The Middle Tier And Its Margin

The eighty-third article was about four words in a bracket. This one is about a pricing structure so common that most owners adopt it without deciding to, and about the one number that determines whether it pays.

The Verdict, Stated First

Five claims, in descending order of confidence.

One. The effect was predicted before it was found, which is rarer and stronger than the usual pattern. The 1989 paper derived it from a theory rather than discovering it and explaining it afterwards.

Two. It is stronger when the buyer must justify the choice, which the abstract states directly and which makes it a business-to-business finding as much as a consumer one.

Three. It has fared better than its sibling. The same paper's other prediction has been, in a later paper's words, strongly questioned, and this one has a meta-analysis devoted to it.

Four. On our own arithmetic the revenue effect and the profit effect can point opposite ways. A share shift to the middle raises revenue by construction and raises profit only if the middle tier's margin clears a threshold we derive.

Five. The businesses most at risk are professional services firms, where the upper tiers mean more of somebody's time, rather than software firms where they mostly mean more software.

Our Grades For These Claims

Applying the scheme from the first article in this series.

Grade A for the 1989 findings, from an abstract obtained verbatim from the publisher, though our reproduction of its third finding is cut off.

Grade B for the divergent fates, which reach us through a bibliographic service's summary of a later paper rather than from that paper.

Grade D for the meta-analysis, which reaches us as a citation in a reference list, with no findings whatever.

Grade C for the practitioner cautions, which come from a commercial pricing consultancy rather than from research and are flagged at every use.

Grade A for our own arithmetic, which is exact and which contains two errors we made and corrected, both documented below.

A Note On Method

Everything here is verified to August 2026.

We obtained the 1989 abstract verbatim from the publisher[1]. Our reproduction cuts off partway through its third finding, which we flag where it arises.

We obtained a bibliographic service's characterisation of the state of the sibling effect[4], and note that this is a summary of another paper's framing rather than a finding.

We obtained citations confirming a meta-analysis of extremeness aversion exists[3] and no part of its contents.

One source is a commercial pricing consultancy rather than an academic one[5], flagged at every use.

We did not obtain the 1989 paper, the 1992 paper, the meta-analysis, or any replication study.

All arithmetic is ours and every price, margin and share in it is invented. This article discusses research on pricing and is not pricing advice.

One Paper, Two Predictions

The structure of the source, which is what makes this article a companion to an earlier one.

Simonson, I. (1989), Choice Based on Reasons: The Case of Attraction and Compromise Effects, Journal of Consumer Research, 16(2), 158–174, September, DOI 10.1086/209205[1][2].

Four observations, ours.

The title names two effects, and the seventy-fifth article in this series covered the first of them under its other name, the decoy effect.

The paper's contribution is not two findings but one theory generating both. That is a stronger scientific move than reporting two effects, and the next section shows what the theory is.

An economics database records that no abstract is available for this item[2], which is a reminder that where you look determines what you find. The publisher carries it.

And the two effects have had markedly different subsequent lives, which is the part of this story worth telling and which we come to below.

Two reasons that divergence is worth a business reader's attention rather than only a methodologist's, ours.

The two effects prescribe different things. One tells you to add a deliberately inferior option; the other tells you to add a genuinely better and more expensive one, and only the second describes a product you would be willing to sell.

And they cost differently. A decoy is a product you hope nobody buys; a top tier is a product you must be able to deliver, which is a real commitment and the reason the arithmetic below matters.

What The Abstract Says

The theory and the findings, in the paper's own words.

The proposal: "Building on previous research, this article proposes that choice behavior under preference uncertainty may be easier to explain by assuming that consumers select the alternative supported by the best reasons. This approach provides an explanation for the so-called attraction effect and leads to the prediction of a compromise effect."[1]

The findings: "Consistent with the hypotheses, the results indicate that (1) brands tend to gain share when they become compromise alternatives in a choice set; (2) attraction and compromise effects tend to be stronger among subjects who expect to justify their decisions to others; and (3) selections of dominating and compromise brands are associated with m", at which point our reproduction is cut off.[1]

Four observations, ours.

The phrase "leads to the prediction of" is doing important work. The compromise effect was derived from the theory before being tested, which is a genuine prediction rather than a post-hoc explanation.

That distinguishes it from a great deal of what this series has covered. A theory that generates a novel prediction and then finds it is doing what theories are supposed to do, and the seventy-first article's complaint about unfalsifiable framing does not apply here.

The mechanism proposed is reasons rather than preferences. People pick what they can justify, and a middle option is justifiable in a way an extreme one is not.

And we do not have the third finding, which our reproduction truncates mid-word. It concerns something associated with selections of dominating and compromise brands, and we will not guess.

Two notes on that truncation, since a reader may reasonably want to know how much is missing. The sentence breaks after "associated with m", so a single word or phrase is absent and the finding's direction is unknown to us.

We could speculate from the paper's theory about what a third finding would likely concern, and we will not. Guessing at a finding from a theory is precisely the failure mode this series documents in others, and the omission costs the article less than the guess would.

The Justification Finding

The second finding, which is the commercially important one and is almost never quoted.

The abstract states the effects are "stronger among subjects who expect to justify their decisions to others."[1]

Four observations, ours.

This makes the finding specifically about accountable buyers, which is nearly every business purchase above a trivial amount.

It also supplies the mechanism in a form a seller can act on. A middle option is the one a buyer can defend: not extravagant, not cheap, and the choice explains itself in a sentence.

Which predicts something testable and specific. The effect should be stronger in business-to-business selling than in consumer selling, and stronger still where a purchase requires sign-off from someone who was not in the conversation.

And it suggests a piece of sales practice that costs nothing, ours and untested. Giving the buyer the sentence they will use to justify the choice is doing the work the mechanism says they are already trying to do.

And it identifies where the effect should be weakest, which matters for whether the structure is worth building. A sole owner spending their own money answers to nobody, and on this mechanism should be least moved by it.

One qualification on that, ours, because it is too neat. People justify decisions to themselves as well as to others, and a sole owner explaining a purchase to a spouse or to their own future self is not obviously outside the mechanism.

What the abstract supports is a comparison rather than an absolute. Stronger among those expecting to justify to others, which leaves the baseline case intact and merely says the effect is larger in the accountable one.

Two Effects, Two Fates

What happened to each prediction, and this is where the two articles diverge.

A bibliographic service, summarising a later paper's framing, records: "Researchers have recently strongly questioned the robustness of the attraction effect, according to which adding a decoy option to an existing choice set affects consumers' choice behavior."[4]

The same service characterises the other effect without any such qualification: "The compromise effect dictates that a decision-maker chooses a middle option over an extreme one given a set of choice alternatives since choosing an intermediate option is easier to", at which point our reproduction is again cut off.[4]

Four observations, ours.

This is a bibliographic service summarising other papers, not a finding, and we grade it accordingly. It tells us how the field is being described, not what is true.

The asymmetry in the language is nonetheless striking. One effect is introduced with "researchers have strongly questioned" and the other with "dictates", in summaries of papers drawn from the same literature.

The seventy-fifth article documented the questioning of the first effect in detail, including the original author's published response[3], and we will not repeat it here.

And the asymmetry is the reason this article exists separately. A business that read the seventy-fifth article and concluded that context effects are unreliable would be drawing a wider lesson than the evidence supports, because the two effects are not in the same evidentiary position.

One caution in the other direction, because we do not want to overcorrect. The absence of a documented controversy is not evidence of robustness, and the compromise effect may simply have attracted less replication attention than its more famous sibling.

The honest statement is narrow. We found a live dispute about one and no dispute about the other, and we could not obtain the meta-analysis that would settle it either way.

A Meta-Analysis Exists

The strongest single piece of evidence about this effect, and we have none of it.

A reference list confirms: Neumann, N., Böckenholt, U., and Sinha, A. (2016), A meta-analysis of extremeness aversion, Journal of Consumer Psychology, 26(2), 193–212[3].

We obtained the citation and nothing else.

Four observations, ours.

Extremeness aversion is the mechanism underlying the compromise effect, named in the title of the 1992 follow-up paper[3], so a meta-analysis of it is a meta-analysis of this.

Its existence tells us the literature is large enough to pool, which the seventy-fifth article's subject also is, and pooling is where decline effects and publication bias become visible.

This is the paper we would most want and did not get, and its absence is the largest gap in this article. A reader who obtains it will know the effect's size, which we do not.

So everything below about magnitude is our own construction rather than a measurement, and we have built the arithmetic to work at any effect size rather than to depend on one.

Why They Might Have Diverged

Speculation, flagged as such. Ours entirely.

Four observations.

The decoy effect requires a dominated option, meaning one that is worse on every dimension, and constructing one in a real market is difficult because nobody sells a product that is worse at everything.

The compromise effect requires only three options along a range, which is a far weaker condition and one that occurs naturally.

Which suggests a reason for the difference in robustness that has nothing to do with psychology. The fragile effect needs an artificial choice set and the durable one does not, and laboratory studies can build the artificial set while the world rarely supplies it.

And we did not obtain any paper testing this, so it is our hypothesis about why two findings diverged, offered as a reading and not as evidence.

What Actually Survives

Our reading, stated directly.

Five statements.

The effect was predicted from a theory and then found, which is a stronger evidentiary position than most findings in this series occupy.

It is reported as stronger when buyers expect to justify their choices, which points it at business purchasing specifically.

It appears to have fared better than its sibling, on the evidence of how a bibliographic service describes each, which is weak evidence about a real asymmetry.

A meta-analysis of the underlying mechanism exists and we have none of its contents, which is the gap in this article.

And nothing above tells you whether adding a tier will make you money, which is a separate question with an arithmetic answer.

What A Third Tier Does To Revenue

Our own arithmetic, on entirely invented figures. Two tiers: Basic at $50 and Pro at $100. Add Premium at $200.

Suppose share moves as the effect predicts, from 60 percent Basic and 40 percent Pro to 40 percent Basic, 50 percent Pro and 10 percent Premium.

Revenue per customer before: $70.00. Revenue per customer after: $90.00. An increase of 28.6 percent.

Four observations.

That is where almost every analysis of this stops, and it is a real increase that would show up in any revenue report.

The share shift is assumed rather than measured. We have no meta-analytic effect size and have chosen figures that are plausible and invented.

But the direction is not in question. If the effect operates at all, share moves toward the middle, and a middle priced above the old top tier raises revenue arithmetically.

And that is precisely why the next section is necessary.

One reason to distrust the revenue figure specifically, ours. It is the number a pricing change is judged on and the number a pricing change most easily moves, which makes it close to useless as evidence that the change was good.

Now Put Margins On It

Ours, same invented figures. A higher tier usually delivers more: more support, more onboarding, more of somebody's time. So its margin can be lower even though its price is higher.

Profit per customer, before and after, at various margin sets:

Margins 80 / 60 / 50 percent: profit goes from $48.00 to $56.00, up 16.7 percent.

Margins 80 / 55 / 45: $46.00 to $52.50, up 14.1. Margins 80 / 50 / 40: $44.00 to $49.00, up 11.4. Margins 80 / 70 / 60: $52.00 to $63.00, up 21.2 percent.

Three observations.

Profit rose in every one of those, though by less than revenue did, and the gap between the two increases as the upper tiers get less profitable.

Which is a useful finding and is not the finding we expected to report. The next section explains why.

And it raises the obvious question: is there a margin set where profit actually falls? There is, and finding it required correcting ourselves twice.

A Correction: The Tiers Were Equal

The first of two errors in our own working, documented here rather than removed.

We drafted a line claiming that a $100 tier at a 40 percent margin is worse than a $50 tier at an 80 percent margin. That is wrong.

One hundred dollars at 40 percent is $40.00. Fifty dollars at 80 percent is $40.00. They are exactly equal, and we had written "worse" because a strict inequality in our own code printed the boundary case on the wrong side.

Three observations, ours.

The corrected version is a better line than the one we drafted. At exactly that margin, the upsell doubled the price and changed nothing at all.

It is also a cleaner statement of the principle. Price times margin is the only quantity that matters, and a tier structure that doubles one while halving the other has accomplished a great deal of work for no result.

And it is the kind of error worth catching precisely because it is small. A boundary printed on the wrong side changes no argument and would have sat in the article indefinitely, which is what makes checking the tidy-looking rows worth the minute it costs.

And a boundary case printed on the wrong side is a small error with a specific danger. It would have made our threshold look more alarming than it is, and this series does not get to overstate in its own favour.

The Break-Even Condition

The rule, stated correctly. Ours.

Moving a customer up a tier pays only if the new tier's price times its margin exceeds the old tier's price times its margin.

Basic at $50 and an 80 percent margin earns $40.00. So Pro at $100 must hold a margin above 40 percent to be worth the move:

At a 30 percent margin, Pro earns $30.00, which is worse. At 40 percent: $40.00, which is exactly a wash. At 45: $45.00. At 50: $50.00. At 60: $60.00. At 70 percent: $70.00.

Four observations.

The threshold is the ratio of the prices. A tier at twice the price needs half the margin to break even, which is easy to state and easy to check.

It is also the calculation nobody runs, because the upsell is discussed in revenue terms and the margin sits in a different report.

And the threshold moves against you as the tier gap widens. A tier at four times the price needs only a quarter of the margin, which sounds generous until you notice that a four-times tier usually involves four times the delivery.

That gives a design instruction worth carrying. The question is never whether customers will move up, it is what they cost you once they have.

Two ways firms get this wrong in opposite directions, ours. Pricing the middle tier off the entry tier, as a multiple of it, sets the price before knowing the delivery cost and is how the margin gets squeezed.

And pricing it off what the market seems to bear has the same defect with better marketing behind it. Either way the number is chosen before the cost is known, and the threshold above is the only test that catches it.

A Second Correction: Deriving The Reversal

The second error, and it was a claim we made before checking.

We drafted a heading reading revenue rose but profit did not, and then printed a table in which profit rose in every single row. The claim was written before the numbers were looked at.

So we derived the condition properly. Under our assumed share shift, Basic loses 20 points, Pro gains 10 and Premium gains 10. The change in profit per customer is:

minus 0.20 times (50 times the Basic margin), plus 0.10 times (100 times the Pro margin), plus 0.10 times (200 times the Premium margin), which simplifies to minus 10 times Basic, plus 10 times Pro, plus 20 times Premium.

Which gives the condition. Profit falls when the Pro margin plus twice the Premium margin is less than the Basic margin.

Four observations, ours.

That condition is harder to satisfy than we had assumed, which is why our first table showed no reversal anywhere.

It requires the upper tiers to be substantially less profitable, not merely somewhat, and stating that honestly narrows who should worry.

The error is instructive about our own process. We wrote a heading that made a claim, and the arithmetic did not support it, which is the second time in three articles that has happened.

And the version we can defend is more useful than the one we drafted, because it says exactly which businesses are exposed rather than implying all of them are.

Two general lessons we would draw from making the same class of error twice, ours. A heading written before the table is a hypothesis, not a summary, and ours read like a summary.

And the direction of both errors was the same: both would have made our own case look stronger. That is the direction errors tend to run when a writer already knows what they want to say, which is a reason to check hardest at exactly the moment the result seems to confirm the argument.

When Profit Actually Falls

The condition applied. Ours, same invented figures throughout.

Margins 80 / 60 / 50: the test value is 1.60 against a Basic margin of 0.80, so profit rises 16.7 percent.

Margins 80 / 40 / 25: test value 0.90 against 0.80, profit rises 2.5 percent.

Margins 80 / 30 / 20: test value 0.70 against 0.80, profit falls 2.8 percent.

Margins 85 / 25 / 15: 0.55 against 0.85, profit falls 8.5. Margins 90 / 20 / 10: 0.40 against 0.90, profit falls 14.3. Margins 85 / 30 / 10: 0.50 against 0.85, profit falls 9.3 percent.

Four observations.

Revenue rises 28.6 percent in every row above, including the ones where profit falls by double digits, because revenue does not know about margin.

The reversal is real and requires a specific shape: a high-margin entry product and much lower-margin upper tiers.

That shape is not the software pattern. Where the upper tiers are mostly more software, margins are similar across tiers and the condition is nowhere near satisfied.

And it is the professional services pattern, which the next section sets out.

Who Is Actually At Risk

Narrowing the warning to the businesses it applies to. Ours.

Four observations.

The condition needs the entry tier to be the most profitable per dollar, which happens when it is standardised, self-serve or productised.

And it needs the upper tiers to consume a person's time, which is what makes their margin fall as their price rises.

Combine those and you have a specific and common business: a firm with a packaged entry offering and bespoke delivery above it. That describes a great many accounting practices, agencies, consultancies and trades.

And it explains a pattern owners of such firms report without being able to name. Revenue grows, the team is busier, and the bank balance does not move, which is what this arithmetic looks like from inside.

Two diagnostics that would show it, ours and cheap to run. Plot gross profit per customer by tier over the period you introduced the structure, rather than revenue, which is the comparison the effect hides in.

And check whether the middle tier's delivery hours per customer have risen since it became the popular choice. Scope tends to expand into the tier customers actually buy, and that expansion is invisible in a price list.

The Professional Services Case

The worked example, and the one closest to home. Ours, invented figures.

A firm with an 85 percent margin on a standardised entry product, 30 percent on a delivered middle tier, and 10 percent on a bespoke top tier.

Profit per customer before: $37.50. After: $34.00. A fall of 9.3 percent.

Revenue per customer before: $70.00. After: $90.00. A rise of 28.6 percent.

Four observations.

A business watching revenue would call that a success, report it as one, and repeat the exercise next year.

The 10 percent margin on the bespoke tier is not an exaggeration. Custom work with an owner's own hours in it frequently earns less than that once the hours are properly costed, and frequently nobody costs them.

And the middle tier at 30 percent is the real problem rather than the top one, because the middle is where the volume went. The top tier is only 10 percent of customers; the middle is half.

Which points at the fix, and it is not to abandon the structure. Raise the middle tier's margin or its price, because that is the tier the effect fills and the one carrying the result.

Two ways to raise that margin that do not involve charging more, ours and untested. Standardise what the middle tier includes, since bespoke delivery at a fixed price is where margin goes, and a defined scope is the difference between a product and a promise.

And move work out of the middle tier into the top one, which raises the middle's margin and makes the top genuinely worth its price rather than merely positional.

What The Top Tier Has To Earn

The last computation, on the tier that exists to make the middle look moderate. Ours, invented figures.

At Basic and Pro margins of 80 and 60 percent, the share shift is worth $4.00 of extra profit per customer. Against a top tier costing $15,000 a year to build and support:

At 100 customers a year, the shift earns $400. At 500: $2,000. At 1,000: $4,000. At 2,000: $8,000. At 5,000: $20,000.

Four observations.

Below about 3,750 customers a year the third tier costs more than the share shift earns, on these invented figures.

Which is a real constraint on small businesses specifically. The structure is cheap for a software company with tens of thousands of customers and expensive for a firm with two hundred.

Which inverts the usual advice, ours. Tiered pricing is presented as a small-business technique and is arithmetically a scale technique, because the benefit is per customer and the cost is annual and fixed.

The costs are the ones firms forget. A third tier needs pricing pages, sales training, contracts, delivery capability and support, whether or not anybody buys it.

And there is a cheaper version of the same idea that we would not dismiss. A top tier that is genuinely sold rather than merely displayed pays for itself, and the arithmetic above assumes it barely sells.

One further way to get the positional benefit at almost no cost, ours and untested. A top tier priced but delivered rarely can be defined narrowly enough that it is real without being a standing commitment, for instance as a bespoke engagement quoted from a starting price.

That keeps the structure honest, which matters on this publication's usual test. An option nobody can actually buy is not a tier, it is a prop, and a customer who tries to buy it and finds it does not exist has learned something about the seller.

What Practitioners Warn About

A commercial pricing consultancy's cautions, which align with the arithmetic. This is a consultancy's marketing page, not research, flagged here and at every use.

It warns: "Leveraging the preference-for-the-middle effect can harm your lower and higher-end products, negatively impacting your KPIs."[5]

And: "If you set an anchor too high, for instance, the portfolio may appear expensive compared to the competition."[5]

Four observations, ours.

The first warning is the arithmetic above stated qualitatively. Share moving to the middle comes out of the other two, and whether that helps depends on their margins.

The second is a point our arithmetic does not capture at all and which is a genuine limitation of it. We modelled the choice among your own tiers and ignored the competitor across the street.

A portfolio that reads as expensive can lose the customer entirely, which appears nowhere in a model where everyone buys something. That is the largest omission in our figures and we would rather name it than let it sit.

And we note the source's interest. A pricing consultancy has a commercial reason to say pricing structure is subtle and requires expertise, which does not make the caution wrong and is worth knowing.

One reason we cited it anyway, ours. The caution runs against the source's commercial interest in the specific sense that it warns the technique can backfire, and a marketing page that undersells its own subject is worth more than one that does not.

Why It Is Always Three

A structural point the arithmetic makes visible. Ours.

Four observations.

The effect requires a middle, and two options do not have one. Three is therefore the minimum structure that produces it at all.

Four or five options would each have a middle too, so the effect does not by itself explain why three is the near-universal choice. Something else must be capping it.

Our reading, offered as reasoning rather than evidence: each additional tier costs the same fixed overhead we computed above, and the positional benefit of a fourth is much smaller than that of a third, because the middle already exists.

And a longer menu runs into the finding the twelfth article in this series covered, where more options can reduce choice rather than improve it. Three is the smallest number that produces a middle and the largest that avoids that problem, which is our explanation and not anybody's finding.

One prediction that follows and could be checked, ours. Businesses with cheap tier overhead should use more than three, and businesses where each tier is a delivery commitment should use exactly three.

That is roughly what you observe, though we would not claim the observation as evidence. Software companies commonly list four or five and professional firms almost never do, and the fixed cost per tier is the obvious explanation.

Not The Same As Anchoring

A distinction worth drawing, since the two get merged constantly. Ours.

Four observations.

Anchoring is about a number shifting a valuation: seeing $200 makes $100 feel smaller than it otherwise would.

The compromise effect is about position in a set: the middle option gains share because it is the middle, and the 1989 mechanism is justification rather than valuation.

They predict differently in a case that distinguishes them. Anchoring predicts a single high number should work, with no third product needed; the compromise effect requires an actual option occupying the top position.

And the practical difference is what you have to build. An anchor can be a reference price on a page; a compromise structure needs a product you would have to deliver if somebody bought it, and the second is a real commitment with the costs computed above.

Two reasons the distinction is worth keeping straight, ours. A firm that thinks it is anchoring will invent a top tier it cannot deliver, which is the prop problem above and is a promise it will eventually have to break.

And a firm that thinks it is running a compromise structure when the effect is really anchoring is paying delivery costs for something a reference price would have achieved, which is the more expensive of the two mistakes.

The Customer Who Already Knows

The boundary condition, from the same practitioner source and consistent with the 1989 mechanism.

It records: "When customers know the exact product they want or need, they are less likely to be influenced by this effect."[5] Again, a consultancy page rather than research.

Four observations, ours.

That is exactly what the 1989 theory predicts. The paper's mechanism is choice under preference uncertainty, and a customer who knows what they want has no uncertainty for the structure to resolve.

So the effect should be strongest with new customers and weakest with returning ones, which is a testable claim about your own book and one you could check.

It also means a tier structure does more work in categories customers buy rarely, which for an accounting practice describes most of what it sells.

And it identifies the customer for whom the whole structure is friction rather than help. Somebody who arrives knowing they want the entry product now has to read three columns, and the design should let them skip it.

One design consequence worth stating, ours. A returning customer should be able to reach what they bought last time without passing through the comparison, which costs nothing and removes friction from the people most likely to buy.

Designing The Three

What follows from all of the above. Ours, and not pricing advice.

Four points.

Compute price times margin for every tier before launching, because that single number determines whether an upsell is a gain, and the threshold is the ratio of the prices.

Put the margin work into the middle tier, since that is where the effect sends volume and where the result is decided. The top tier's job is positional; the middle tier's job is to pay.

Cost the delivery honestly, including your own hours. The reversal condition is only reached when upper-tier margins are genuinely low, and a firm that does not cost owner time will not know whether it is there.

And check what the whole portfolio looks like against a competitor, which is the thing our arithmetic cannot see and which a practitioner source flags as a real risk.

One sequencing note, ours, because the order of these matters more than any of them individually. Do the margin arithmetic before designing the tiers, not after launching them, because a structure is far harder to withdraw than to price correctly at the outset.

A firm that raises a middle tier's price a year in is asking existing customers to absorb an increase. A firm that priced it correctly at launch is not asking anyone for anything, and the whole difference is a calculation that takes an afternoon.

Bibliographic Note

The series keeps a count.

A practitioner page gives the 1992 companion paper's page range as "281-29"[5] where the standard form is 281–295, a truncation.

Three observations, ours.

This one is on a commercial page rather than in an academic reference list, which is where we would expect looser citation practice and where it matters less.

It would nonetheless defeat a page-range lookup, which is the practical test we have applied throughout.

That brings the running count of bibliographic variants across this series to twenty-nine.

Two observations on where these now come from, ours. The recent variants have increasingly been on commercial and summary pages rather than in journal reference lists, which is where a business reader is most likely to encounter a citation and least likely to check it.

Which is a mild argument for the practice this series has followed throughout. Chasing a citation to the publisher takes about a minute, and it is the minute in which most of these would have been caught.

What To Do

Compute price times margin for each tier. That product, not the price, decides whether moving a customer up is worth anything, and a tier at twice the price needs half the margin merely to break even.

Check the reversal condition if your upper tiers are delivery-heavy. On our own arithmetic profit falls when the middle margin plus twice the top margin is less than the entry margin.

Watch profit per customer, not revenue per customer. On our invented professional services example revenue rose 28.6 percent while profit fell 9.3.

Put the margin work into the middle tier, because that is where the effect sends volume.

Cost your own hours into the upper tiers. The reversal only appears when upper-tier margins are genuinely low, and uncosted owner time is how a firm fails to notice.

Ask whether you have the volume to justify a third tier at all. On our invented figures the break-even was around 3,750 customers a year against a $15,000 annual cost.

Expect the effect to be strongest with accountable buyers and new customers, since the paper reports it stronger among those expecting to justify their decisions, and a practitioner source notes it weaker where the customer already knows what they want.

Look at the portfolio from outside. Our arithmetic assumes everyone buys something, and a lineup that reads as expensive can lose the customer entirely.

The Limits Of This Analysis

Several caveats matter. This article discusses research on pricing and is not pricing advice; the applications are our own reasoning and untested. Everything is verified to August 2026. We did not obtain the 1989 paper, only its abstract from the publisher, and our reproduction of that abstract cuts off mid-word in its third finding, which we have therefore not reported. We did not obtain the 1992 companion paper, any replication study, or the 2016 meta-analysis of extremeness aversion, which reaches us as a citation with no contents whatever and which is the largest gap here: we cannot tell you how large this effect is. Our account of the two effects' divergent fates rests on a bibliographic service's summaries of other papers' framings, which is weak evidence about a real asymmetry, and our explanation for why they diverged is our own speculation supported by no source. Two of the sources for practical cautions are a commercial pricing consultancy's marketing page rather than research, flagged at every use, and that source has an interest in pricing being seen as subtle. All arithmetic is ours and every price, margin, share and cost in it is invented; in particular the share shift from 60/40 to 40/50/10 is assumed, not measured, and we have no effect size from any source to justify it. This article contains two documented corrections to our own working: a boundary case printed on the wrong side, and a claim about profit falling that our own first table contradicted. And the arithmetic models only the choice among your own tiers: it assumes every customer buys something, ignores competitors entirely, and therefore cannot see the risk that a portfolio reading as expensive loses the sale altogether.

Related Reading

No. 85 The Paper Says The Opposite The study cited for 23 minutes of lost focus reports that interrupted tasks were finished faster. And the two figures everyone pairs cannot both mean... Behavioural Finance What Free Costs The Person Taking It The other pricing device whose break-even divides by your margin. Behavioural Finance The Parenthetical That Changed Everything Another case of watching the number that moves rather than the one that matters.

Frequently Asked Questions

What is the compromise effect?

The 1989 abstract reports that brands tend to gain share when they become compromise alternatives in a choice set. It was derived from a theory that people choose what they can best justify, and was predicted before it was tested rather than explained afterwards.

Is it more reliable than the decoy effect?

It appears to be, though our evidence for that is weak. A bibliographic service introduces the decoy effect by noting researchers have strongly questioned its robustness and introduces this one without qualification. A meta-analysis of the underlying mechanism exists, and we obtained none of its contents.

When is the effect strongest?

The abstract reports it stronger among subjects who expect to justify their decisions to others, which points it at business purchasing and at buyers who need sign-off. A practitioner source adds that customers who already know exactly what they want are less influenced, which matches the paper's own mechanism of choice under preference uncertainty.

Does a third tier make money?

It raises revenue arithmetically if share moves to a middle priced above your old top tier. Whether it raises profit depends on margins. On our own invented figures a shift lifting revenue 28.6 percent lifted profit anywhere from 25 percent down to minus 14, depending entirely on the margins of the upper tiers.

When does profit actually fall?

On our own derivation, when the middle tier's margin plus twice the top tier's margin is less than the entry tier's margin. That requires the upper tiers to be substantially less profitable, which is not the software pattern but is the professional services pattern where upper tiers mean more of somebody's time.

How do I know if the upsell is worth it?

Compare price times margin across the tiers. A $50 tier at an 80 percent margin and a $100 tier at a 40 percent margin earn exactly the same $40, so that upsell doubles the price and changes nothing. The threshold is the ratio of the prices.

Is a third tier worth building for a small firm?

On our own invented figures, a share shift worth $4.00 per customer against a tier costing $15,000 a year to maintain needs about 3,750 customers annually to break even. The structure is cheap for a business with tens of thousands of customers and expensive for one with a few hundred.

References

1. Publisher record for Simonson, I. (1989), Choice Based on Reasons: The Case of Attraction and Compromise Effects, Journal of Consumer Research, 16(2), 158–174, September, DOI 10.1086/209205, reproducing the abstract: that building on previous research the article proposes that choice behaviour under preference uncertainty may be easier to explain by assuming consumers select the alternative supported by the best reasons; that this approach provides an explanation for the so-called attraction effect and leads to the prediction of a compromise effect; and that consistent with the hypotheses the results indicate, first, that brands tend to gain share when they become compromise alternatives in a choice set, second, that attraction and compromise effects tend to be stronger among subjects who expect to justify their decisions to others, and third, that selections of dominating and compromise brands are associated with something our reproduction truncates mid-word. Note: the publisher's record. Our source for the theory and the findings. We obtained the abstract and not the paper, and the third finding is cut off in our reproduction and is therefore not reported. academic.oup.com
2. Economics database record for the same paper, confirming Simonson, Itamar, 1989, Journal of Consumer Research, vol. 16(2), pages 158–74, September, DOI 10.1086/209205, and recording that no abstract is available for this item and that it holds no bibliographic references for it. Note: an economics database record; citation only. Recorded because it illustrates that the same paper carries an abstract at the publisher and none at a major indexing service, which bears on how a reader's source determines what they can check. ideas.repec.org
3. Reference list carried on a peer-reviewed article in Marketing Letters concerning attribute ratings and their impact on attraction and compromise effects, confirming Neumann, N., Böckenholt, U., and Sinha, A. (2016), A meta-analysis of extremeness aversion, Journal of Consumer Psychology, 26(2), 193–212; Sheng, S., Parker, A. M., and Nakamoto, K. (2005), Understanding the mechanism and determinants of compromise effects, Psychology and Marketing, 22(7), 591–609; Simonson, I. (2014), Vices and virtues of misguided replications: The case of asymmetric dominance, Journal of Marketing Research, 51(4), 514–519; and Simonson, I., and Tversky, A. (1992), Choice in context: Tradeoff contrast and extremeness aversion, Journal of Marketing Research, 29(3), 281–295. Note: a reference list; citations only. Establishes that a meta-analysis of extremeness aversion exists, which is the paper we would most want and did not obtain. We obtained none of the works named here. link.springer.com
4. Bibliographic service record for the 1989 paper, carrying summaries of related and citing works, including one characterising the compromise effect as dictating that a decision-maker chooses a middle option over an extreme one given a set of choice alternatives since choosing an intermediate option is easier to do something our reproduction truncates; and another stating that researchers have recently strongly questioned the robustness of the attraction effect, according to which adding a decoy option to an existing choice set affects consumers' choice behaviour. Note: a bibliographic service's summaries of other papers' framings, not findings from those papers. Our only evidence for the asymmetry in how the two effects are currently described, and weak evidence at that. We obtained none of the summarised papers. semanticscholar.org
5. Glossary page published by a commercial pricing software company, citing Simonson (1989) and Simonson and Tversky (1992) and offering practitioner cautions: that when customers know the exact product they want or need they are less likely to be influenced by this effect; that leveraging the preference-for-the-middle effect can harm lower and higher-end products and negatively impact key performance indicators; and that setting an anchor too high may make the portfolio appear expensive compared with the competition. Note: a commercial pricing consultancy's marketing page, NOT an academic source, flagged at every use in this article. The publisher has a commercial interest in pricing structure being seen as requiring expertise. Recorded also as a bibliographic variant: it gives the 1992 paper's page range as 281-29, a truncation of 281-295. buynomics.com

This article discusses research on pricing and is not pricing advice. The 1989 paper was not obtained, only its abstract, whose third finding is truncated in our reproduction and therefore unreported. The meta-analysis of extremeness aversion was not obtained, so no effect size is available and every share figure below is an assumption. All arithmetic is the authors' own and every price, margin and cost in it is invented. This article contains two documented corrections to its own working.