Testing a price change before launch

A price change is hard to test with real customers because asking them reveals that the change is being considered. For Artificial Societies’ private equity client, putting future pricing, product consolidation or ownership changes directly to customers could expose its investment thesis and unsettle the market, as the commercial due diligence case study records.

Live price testing charges real customers different prices for the same thing. In September 2000, Amazon said it had run a five-day random price test on 68 DVD titles and had refunded 6,896 customers an average of $3.10 (Amazon.com statement, September 2000). In retrospect, the company said, the test was a mistake. Learning the answer at launch costs more. After Netflix split its DVD and streaming plans in 2011, raising rates for many members by 60%, it lost 800,000 US subscribers in the third quarter (Broadcasting & Cable, 24 October 2011). Its letter to shareholders said many long-term members “felt shocked by the pricing changes”.

In Artificial Societies’ commercial due diligence engagement, buyers’ stated loyalty looked strong until a price or product change was made concrete. A private equity firm was in early due diligence on a specialist software group with a portfolio of established brands. Working with a panel partner, Artificial Societies fielded the client’s survey with 130 software decision-makers, then built 130 digital twins, each anchored in one participant’s survey responses. The survey and the simulation answered different questions:

• •

  In the client’s primary survey, 98% of accepted responses said the organisation was likely or very likely to renew, and 83% described transferring historical data as difficult or very difficult.

• •

  Asked about moving to a new cloud-based product, 52% of Artificial Societies’ personas would accept no price increase for the move.

• •

  Offered a three-year agreement with a 10% discount, 48% of Artificial Societies’ personas would accept it only if it included a price freeze and an exit clause if service levels fell.

• •

  Told their current product would be retired in 24 months, 42% of Artificial Societies’ personas would review alternatives or leave rather than migrate.

For a pricing team, the second and third results matter most. With more than half of Artificial Societies’ simulated buyers accepting no increase at all, the result suggested that migration alone would not establish a compelling case for higher pricing. For 48% of the personas, the longer contract came with two conditions, a price freeze and an exit clause, that a proposal aimed at them would have to price in.

Observations of the real buyers come first in a simulated price test. In the commercial due diligence study, Artificial Societies enriched the 130 survey-anchored twins with deep market research, proprietary industry intelligence and client-supplied expert interview transcripts on procurement, implementation and operational workflows. Those sources informed modelled context around each persona, while observed responses stayed distinct from inferred characteristics.

The scenarios then name their terms. The due diligence follow-ups specified a three-year contract, a 10% discount and a retirement date 24 months away. A scenario that precise gives a persona something definite to accept, refuse or attach conditions to.

A test of several price points or tiers also needs each option kept apart from the others. Artificial Societies’ consumer goods study ran a five-armed randomised controlled trial on 1,498 simulated health-conscious consumers, comparing five product concepts without cross-contamination and exploring product names and feature options. That study tested concepts, names and features, not prices; a pricing version of the design would put one price point or tier in each arm.

In the commercial due diligence study, the client could explore results across the society and at individual persona level, combining distributions with the reasoning behind each response. Every survey question in Artificial Societies also asks for a verbatim reason behind the answer by default, as the consumer goods study notes.

A simulated audience fits before the standard pricing research methods, not in place of them. Van Westendorp, Gabor-Granger and choice-based conjoint all ask real respondents about price, and all three rest on what people say.

Sources: Huettermann, Haefliger and Stallone, Frontiers in Sports and Active Living, September 2021 (Van Westendorp); West and colleagues, PLOS Global Public Health, April 2023 (Gabor-Granger); Dezwaef and colleagues, PLoS One, July 2019 (conjoint); Artificial Societies’ commercial due diligence case study (simulated scenarios).

Even conjoint, the indirect method, still depends on stated preferences, Dezwaef and colleagues note (PLoS One, July 2019). A simulated audience does not escape that, because its answers are stated too. What it changes is the cost of asking again: Artificial Societies’ due diligence client could return to its society with new questions without commissioning a fresh panel. That suits the early stage of pricing research, cutting a long list of price points, tiers and terms to the few worth fielding, and finding conditions such as a price freeze to write into a Van Westendorp or conjoint study for real buyers.

AI pricing simulation predicts stated reactions, and stated reactions overstate what people pay. Artificial Societies measures its accuracy on survey answers: 86% distribution accuracy, the overlap between simulated and human opinion distributions, against a 91% ceiling set by humans answering the same question twice, across 1,000 surveys (Survey Evaluation Report, January 2026, on the evaluation page). That benchmark covers opinions, not purchases. Human stated willingness to pay runs high as well. A meta-analysis of 28 studies found stated values 35% above actual payments at the median, a ratio of 1.35, and choice-based questions reduced the gap (Murphy and colleagues, Environmental and Resource Economics, March 2005).

Research on language models queried directly, with or without fine-tuning, marks a further limit. Brand, Israeli and Ngwe found willingness-to-pay estimates from model responses realistic and comparable to human studies (Marketing Science Institute, October 2025). Fine-tuning on earlier survey data improved alignment for existing and new features, but not for new product categories or for differences between customer segments. That finding shows what to check first in any simulated price result: the differences between customer segments.

Artificial Societies starts a first price test from the options on the table and the evidence about the buyers. Bring the price points, tiers or contract terms you are choosing between, and whatever you already know about the audience. In the commercial due diligence study, the client brought its survey and expert interview transcripts, and Artificial Societies fielded the survey with 130 software decision-makers through a panel partner. In the consumer goods study, Artificial Societies gathered over 100,000 anonymised profiles with social listening data partners.

We build the society, run the scenarios and return the results as a report, with access to the society in Radiant so you can re-run a scenario when the pricing question changes. In the due diligence study, the private equity client could explore new scenarios without commissioning a fresh panel for every follow-up question. To scope a first test, book a call with the team.