AI Purchase Intent Research: Testing Trade-Offs Before You Launch
On this page
A launch owner needs to choose an offer against the incumbent and the option to buy nothing. Subconscious can compare feasible offers under those conditions and turn estimated choice differences, uncertainty and operating constraints into a launch recommendation. The useful result is an inspectable offer brief: which configuration to commit to, under which rule, and who owns the next action.
The Mintel comparison shows why the evidence behind that recommendation matters. The model reproduced a moisturizer choice task using demographics without human choice data as input. Reported distribution similarity was 0.79–0.88 across attributes against human results, including a second 1,000-person wave. The model diverged at the upper prices, where human choices fell more sharply. This is relevant evidence for inspecting represented preferences, with a visible price boundary; distribution similarity does not establish sales accuracy or a guaranteed launch result.
Ask for a comparison that can change the launch brief
Choose Subconscious when the unresolved question concerns how defined product, price or package alternatives affect modeled choice. The documented methodology assigns attribute levels across choice tasks and estimates effects with uncertainty. The buyer can inspect the intervention, baseline and response rather than relying on a single favorable purchase-intent rating.
Agree the task and deliverables through the study compatibility guide. Include feasible configurations, tested prices, meaningful competitors and an outside option. Review audience support, assignment, estimator, repeated responses and decision-relevant buyer differences. A complete-package comparison measures the package difference; attributing the result to one feature requires a design that isolates that feature.
The recommendation connects the returned contrast to implementation cost, feasibility and the smallest difference that would change the commitment. Product receives the accepted configuration and excluded variants. Marketing receives the tested promise and audience. The launch owner receives the prior rule, uncertainty and reason to select an offer, retain the current proposal or leave the decision unresolved. Agree these outputs in the brief rather than assuming every analysis delivers every business metric.
Represent the incumbent and the cost of switching
A purchase-intent rating records the response to the question and context supplied. Budget, availability, habits and switching costs can change the actual decision in either direction. Inspect existing customer records and previous comparisons first; adequate evidence can already settle the launch question.
Name the incumbent, complete offer and price instead of presenting the proposed product in isolation. Include implementation effort, contract commitments or an existing owned product when relevant. Buying nothing must remain possible when that is a realistic action. Keep the alternatives and surrounding information identical across the comparison except for the assigned changes.
A human conjoint validation study compared relative applicant-attribute effects with Swiss naturalization decisions. That human-method precedent does not validate synthetic purchase demand. A modeled choice response is also different from an observed purchase. A sales or demand forecast needs relevant market inputs, calibration and evidence for the actual endpoint, as the compatibility guide specifies.
Work through an offer against an incumbent
Consider an invented launch example, with no study run or claimed result. Northline plans a 20-litre black backpack weighing 600 grams. Offer A costs $49 and includes one main compartment, one front pocket and adjustable shoulder straps. Offer B costs $59 and adds a removable padded laptop divider; capacity, weight and all other stated specifications remain the same. The supplier confirms both complete configurations are feasible before the study.
Both offers include applicable tax and shipping within seven calendar days to a US address. Each includes a 30-day return of an unused bag for a full refund, with prepaid return postage and no extra charges. The fictional named incumbent, Harbor, offers a 20-litre black backpack with a fixed laptop divider for $45, weighing 650 grams, under the same delivery and return terms. The outside option is keeping the buyer's existing bag with no new purchase; the design does not invent a resale value or replacement deadline.
Confirm support for the proposed modeled audience of US adults considering an everyday backpack. Randomly assign the Northline A or B offer beside the same Harbor offer and no-purchase option. Keep images, instructions, placement controls and context equivalent. This estimates the complete Northline package contrast within the submitted set, including the changed price; the result cannot isolate a divider effect.
Request the B-minus-A difference in modeled probability of choosing Northline, arm counts and uncertainty accounting for repeated responses. Inspect the corresponding incumbent and no-purchase choices. Retain the exact offers, assignment, population settings, model, run and available response records. Generated explanations can suggest follow-up questions; explanations do not establish every psychological reason for switching.
Turn the result into an accountable offer commitment
Before results, the launch owner sets the smallest useful choice improvement, unit-cost ceiling and maximum production commitment. Recommend B only when the agreed interval supports that improvement and the supplier can meet the cost and commitment limits. Retain A when the added package does not justify the change. Keep the choice unresolved when neither action satisfies the rule. No illustrative winner or sales volume is supplied.
The Subconscious recommendation becomes the production brief: exact accepted offer, incumbent and outside option, estimated contrast, uncertainty, prior rule, actual cost assumptions and responsible launch owner. The result can support a bounded configuration commitment. A production quantity justified by expected sales additionally needs the market and adoption evidence required for a demand forecast.
Preserve the different decision in each application
New product launch. Compare the proposed offer with the incumbent and no purchase. Use the accepted comparison and operating limits to choose the feasible configuration for the launch brief.
Pricing and packaging. Define currency, billing period and complete terms. Test realistic levels and inspect uncertainty before selecting a price or package; calibrate separately when the requested output is market revenue or a demand curve.
Competitive positioning. Compare actual feasible alternatives. Closed-lost records and generated explanations can identify questions; a randomized attribute contrast tests a defined change rather than proving every cause of losing a deal.
Go-to-market allocation. Connect the relevant offer contrast to the cost and reversible scope of the proposed commitment. A conditional modeled choice difference does not supply campaign reach or return on spending by itself.
Resolve the remaining claim before expanding the commitment
Additional evidence follows the unresolved claim and the consequences of being wrong. Inspect existing human comparisons, purchases, product records and quality evidence before commissioning more research. Novel behavior, unsupported audiences, physical experience or an irreversible promise can leave a material gap that the modeled task cannot answer. Select the relevant evidence source for that gap; another human study is not an automatic step after every comparison.
Preserve specialist, legal and private-customer review where applicable. Keep the accepted brief, evidence and assumptions so a later owner can distinguish a changed condition from a decision already supported.
Book a Subconscious decision review with the proposed offer, named incumbent, no-purchase alternative, price and switching conditions, existing evidence, deadline and commitment cost. Agree the supported task, returned comparison and prior action rule. Make the next launch brief specific enough for the product and implementation owners to act.