change the words on top of it, so every ad, product page and email ends up telling the same story without a revenue gap in between.
What does the first quarter look like?
- Weeks one and two: diagnose the account, choose the high margin products to lead with, produce new product imagery and a first round of creative.
- Month one: rebuild the campaign structure around proven winners, fix tracking so every purchase is measured, start the test budget.
- Months two and three: scale what passes, retarget with creator, testimonial and feature ads, and push the proven products hard into the peak season.
- After peak: rebuild and experiment, replicating winning ads across new products and accounts.
The 20% can test products as well as ads, before any stock exists. Test the product before you make it