Conventional wisdom in 2026 calls single keyword campaigns a relic, says broad match plus Amazon's algorithm has made granular structure obsolete, and writes off anyone still isolating keywords as busywork. We run advertising across a portfolio of brands for a living, and we still build single keyword campaigns for every term that matters, because the structure solves two problems blended campaigns cannot touch.
Attribution clarity comes first. With one exact match keyword in a campaign, every dollar of spend, every impression, every placement report, and every attributed order ties back to a single decision we made. A blended campaign report shows that forty targets shared the spend, and we end up inferring which one earned the sales, which is exactly where quiet errors creep in. With a SKAG setup the campaign report is the keyword report, so nothing needs untangling.
Ambiguity also breaks automation. A bid rule aimed at a mixed campaign has to guess which target caused the outcome it is reacting to, and a rule that guesses is not one we trust with money. A bid rule aimed at a single keyword campaign acts on a fact. If this target racks up clicks with no orders, the rule cuts the bid, and there is no question about which product or which search intent it is punishing. That determinism lets us hand the daily work to software, and it is why we open sourced the engine that runs it at the public tool archive. The engine reads performance on a schedule and applies the same rule to the same situation every single time, which is a standard no human operator meets across two hundred campaigns.
Launch conventions make the structure behave. We launch on fixed bids rather than the dynamic down only setting, because we want the data to reflect the bid we chose instead of whatever Amazon discounted it to in each auction, and clean inputs are the point of the exercise. Every new campaign starts at ten dollars per day, which is enough for a single keyword to prove itself without letting any one term starve the rest of the account. Starting bids sit at the low end of Amazon's suggested range, because discovery should be cheap and the rules will raise a winner on their own.
Hygiene keeps the structure honest. When a search term proves itself inside a research campaign, it graduates into its own exact match campaign, and we negate it at the source in the same motion so the term converts in exactly one place. Skip that dedupe step and the same query bleeds across campaigns and the attribution clarity we built the structure for evaporates.
The tradeoff is surface area. Single keyword campaigns multiply the number of objects you manage, and at portfolio scale nobody can review them all by hand, which is why the structure and the automation have to ship together. The campaigns make the data unambiguous, and the engine makes the workload irrelevant. People who abandoned SKAGs usually abandoned the manual labor, and we agree about the labor. We kept the structure and removed the human from the loop instead.
← Back to all posts