← Sapien

Case study: Deodorant category

What is an aluminum-free claim really worth?

Sapien measured how eight common deodorant claims change consumer choice, estimated the premium each claim supports, and compared those premiums with pricing observed on the shelf. The results separate the claims that carry real value from the ones the category overprices.

8 claims tested1,248 consumer profilesReal shelf pricing

What we found

Sapien estimated that an aluminum-free deodorant claim supports a +$0.49 premium4. The human benchmark for the same claim is +$1.97, measured in a large-scale incentive-compatible conjoint by Michigan, Duke, and Chicago researchers1.

Consumer purchase data puts the realized market premium near +$0.03 per equivalent stick, about one percent at the median price and statistically indistinguishable from zero3. Against that commercial benchmark, Sapien came substantially closer to the value brands actually capture.

Sapien’s estimate sat $0.46 above the observed premium, while the human conjoint estimate sat $1.94 above it. On an absolute-error basis, Sapien was approximately 4.2× closer to observed shelf pricing.

Human conjoint estimate1+$1.97
Sapien estimate4+$0.49
Observed shelf premium3≈+$0.03

Aluminum-free premium, dollars per equivalent stick

Human research measured a strong theoretical preference. Sapien more closely approximated the value the market actually captured.

The figures

EstimateAluminum-free premiumDifference from market
Sapien estimate4+$0.49+$0.46
Human conjoint estimate1+$1.97+$1.94
Observed shelf premium3≈+$0.03 (+1.0%)Reference

Interpreting the result

The human benchmark comes from the Michigan researchers’ conjoint1. Their deodorant experiment, fielded in September 2020 with 155 participants, measured the value of aluminum presence as −$1.97. Because Sapien tested the attribute as aluminum-free, the equivalent human willingness-to-pay benchmark is +$1.97. Sapien estimated the same claim at +$0.494.

At first glance, this appears to be a substantial understatement relative to the human study. Shelf pricing, however, suggests that the human estimate overstates the premium consumers actually pay. Stated-preference research is known to inflate willingness to pay: respondents spend hypothetical money under controlled trade-offs, with none of the friction of a real purchase.

Across consumer purchase data covering 1,400 deodorant listings, aluminum-free variants sell at roughly the same price as conventional alternatives once brand and package size are held constant. The estimated realized premium is approximately +$0.03 per equivalent stick, about one percent, and statistically indistinguishable from zero3.

Sapien’s +$0.49 estimate therefore came within $0.46 of observed market pricing. The human estimate was $1.94 higher than the market benchmark, roughly 4.2× further away.

This does not necessarily mean the human estimate is incorrect. Conjoint research measures the value consumers assign to an attribute under controlled trade-offs, while shelf pricing also reflects competition, retailer strategy, production costs, brand positioning, and promotions. But when the commercial question is how much of a claim’s value currently appears in market pricing, Sapien was much closer to reality.

Synthetic versus human research

Sapien’s deodorant result sits inside a broader research picture. The Harvard Business School paper “Using LLMs for Market Research”2benchmarked GPT-generated conjoint responses against the same human research tradition: their fluoride pilot landed close to the human estimate, their follow-up studies overshot it by roughly 3.2×, and their fine-tuned model recovered the correct negative sign for an implausible concept. Across every attribute shown, synthetic and human methods agreed on direction; the magnitudes are where they diverge.

parityAluminum-free (Sapien)Fluoride (pilot)Fluoride 1AFluoride 1BPancake flavorHUMAN CONJOINT ESTIMATE ($)SYNTHETIC ESTIMATE ($)
SapienGPT baseline, Harvard Business School2

Willingness-to-pay estimates in dollars. Every point shares its sign across both methods.

The other claims tested

Aluminum-free was one of eight claims tested against the same simulated population. The full report ranks every claim by willingness to pay and states the premium each one supports.

01
Aluminum-free
02
Long-duration protection
03
Claim name available in the full report
04
Claim name available in the full report
05
Claim name available in the full report
06
Claim name available in the full report
07
Claim name available in the full report
08
Claim name available in the full report

Relative willingness to pay. Full ranking in the report

Commercial takeaway

The Michigan researchers’ conjoint1 suggested that aluminum-free deodorant could support a premium of +$1.97. Consumer purchase data shows the market currently captures approximately +$0.033. Sapien estimated +$0.494, placing it $0.46 from the observed market premium and approximately 4.2× closer than the human conjoint estimate on an absolute-error basis.

Human research measured a strong theoretical preference. Sapien priced the claim the way the market does.

Exploratory single-category result. Shelf premiums reflect competition, retailer strategy, promotions, and production costs as well as consumer willingness to pay.

Sources and methodology

  1. [1]

    Fong, Guo & Rao (2024), "Debunking Misinformation About Consumer Products: Effects on Beliefs and Purchase Behavior," Journal of Marketing Research, 61(4). The deodorant conjoint, fielded September 2020 with 155 participants, measured aluminum presence at −$1.97; the equivalent aluminum-free contrast is +$1.97, as reported in the Harvard Business School paper [2]. doi.org/10.1177/00222437221147088

  2. [2]

    Brand, Israeli & Ngwe (2025), "Using LLMs for Market Research," Harvard Business School Working Paper 23-062. Source of the GPT baseline estimates shown for comparison: the fluoride pilot (+$3.40 vs. +$3.27 human), Studies 1A and 1B (+$8.20 and +$9.30 vs. +$2.60 and +$2.90), and the fine-tuned pancake-flavor concept (−$2.40 vs. ≈−$7.00). hbs.edu, working paper 23-062

  3. [3]

    Amazon US marketplace data, licensed through the Keepa Product API, July 2026. Ninety-day Buy Box listing prices across a 1,400-ASIN deodorant frame; the premium is estimated with brand fixed effects and package controls on 345 within-brand comparison rows. Aluminum-free carries approximately +$0.03 (+1.0%) at the median price, statistically indistinguishable from zero. These are listing prices, not transaction records.

  4. [4]

    Sapien estimate: a simulated consumer panel answering the same attribute-and-price choice design tradition as [1], scored with a conditional logit with an opt-out option and respondent-cluster bootstrap intervals.

Get the full analysis

Every figure, the full comparison with human research, and where the premium actually sits.

Tell us where to send it. The full analysis PDF arrives in your inbox within minutes.

We’ll email you the full case study PDF. No drip sequence, no surprise sales calls.