Sawtooth Software vs an AI Conjoint Platform: Which Fits Your Pricing Decision
On this page
Sawtooth Software and an AI conjoint platform answer the same question, which package at which price do buyers choose, with different respondents. Sawtooth's Lighthouse Studio and Discover field choice-based conjoint to human respondents you recruit or buy; its CBC page lists multinomial logit, latent class and hierarchical Bayes estimation, calling HB "the most popular approach". An AI conjoint platform such as Subconscious, the one AI platform this page covers, runs a designed choice experiment of the same kind on simulated respondents, without recruiting anyone, and reports the estimated effect of each price and feature level with a confidence interval (method described on fidelity.subconscious.ai); agreement of those simulated intervals with human-study intervals has not yet been shown experiment by experiment. Choose Sawtooth when the decision needs a human sample on record or an adaptive design. Choose a simulated experiment when you need candidate price structures to test before a human study, or when fielding the audience is slow or costly; the simulated result is a set of candidates, and the registry record for a matched published study shows whether direction and rank held. The sequence below uses both.
What each one is
| Sawtooth Software | Subconscious (AI conjoint) | |
|---|---|---|
| Respondents | Humans you recruit or buy. The software licence carries no per-respondent fee; Sawtooth's Panel service advertises "over 200 million verified respondents worldwide" and shows no sample price on the pages read ([home page](https://sawtoothsoftware.com/) and [pricing page](https://sawtoothsoftware.com/pricing), read 2026-11-12) | Simulated respondents matched to census statistics from the American Community Survey and the National Household Travel Survey ([fidelity.subconscious.ai](https://fidelity.subconscious.ai/), read 2026-11-12) |
| Products | Lighthouse Studio (installed Windows software for choice-based conjoint, with free survey hosting; its Premier tier is the only one that includes ACBC, ACA and CVA) and Discover (web-based, "Simple CBC capabilities", in the Free and Professional tiers); consulting and panel services ([CBC page](https://sawtoothsoftware.com/conjoint-analysis/cbc), [pricing page](https://sawtoothsoftware.com/pricing) and [home page](https://sawtoothsoftware.com/), read 2026-11-12) | A market simulation that runs designed choice experiments and reports effects with confidence intervals |
| Methods | CBC, which Sawtooth notes "is known as discrete choice modeling (DCM) or discrete choice experiments (DCE)"; ACBC; ACA; CVA; MaxDiff. Estimation by MNL, latent class or hierarchical Bayes, "the most popular approach" (CBC page) | Factorial experimental designs, simulated respondents matched to ACS and NHTS statistics, and discrete-choice estimation with confidence intervals, as described on [fidelity.subconscious.ai](https://fidelity.subconscious.ai/) |
| Licensing | Annual subscriptions only; a free tier up to 50 respondents per survey; Professional and Premier tiers on request, no public amounts; invoices under $13,000 can be paid by card (pricing page FAQ, read 2026-11-12) | Studies are scoped through a decision review |
| AI features | "Smart follow-up" and "Open-end analysis" listed as AI features; no synthetic or simulated respondents found on the pages read | Simulated respondents in place of recruited ones |
| Time to a first result | Sawtooth states no turnaround on the pages read; fielding depends on the sample | No recruitment or fielding step |
Sawtooth's pages did not show publication dates; every Sawtooth fact above was read on 2026-11-07 and rechecked on 2026-11-12. Sawtooth's tier table lists up to 100 attributes per conjoint exercise in Free and Professional (Discover) and up to 250 in Premier (Lighthouse Studio), and up to 120 MaxDiff items against up to 2,000 (pricing page).
What the evidence says about each
Each vendor's case figure below is that vendor's own account of a single engagement. Sawtooth's home page carries a Google NBU customer story in which "predicted market share matched actual results within 2%" (home page, read 2026-11-12). A fielded CBC study measures human choices for the tested offers. No matched comparison of the two methods exists in these sources, so the 2% figure and the registry figures below do not rank the vendors.
Subconscious's evidence is public and study by study. The July 2026 working paper, not peer reviewed and summarized on fidelity.subconscious.ai, reports a mean Spearman rank correlation of estimated choice parameters of 0.55 across roughly 300 replicated published studies and 0.73 across the 43 that pass design filters. The public study registry (read 2026-11-12) adds per-study comparisons: 308 studies carried a detailed comparison; the registry labels direction as holding in 127, mixed in 92 and failing in 89, and rank as holding in 103, mixed in 93 and failing in 112. In the 27 studies that state a size ratio, simulated effects ran from 1 percent of the human size to 3.8 times it, and none was graded equivalent. The validation page lists the registry's study-level comparisons and their misses.
That evidence sets the division of labor. Where a matched published study holds on direction and rank, the registry's supported uses are screening which attributes move choice and ranking levels by how strongly they move it; where it does not, the registry records for Mele 2021 and Unterberger 2021 list "Nothing yet without human confirmation", and a matched study does not validate a new population or design. On Mintel's moisturizer conjoint, run on demographics alone against two waves of 1,000 human respondents, the simulated attribute distributions reached 0.79 to 0.88 Jensen-Shannon similarity, and the model missed the sharp drop in human demand between $34 and $48 (Mintel case). A human choice study on the leading offers is the step that measures direction, order and size for the offers you will ship.
Choose Sawtooth when
- A human sample must be on record for the decision.
- The design needs adaptive methods (ACBC), which Sawtooth lists.
- Your audience is reachable and affordable through a panel.
- Procurement or a regulator requires human respondents for this study.
Choose a simulated choice experiment when
- You need to screen several price structures or packages before you can afford to field one.
- Fielding the audience is slow or costly; validated performance for a specific specialist audience is not established.
- You want candidate rankings with uncertainty, and a matched registry study holds on rank.
- You are deciding what to test live, not what the live result will be.
Doing both
The proposed sequence: run the simulated experiment across the full price grid and package set, take the leading offers into a fielded CBC or a live test, and forecast from the human results. The Mintel price-threshold miss is the kind of question the human step answers.
The pricing pillar sets out that workflow with what the Finta, Mintel and New Age Floral cases each document. To scope a study, book a decision review.
How this page was made
By: Avi Yashchin, CEO and Co-Founder. Drafted with AI assistance on 2026-11-07 and updated on 2026-11-12. Sawtooth facts were read on sawtoothsoftware.com (home, CBC and pricing pages) on 2026-11-07 and rechecked on 2026-11-12; the pages show no publication date. Subconscious figures come from the fidelity.subconscious.ai summary of the July 2026 working paper, the public study registry (read 2026-11-12) and the public Mintel case page.