Alright folks, I need to vent and maybe get a reality check. I've been running Amazon ads for a few years, mostly in the pet supplies space, and I dove into Rule-Based Bidding (aka Target ROAS) when it first rolled out, thinking it was the "set it and forget it" holy grail.
Spoiler: It wasn't. Not for new launches, at least.
My experience? I'd set a target, watch impressions and spend shoot up initially (because the system is "learning"), and then... crickets. The campaign would just flatline. I burned through a decent chunk of change before I realized the core problem everyone talks about but glosses over: This thing is data-hungry, and a new ASIN has no data to feed it.
I went back and tried to understand the logic. Here's my take, and I'd love to hear if yours matches up:
It's Just "Dynamic Bids - Up and Down" on Steroids. The old strategy lets Amazon decide when to raise/lower bids based on perceivedconversion chance. Rule-Based Bidding just adds a clear goalpost: "Hey Amazon, here's my target ROAS. Now, use that goal to decide when to bid up/down." The problem? The system still needs a mountain of historical data to build that "high-conversion customer" profile. A new listing has zero history, so the AI is just guessing wildly.
"Good" and "Bad" Traffic Are Relative to YOUR ASIN. This was the lightbulb moment. Amazon doesn't label customers as "good" or "bad" globally. A customer who loves buying premium dog toys might be "good" traffic for my 30indestructiblechewtoybut"bad"trafficfora5 plush squeaker. Without my own sales data, Amazon can't map this. So, the rule either bids too aggressively on the wrong people (wasting money) or too conservatively on everyone (killing impressions).
Your Target ROAS is a Traffic Valve. Set it too loose (like a ROAS of 1), and the system will bid on almost anything, burning cash. Set it too tight (a high ROAS), and the system, having no positive data to work with, will conclude "NO traffic meets this standard" and shut off your impressions. That's why new listings often see a spike and then a total drop-off.
So, who is it good for? My conclusion, and the reason I "stopped" using it for launches, is that Rule-Based Bidding is a tool for mature, data-rich products. If you have an established ASIN with consistent sales, thenyou can feed it 60+ days of data, set a target, and let the AI efficiently optimize towards that goal. It can finally differentiate between traffic worth bidding up for and traffic to avoid.
My current playbook:
New Listings: Start with manual or standard dynamic bids. Gather 30-60 days of conversion data first. No exceptions.
Established Products: Then, and only then, do I switch to Rule-Based Bidding. I calculate the initial target as 1 / (Current ACOS)and adjust in increments of no more than 10%.
Am I crazy? Has anyone else had success using Target ROAS right out of the gate on a new product? Or is it truly a "mature SKU only" tool like I'm thinking?
Answers (6)
I only use it on mature listings with tight copy, good reviews, and consistent sales. Even then, I adjust the ROAS target in 10% increments max — don’t go too tight too fast, or you’ll kill your exposure.
For small sellers especially, it’s a waste of time and money. Stick to manual or standard dynamic bidding for launches — you’ll have way more control, and you won’t watch your budget go up in smoke.
Problem is, for new listings, it never gets to the “optimizing” part — it just keeps bidding up to gather data, and you burn through your budget before it figures anything out. Feels like the algorithm gets more unpredictable every month, tbh.
Ended up switching back too — it felt like the AI was messing with my organic rank and eating into my profit margin. Turns out, even with a mature SKU, you need tons of consistent data for it to work right. If your sales are spotty, don’t bother.
Two days in, Amazon jacked my CPC up to $1.00, conversion stayed okay, but ACOS didn’t drop at all. Then impressions and clicks tanked — went from thousands to a few hundred a day. Total waste of cash.
Switched back to dynamic bidding, let it gather 60 days of data, and only then tried rule-based again. Worked way better once the AI had something to go off of.