Say you spent $1 million on a channel and it returned $2 for every $1 a clean 2x ROI. Now ask the question that should actually drive your next budget meeting: if you spent $1,000,001, would that last dollar still return $2? It likely wouldn’t and most brands don’t have visibility into the long-term impacts when evaluating short-term metrics.
That gap between your average ROI and your marginal ROI is where most media plans quietly go wrong. In this episode, Carve Designs’ Hannah Fleming and Keen’s Jesse Math walk through exactly where that gap shows up in direct mail, Connected TV, TikTok, and the bets that looked wrong before they looked right.
Why ROI breaks down the bigger you get
The first dollars you spend on any channel are almost always your best dollars. They hit your most responsive audience, your cheapest inventory, your most obvious demand. Every dollar after that works a little harder for a little less. That is the diminishing returns curve, and it’s the real decision layer hiding underneath ROI.
Most brands never see it, because most measurement tools report one number: the average. Hannah describes how Carve handles this directly, treating contribution per piece as a core KPI for direct mail rather than relying on blended ROI alone. A distinction that matters more as a channel scales, not less.
The halo effect is real and it’s invisible to last-click
Carve has run catalogs for ten years. The lift into email and SMS shows up clearly in match-back data and holdout panels, but the harder question, how much of that revenue was incremental versus just pulled forward from another channel, only shows up when you measure causally, not deterministically.
Jesse makes the case that this is true everywhere, not just in direct mail: Connected TV shows roughly 30% of its impact in the short term and 70% over time, which means a channel can look underwhelming in-platform while quietly reshaping performance in search, direct traffic, and even Amazon. Hannah’s team sees exactly this pattern — conversion rates lift across every other channel once CTV enters the mix, even though CTV itself rarely gets the credit in last-click reporting.
Patience is a strategy, not an excuse
Not every channel pays back on day one, and that’s not automatically a red flag.
TikTok showed no measurable ROI for six months before momentum arrived on both organic and paid. Whitelisting took Carve eighteen months to become one of its top-performing formats. Podcasts and out-of-home tend to run even longer. Jesse notes it’s rare for a channel to have long-term impact without some short-term signal first, which means the real question isn’t “did it work,” it’s “how much of it worked, and on what timeline.”
That’s the shift Hannah describes in her own planning: instead of asking whether a channel is binary pass or fail, she’s asking what percentage is working and tuning from there a more granular read on incrementality than most teams are set up to make.
Budget allocation that respects the curve
Scaling a channel that works isn’t automatic. Carve segments direct mail by recency (0 to 12 months versus 13 to 24 months) and continuously tests catalog size, frequency, and in-home timing, because growing a channel 20% doesn’t mean it stays as efficient as it was at its current size.
This is the next-dollar question in practice: not “is this channel good,” but “where, specifically, does the next dollar create the most incremental value right now.” It’s the question Jesse poses directly to Hannah mid-episode, and it’s the same logic behind how Keen’s AI Cortex models response curves across a brand’s full mix — short-term and long-term, deterministic and halo, channel by channel.
Watch the full conversation
Hannah and Jesse go deeper on all of this in the episode above: Carve’s live match-back testing process, how CTV planning changes the shape of the rest of the media calendar, and why Pinterest — despite looking like an obvious fit for a women’s apparel brand — simply hasn’t worked yet. It’s a case study in a team that measures first and reacts to the data, not the hunch.
