Creative production has historically operated on a hero model. A brand commissions a small number of expensive assets, develops them carefully, and commits media budget behind whichever concept survived internal review. The economics enforced that structure, since production costs made generating dozens of alternatives impractical.
That model concentrates risk in a way most marketers acknowledge but few can avoid. If the central concept fails to resonate, there is limited recourse beyond commissioning another expensive asset and hoping. The failure is expensive, slow to detect, and produces little usable information about why it failed.
Street interview advertising inverts that structure, and Street Poller Media has built its commercial model explicitly around the inversion.
The agency operates on three-month engagements targeting a reduction of at least 20 percent in a brand's blended cost per acquisition. Founder Shane Ginsberg has explained why he treats that target as realistic rather than aspirational, and the reasoning is closer to statistics than to creative confidence.
A typical engagement produces roughly 10 street polls monthly, accumulating to 30 unique polls across three months, which expand to approximately sixty total video assets once hook variations and edits are accounted for. Given that volume, Ginsberg has argued the probability of failing to produce at least one genuinely high-performing advertisement approaches one percent.
The claim is not that the agency can predict which asset will succeed. It is that sixty attempts against a format with a demonstrated hit rate makes at least one winner close to inevitable, in the same way no individual coin flip is predictable while the distribution across sixty is quite constrained.
That framing represents a different philosophy than the hero model rather than a refinement of it. Individual asset cost is low enough that volume becomes the strategy rather than a luxury. Instead of predicting in advance which concept will resonate, the model generates enough attempts that a winner emerges reliably, then redirects media spend toward whichever assets the data identifies. The approach borrows more from direct-response testing than from traditional creative development.
Several assumptions deserve to be visible, since the argument is stronger when they are. The volume logic depends on the underlying format possessing a genuine hit rate rather than a random one, which is an empirical claim resting on prior campaign history rather than anything inherent to the arithmetic. A brand in an unusual category, or one whose product resists the format structurally, would not necessarily see comparable distribution.
The 20 percent target also measures against existing performance, meaning achievable improvement depends substantially on how well-optimized that baseline already is. An advertiser running poorly performing creative has considerably more room than one already operating efficiently, and the same percentage improvement represents very different underlying work in each case.
What the model does address is the most common failure mode in creative marketing, which is committing significant budget to a single concept and discovering post-launch that audiences do not respond. Sixty attempts distributed across a quarter makes that specific outcome structurally difficult.
The secondary output may be more valuable than the winning asset. A campaign structured this way produces a documented record of which hooks, questions, and framings failed, which is information the brand did not previously possess and which carries forward into subsequent work regardless of who executes it.
There is a cultural obstacle to adopting this approach that has little to do with its merits. Volume-based creative production requires marketing organizations to become comfortable with most of their output failing publicly, which conflicts with how creative work is typically reviewed, approved, and defended internally. An asset that underperforms is treated as a mistake rather than as an expected outcome of a testing process, and that framing makes the model difficult to sustain in organizations where individual campaigns carry reputational weight.
That accumulated learning is the part hero-model production never generates, because three expensive assets that underperform tell a marketer very little about why.
Copyright © 2026 California Business Journal. All Rights Reserved.
For California Business Journal Disclaimers, go to /terms-conditions/.
