Aaron Schwartz

The through-line

Aaron Schwartz's recurring argument is that retention marketing has become a data problem that brands are trying to solve with rules of thumb, and that the gap between the two is where their revenue is leaking. Acquisition has got more expensive and harder to target, so struggling brands turn to what they treat as their free channel and, as he puts it, "go email the heck out of my audience" [1]. The short-term lift is real, which is exactly what makes it dangerous: the brand sends more the next week, and more the week after, and "it works for a month and then you fall off a cliff" [1]. His fix is not to send less as a matter of principle but to stop deciding by calendar and start deciding by evidence, using the customer engagement data brands already hold to work out who actually wants to hear from them.

Underneath that sits a consistent posture about which decisions a brand should own and which it should not. Content, creative, merchandising, inventory, cash flow are the brand's own; working out who wants to hear from you on email is "a big data problem" where a single brand only ever sees its own data, and where the learnings of hundreds of other brands are worth more than in-house intuition [1].

On the over-emailing treadmill

The pattern he sees across the couple of hundred brands Orita serves is a brand that used to send three campaigns a week moving to four or five, experimenting with anonymisation tools, and treating the increase as free upside [1]. He is careful not to moralise about frequency, saying "none of these things are right or wrong" and that it depends on the brand and the strategy [1]. What he objects to is the escalation loop, and he gives three reasons the revenue drop that follows is steeper than the lift that preceded it. People unsubscribe because you are annoying them, and "there's probably a reason you had the cadence you had" [1]. Click rate falls, "Gmail starts downranking you," and a higher proportion of mail lands in spam rather than the inbox, which he distinguishes explicitly from the promotions tab [1]. And because acquisition is hard, there is no new audience arriving, so it is "literally the same message to the same people" [1].

He dates the current reckoning fairly precisely: Google changed its rules in 2024, brands panicked about infrastructure and spam rate, acquisition got harder at the same moment, so they sent more email and "forgot to think about click rate" [1]. The consequence now showing up is a well-loved, well-regarded brand with good collateral looking at a list of two million people and making $8,000 on a campaign, and the answer is simply that "you're not even hitting the person's inbox" [1]. Brands still on the escalation path, he warns, are in for a rude awakening after Black Friday and Cyber Monday [1].

On why time-bound segmentation breaks

His sharpest critique is of the standard engaged-segment definition: anyone who clicked, opened or visited in the last 60 or 90 days, stretched to 360 for something like a cookware company [1]. The tools available, he says, essentially only let you ask how many times somebody did an action over what period, then chain if-then logic on top [1]. He works the failure case through with an example. One person clicked 89 days ago and was on the site for a tenth of a second and makes the cut; another clicked 91, 92 and 93 days ago, has spent $10,000 with the brand and purchases every 108 days, and gets excluded, even though "I actually might be the most engaged person" [1]. Nor can a marketer easily build the obvious refinement, extending the window by ten days for each additional click, and even doing it by hand as an "insane consulting level" spreadsheet project leaves you wrong the next morning, because you are a day closer to a sale, a day further from Black Friday, or facing a product launch or different inventory [1].

He is explicit that the old approach was not stupid, agreeing that it was the only thing available at the time; what has changed is that machine learning at scale, in a timely fashion and at reasonable cost, now exists [1]. The scale of the waste is what he keeps returning to: the average brand Orita works with has around 250 million customer engagement data points in its email platform and is "probably using 10,000 to make a decision, maybe 100,000" [1]. The alternative he describes is a model updated every day that says whether a given person wants to receive an email, or force-ranks the list from 1 to 100 [1]. The compounding benefit is what he calls the magic: you stop emailing people who do not want to hear from you, so they do not unsubscribe and they do not register a non-click, and "by saving a not click ... you're actually improving your deliverability for the next person" [1].

On profit-optimal sending, not perfect deliverability

He resists the instinct to treat deliverability metrics as the goal. His analogy is a credit score: if you are at 770 it would take an insane amount of work to reach 800, so do not do the work unless a tenth of a point on an interest rate genuinely matters, just as a company can get by with a double-A rather than chasing AAA [1]. He also notes that plenty of brands under-send, so the exercise is not about shrinking volume; the thing that matters is inbox placement, and fixating on the metric leads you to optimise click rate or spam rate rather than think about the customer, when "the results that you're talking about will follow" if you start with when the customer wants to hear from you [1].

That framing lets him argue the opposite direction where the data supports it. Some brands send five emails a week and see 3 or 4% click rates in their top tier, and those people "could literally stand another email and they would make a bunch more money" [1]. Yes, some would unsubscribe and the click rate would drop to around one and a half percent, but he calls that "the profit optimal choice" [1]. The test is profit, not the health of any single metric.

On the consumer hat

He grounds the argument in his own history as a brand owner who "sent every email to every person every time," thoughtful about content and not at all thoughtful about whether the recipient wanted it [1]. His prescription is a switch of perspective: "take off your marketer hat and put on your consumer hat," and "don't be the brand that emails somebody when you want to. Be the brand that emails somebody when they want to hear from you and the data shows they want to hear from you" [1]. He is sympathetic about why this does not happen, since a brand owner is juggling four thousand things, a marketing lead two thousand, and someone in CRM still a thousand [1].

On multichannel retention and direct mail

He rejects the idea that older channels are dead, treating "catalogues don't work" as the same lazy claim as "email doesn't work" [1]. Orita started with email, is building its SMS product on the same platform, and intends to move onto as many platforms as possible so brands can bring whatever tool they want [1]. Direct mail through a partnership with Postpilot is the case he makes most enthusiastically: it is "a wildly profitable retention channel" provided you go deep enough on segmentation to know which part of the audience a postcard will actually drive incremental profit from, and if it will not, "don't ship it" [1]. Part of its appeal is structural, since a postcard is effectively an open by default: "it's on my counter, like I have to look at it before I throw it out" [1]. It is expensive, he concedes, and then adds the qualifier that runs through everything he says, "not if you target the right people" [1]. He also observes that brands are generally stagnant about trying new channels, and that someone who opens every one of your emails may still need a different medium at the right moment with the right discount to actually buy [1]. The endpoint he describes is optimising email, SMS and direct mail together into a much bigger customer journey [1].

On where growth actually comes from

Retention discipline does not, in his telling, remove the need for new audiences. Since more of the same message to the same people is a dead end, the question becomes how to create viral moments and how to run collaborations that bring in an audience you are not currently talking to but who might like the product [1]. The brands doing that are the ones thriving; the ones "saying the same thing just more of it" are struggling now and will struggle worse later [1]. He also refuses to treat expensive acquisition as permanent: "I don't think CAC being high is a fiat," describing a cycle in which costs expand, drop, a new channel appears, and the cycle repeats [1].

Takeaways

• Escalating email frequency to compensate for weak acquisition produces a short-lived lift and then a sharp fall, through unsubscribes, Gmail downranking clicks into spam, and repeating the same message to the same people [1].
• Standard engaged segments defined by a 60 or 90 day click window misclassify high-value buyers, and the manual refinements that would fix them are stale the day after you build them [1].
• The average brand Schwartz works with holds roughly 250 million customer engagement data points and uses perhaps 10,000 to 100,000 of them when deciding who to email [1].
• Suppressing a send to someone who would not have clicked improves deliverability for everyone else on the list, which is the compounding effect he is chasing [1].
• Chase inbox placement and profit rather than metric perfection; some top-tier segments should receive more email even if the click rate falls, because that is the profit-optimal choice [1].
• Brands should keep content, creative, merchandising and inventory decisions in house, and get outside help on who wants to hear from them, because that is a data problem where seeing only your own data is a handicap [1].
• Direct mail, run through the Postpilot partnership, is highly profitable as a retention channel when targeted, and worth skipping entirely for customers it will not move [1].
• With acquisition harder, growth depends on reaching audiences you are not currently talking to through collaborations and viral moments, and high CAC is cyclical rather than permanent [1].