Ramping new products (incubation)
Why new and thin-data products struggle under tROAS, what incubation is, the graduation criteria that promote a product to your main campaign, and how BidSmart's incubator runs the whole process approval-first.
New products have no conversion history, so Target ROAS has nothing to predict from and either won't spend or spends wildly. Incubation solves this: run new products in a separate campaign whose only job is buying data cheaply, then promote each one to your main tROAS campaign once it has proven it can convert. Products graduate on data, not on a calendar.
Why new products break tROAS
Target ROAS works by predicting the conversion value of every individual auction and bidding to hit your ratio. That prediction runs on history: this product's conversion pattern, its order values, its shopper behaviour. A brand-new product has none of that, so one of two bad things happens:
- It goes invisible. With no evidence a click will convert, the algorithm judges most auctions not worth entering at your target and simply stops bidding. The product never gets the impressions it would need to prove itself, a chicken-and-egg trap.
- It spends erratically. Where the algorithm does bid, it's guessing off category priors, and a few early conversions (or none) swing its estimate wildly. You get feast-or-famine spend with no stable signal.
Either way, dropping a new product straight into a tROAS campaign is not a launch, it's a coin toss. The fix is to change the product's job for a while.
What incubation actually is
Incubation runs new and thin-data products in a separate campaign whose goal is buying conversion data cheaply, not making a profit. In practice that means a Maximise Clicks strategy with a modest budget and a max CPC cap: enough traffic to let conversion behaviour emerge, without betting real money on an unproven product.
The mental shift is the important part. Inside the incubator, efficiency is not the metric: data accumulation is. You are deliberately paying for clicks to learn whether the product converts, at what value, and against what queries. That's information you can't get any other way, and it's cheap compared to the cost of an unproven product mispriced in your main campaign.
Keeping incubation in its own campaign matters for a second reason: it protects your main tROAS campaign's learning. Thin, noisy products don't drag the average or add variance to the auctions your proven products depend on.
Graduation criteria
A product leaves the incubator when it has earned its way out, on evidence, not elapsed time. The criteria that matter:
- A conversion threshold: roughly 15–30 conversions in the product's own history, enough to compute a ROAS that isn't just noise.
- Stability: those conversions spread over time, not one lucky day, so the pattern is real.
- A sustainable target emerges: the product's history implies a tROAS it can actually bear, given its price and margin.
Meet those, and the product moves to the main tROAS campaign at its own derived target. Fail to meet them after a fair run, and the product is telling you something worth hearing: see the FAQ above. For why thin data is so hard to reason about, see bid simulators and the data-thinness problem.
Deep dive How BidSmart's incubator runs this, approval-first
Doing incubation by hand means creating a separate campaign, filtering the right products into it, watching each one against a conversion threshold, and manually moving graduates across at the right target, repeated forever as new products arrive. BidSmart's incubator automates that loop, but never takes the decision out of your hands.
Here's what it does:
- Runs the separate Maximise Clicks campaign for new and thin-data products, with its own budget and CPC cap, kept apart from your main tROAS campaign so incubation noise never touches proven-product learning.
- Watches every product against the graduation criteria continuously: conversion count, stability, and the target its history now supports.
- Stages the graduation for your sign-off. When a product is ready, BidSmart proposes the move to the main campaign at a specific target and shows you the evidence behind it. The move does not execute until you approve it. Nothing in BidSmart auto-executes, recommendations are always queued for explicit approval first.
The plumbing that makes graduation painless is the feed: moving a product between the incubator and the main campaign is a lifecycle-stage custom label change, not a campaign rebuild, so the receiving campaign's learning isn't disturbed. The mechanics of the incubator campaign itself are covered in incubator campaigns.
Common mistakes
- Graduating on a calendar. "It's been a month, move it" ignores whether the product actually gathered data. Promote on conversions, not dates.
- Skipping the incubator to save effort. Dropping new products straight into tROAS is the invisible-product trap above. The incubator is faster overall because it avoids weeks of erratic spend.
- Setting the incubator budget too high. The goal is cheap data. A generous budget on Maximise Clicks buys volume you don't need and clicks you can't yet value.
- Never revisiting stuck products. A product that won't graduate is a diagnosis waiting to happen, usually a feed health issue throttling its impressions before it can prove anything.
Frequently asked questions
Why does tROAS struggle with new products?
tROAS bids by predicting each auction's conversion value from history. A product with no history gives the prediction engine nothing, so it either refuses to bid (the product goes invisible) or bids erratically on guesses. Neither is a controlled way to launch a product.
How long does incubation take?
There's no fixed duration, it's data-driven, not time-driven. A fast-moving product might gather enough conversions in a couple of weeks; a niche one might take months. Graduating on a conversion threshold rather than a date is the whole point.
What if a product never gets enough data to graduate?
That's a signal in itself. A product that can't accumulate conversions cheaply in the incubator is telling you something, weak demand, a feed problem, or an uncompetitive price. Fix the cause (often the feed) or accept that it isn't ready for a profit-based target.