Myth 1
Most clients select one retargeter
The reality:
The vast majority of clients have one or more all-purpose retargeting solutions (RTB House, Criteo, Google DV360, TTD, AppLovin).
Such clients also use Meta tags on their websites. This allows Meta to push high-quality retargeting ads (DPA) to the same audience.
Clients also use Google's PMAX. PMAX, with little ability for clients to opt out, pushes retargeting ads across a wide range of Google- and non-Google-owned surfaces, as determined by Google’s algorithm.
An increasing number of clients have added complementary retargeting solutions to fill gaps and improve performance, including email and postal retargeting.
Deep Learning is different in kind, and clients are aware that a high-performing retargeting stack is built by using a set of partners with different underlying technical characteristics.
Myth 2
Multiple partners increase what you pay for inventory in auctions (“bid inflation”)
The reality:
The second-price auction mechanism, in which you paid what the next-highest bidder would pay, was sunset in 2018-2019.
You now bid what you pay. Only Machine Learning systems that try to game the SSP auction are fearful of more competition.
Using two different-in-kind retargeting bidding algorithms allows each one to shine, giving the marketer more scale and better ROAS.
Most vendors respond to competition by “giving you their best.” Single-sourcing anything leads to poorer performance.
It is true that outdated bidding models using Machine Learning are less able to compensate for the complexity of a retargeting stack when thinking about how to bid. However, modern bidding models using Deep Learning handle this degree of complexity by design. Deep Learning seamlessly keeps bids optimal and doesn’t rely on yesterday’s concepts, such as "perceived demand.” It uses straightforward true-value bidding: you bid based on what the ad is actually worth. If Deep Learning doesn’t value the ad highly—or win the auction—there is no increase in the bid of any other player, meaning no bid inflation. If Deep Learning values the ad the most, the client gets additional incremental inventory at the right price, resulting in a better return on investment.
Myth 3
Audience “overlap” hurts performance
The reality:
The vast majority of users require multiple touches, multiple approaches, and multiple versions of creative to move fully from site visitor to purchaser.
Integrated marketing is built on the premise that the variation in approaches is synergistic. Decades of marketing practice strongly suggest that a “stack” approach in any programmatic marketing tactic outperforms a static single-source vendor approach.
Using a single retargeter is fighting with one hand tied behind your back. The “Before They Buy” Consumer Research study from early 2026 demonstrates that both in-cart dwell times and the time from product discovery to purchase are increasing. Marketers need more touchpoints than ever to encourage users to convert—and different users are moved toward conversion by different combinations of placements, creative, product recommendations, and messages.
Imagine a typical ecommerce site visitor with a potential purchase set of 30 products. If we consider a retargeting ad with 4 product slots, 6 potential sizes, and 8 ad formats—that’s over 1.3 million combinations. A one-vendor approach is likely to leave significant gaps, even within a large user population. A retargeting stack lets each vendor find micro-segments or users to match across those 1.3 million combinations.
Finally, from a marketing KPI standpoint, overlap is meaningless. It’s yet another outdated vanity metric—like simple impressions—that legacy vendors use to lead marketers towards sub-optimal solutions. What matters is the ROAS generated by the marketing—and the proof that return on investment is incremental. RTB House has run dozens of randomized controlled experiments that demonstrate this principle. The fact that RTB House generates incremental sales, even when added to an existing stack, proves conclusively that “overlap” is not a flaw but a feature of best-in-class marketing.
Myth 4
Multiple providers create frequency blindness
The reality:
It is true that users see ad impressions from other vendors and are exposed to multiple touchpoints. However, this is part and parcel of marketing in general. Your users might, for example, see 6 ads on Meta, 5 on YouTube via PMAX, and 8 ads on open internet sites. And don’t forget about your touches via email, SMS, and non-addressable media (TV, radio, etc.). You therefore need an open internet partner who isn’t afraid to work within these real-world internet constraints.
Machine Learning algorithms are particularly at risk when they can’t accurately count or estimate the frequency of an ad exposure. Their outdated reliance on frequency count leads to difficulty for users who might also be on Meta or Google inventory. When a vendor complains about “frequency blindness,” it’s a clear signal that they're relying on outdated Machine Learning.
Deep Learning, by contrast, is different in kind. It isn’t nearly as dependent on frequency and doesn’t require summary statistics when making predictions. Instead, Deep Learning uses the full, glorious detail associated with a user’s web-browsing and ecommerce experience.
Machine Learning suffers from Frequency Dependency: models that are overly reliant on data points that are hard to know precisely. Break free from this ailment by upgrading to Deep Learning.
The cost of single-sourcing critical marketing tactics
Supply Chain 101 teaches buyers of goods that single-sourcing any key input is dangerous. In digital media, that’s even more true. Most performance systems work by starting with the lowest-hanging fruit first and then moving on to more challenging users. That’s fair—and how it should work. It’s a mathematical truth that if you split a budget across multiple providers, you are very likely to increase your overall ROAS.
In addition to the fundamental nature of performance (better with a stack), RTB House has modeled the specifics related to ecommerce. We’ve simulated over 10 M bid requests and followed what happens across website micro-segments when ecommerce companies single-source retargeting vs. when they employ a retargeting stack.
Performance increases substantially (8-11%) when a stack is employed, but principally when the underlying technology of the stack’s components is different. For example, adding Deep Learning to a stack that contains Machine Learning / Meta / PMAX produces clear benefits for marketers.
The proof is in the results
There is a reason RTB House’s retargeting product is growing far faster than offerings from other AdTech companies: once you test a stack, it’s hard to go back to a lower-performing setup.
Even better, because Deep Learning understands user behavior more holistically than Machine Learning, it drives not just attributed sales but also incremental sales in the context of your existing marketing stack. Deep Learning seamlessly layers on top of what marketers are already doing, without requiring endless split-testing or complex integrations.
Real sales. Incremental results. Minimal disruption.
