A write-up of a research project we did with Claude: first on a practical sales question, then on whether the famous psychology behind the usual advice actually holds up.
We started with a narrow, practical question for our info-product funnels:
What makes more money — upselling or downselling?
Do you open with a cheap product and work up to the expensive one? Or lead with the expensive offer and, if the customer declines, catch them with something cheaper on the way out?
We expected a clean answer with a percentage attached. We got something more useful and a little less tidy — and then, because we kept pulling the thread, we ended up checking whether the studies everyone cites for this stuff actually replicate. Some do. Several don't. Here's the honest version.
Part 1: The question "which is better?" doesn't have a clean answer
The most relevant academic evidence isn't a head-to-head test of sales funnels — it's a comparison of the two psychological mechanisms underneath them. Upselling leans on foot-in-the-door (a small yes makes a bigger yes more likely). Downselling leans on door-in-the-face (you ask for a lot, the customer says no, and you "concede" to a smaller ask). When researchers compared these two techniques across many studies, they found no clear winner — neither was reliably stronger than the other.
So "which is better in general" is probably the wrong question. A more useful one is: how warm is this customer at this point in the funnel? Our working framework — and we'd flag this as reasoning from the evidence, not a proven law — is roughly:
- Cold traffic, no trust, no reference price: work from the bottom. It's hard to sell an expensive offer to someone who has no anchor to judge it against and no reason yet to trust you. A cheap entry product that earns a first yes tends to do more work here.
- A customer who has just bought, card still in hand: there's trust and momentum, so leading with a higher-priced offer can set a useful anchor, with a cheaper fallback for anyone who declines.
That fallback is, in effect, door-in-the-face built into the checkout. It's a reasonable thing to add, and many sellers skip it — they stop at the first "no." But one honest caveat we'll come back to: the same research that supports door-in-the-face also suggests it's better at producing a verbal "yes" than an actual payment. So a downsell is worth testing, not a guaranteed win.
The broader takeaway from Part 1 is undramatic but practical: the money is less about choosing one tactic and more about ordering offers sensibly for where the customer is.
That would have been the whole post. But it rests on a stack of famous experiments — and lately a lot of famous experiments haven't held up. So we checked.
Part 2: Which of the "classic" effects actually replicate
Psychology has spent the last decade reckoning with a replication problem. When a large project re-ran 100 well-known studies in 2015, only about a third reproduced clearly, and the effects that survived were on average roughly half the size of the originals. So we asked Claude to look specifically at the studies our sales advice depends on, favoring replications and meta-analyses from 2015–2025. The results fell into two groups.
The ones that held up
Anchoring. The first number you see pulls your later judgment toward it. This is among the most reliably replicated effects in psychology: a large multi-lab project found it to be one of the strongest effects it tested, and a 2025 meta-analysis covering thousands of comparisons found a sizeable, stable effect that barely changes after correcting for publication bias. If you build pricing on one thing, this is the safest choice.
Door-in-the-face. In 2021 researchers directly re-ran Cialdini's 1975 experiment with about five times the sample, and the numbers came back close to the original — compliance rose from 34% to 51%. The effect is real and durable. The fine print, which marketing advice usually omits: it's noticeably better at getting a verbal yes than at producing actual behavior or payment. A genuine lever, not a money printer — and, as noted, this is the mechanism a downsell relies on.
Foot-in-the-door. A small initial request does make a larger one more likely, confirmed across many studies including online. The effect is modest, though, and a strong enough core offer can wash it out.
The endowment effect. People value something more once they feel they own it — the gap between what they'll accept to give it up and what they'd pay to get it runs to roughly 3x in the research. This is the mechanism a $1 trial leans on. The gap is smaller for experienced buyers.
The shakier ones
The decoy effect. This is the well-known "Economist subscription" idea: add a deliberately worse third option and people shift toward the expensive one. Well-powered replications in 2014 largely failed to reproduce it, suggesting much of the original effect came from artificial, two-number examples. The most realistic test to date — 3.6 million real grocery wine purchases analyzed in 2025 — found the effect does exist outside the lab, but it's small, on the order of a 1% shift in preference. Worth knowing about; not a reliable revenue lever.
Choice overload (the "jam study"). The claim that too many options reduces sales became a staple of marketing decks. But a meta-analysis of 50 experiments put the average effect near zero. A later analysis reconciled this: overload is real, but only under specific conditions — genuinely complex choices, high uncertainty, a tired or rushed decision-maker. As a blanket rule, "fewer options always sells more" doesn't hold; as a conditional one, it sometimes does.
Price-presentation order (show expensive first). The original study is well-designed and the theory is reasonable, but we couldn't find a large independent replication. Treat it as a hypothesis worth A/B testing rather than a settled fact.
The social-security-number anchoring demo. General anchoring is solid, but that one famous demonstration — where the digits of your SSN supposedly shift what you'll pay — has produced much weaker or null results on re-testing. Use the principle; don't lean on that particular story as proof.
What we took away
Two things, both fairly grounded.
First, the practical playbook for info products mostly rests on the effects that did replicate — anchoring, door-in-the-face, foot-in-the-door, the endowment effect. Start lower for cold traffic, consider anchoring higher inside a purchase, test a downsell rather than stopping at the first no, and don't pile on more than a couple of offers in a row. That's reasonably solid footing, with the honest caveat that the real-world effects are smaller than the famous studies imply.
Second — and this is the part we found more interesting — a fair amount of marketing psychology gets repeated without anyone checking whether it still stands. The pattern we noticed: the more vivid and quotable an effect is, the more likely it's been overstated, while the less exciting structural stuff (reference prices, sequence, ownership) tends to be what actually holds. That's a tendency, not a law, but it's a useful prior.
Mostly, the project was a reminder to test the "best practices" instead of inheriting them. The tactics matter. Checking whether they're true matters at least as much.
The full research — every study, sample size, effect size, and verdict, with the caveats about which numbers to trust — is here: the complete research report. We ran it with Claude; the report is in English, but a browser translator handles it fine.