Product planning & marketing
Who will this product resonate with?
Compare product, price, and messaging across customer segments. Narrow down the candidates to try first and the points to revisit.
Customer surveys, purchase history, etc.
Urekata × CoeSignal — validation for enterprise
A simulation
built just for your company —
shall we create it?
Put your own customer data and reviews to work in
a validation framework matched to your products, customers, and markets.
We design it together, from data integration to the validation method.
Standard Urekata is a self-service web app: sign up to compare options with a fixed panel of 100 synthetic personas. This page covers custom development using your company data or a tailored panel.
Standard product & pricing
View the sample without signing in
Products and themes still under consideration are welcome.
See it through your customers’ eyes.
Easy to use
on busy mornings?
on busy mornings?
How does it differ
from my usual?
from my usual?
Your product
PriceMessaging
LifestylePrice sensitivityProducts compared
Test hypotheses under conditions close to your customers.
Under conditions close to your customers
Uses your reviews and real data
Designed through comparison with real research
What you can discuss
Start with what you want to decide now.
From new product launches to improving existing products and expanding overseas.
Product improvement & customer understanding
Why aren’t they choosing it?
Organize the expectations and complaints in reviews. Build improvement hypotheses from the reasons customers hesitate to buy.
Reviews, word of mouth, inquiries, etc.
Overseas expansion & new markets
Will it resonate overseas?
We design validation conditions based on local lifestyles, price sensitivity, and substitute products.
Local reviews, market data, research, etc.
What validation gives you
Material to answer "so,
what do we do?" in the meeting.
It doesn’t end with collecting reactions —
we organize the options to try next and the questions to verify.
Validation reportExample structure
Three perspectives that reveal the next move.
- 01
Who to reach, with which option
Reactions by customer segment and differences between options
- 02
What to change to get through
Barriers to purchase and hypotheses for revising price and messaging
- 03
What to verify next
Items to confirm before real research or sales tests
An example of the deliverable structure. Items and format are decided to fit the validation scope.
Validation design to improve accuracy
How to bring AI closer to your customers.
The key is incorporating human data and checking it against other data.
Featured research / Kolluri et al., EMNLP 2025
Fine-tuning on human responses
improved unseen experiments.
26%
improvement in alignment with
human response distributions
A study using 210 social science experiments in the US with about 2.9 million responses. The 26% is a comparison against the model before fine-tuning — not a purchase-prediction accuracy or an Urekata result.
Read the original paperHow we design enterprise validation.
- 1
Build the real customer into the conditions
Beyond demographics, we organize usage, purchase reasons, and complaints from your data and reviews.
- 2
Compare against responses not used for tuning
We compare rankings and segment-level tendencies with real research to check where it fits and where it diverges.
- 3
Identify where it works, then test the next step
We narrow hypotheses from the results and connect them to the real research or sales tests needed.
About the study’s conditions and methods
Paper: Finetuning LLMs for Human Behavior Prediction in Social Science Experiments. The figures compare Socrates-Qwen-14B with Qwen2.5-14B. The study covered multiple-choice responses in the US and does not show that the same results hold for product purchases in Japan or elsewhere. Incorporating data into inputs and additional model training are separate things. We apply the research findings to validation design; specific methods and whether fine-tuning is needed are considered case by case.
From consultation to validation
Start with one thing you want to decide.
- 01
Tell us what you’re unsure about
We ask about your product, customers, and what you want to decide. Unorganized materials and data are fine.
- 02
Review the validation plan
We outline the approach, required data, deliverables, estimated cost, and schedule.
- 03
Start within the agreed scope
We validate a narrowed theme and organize the options to try next and what else to verify.
Before you reach out
Can we consult you even if our theme and data aren’t decided yet?
Yes. You can consult us from the stage of "what could we validate with this product?" We review the information you have and organize the additional data or customer questions needed. You do not need to attach real data to the first form.
How is this different from regular Urekata?
Regular Urekata asks 100 public synthetic personas for reactions to a product. For enterprise, we design individually around your goals — using your own data, organizing reviews, target markets, and comparison with real research. The scope is decided through consultation.
How are cost, timeline, and deliverables determined?
They vary with the validation theme, target markets, the state of your data, and whether real research is involved. In the first consultation we ask about your goals and conditions, then outline the scope, deliverables, cost estimate, and schedule. Work begins after you agree to the proposal.
Can AI results be used directly as purchase predictions?
AI interest levels are not actual purchase rates. The intended use is to narrow options and hypotheses first, then verify with consumer research or sales tests before major investment decisions. The figures in the cited paper do not represent Urekata’s accuracy or track record.
Urekata × CoeSignal
The right framework for your company —
it starts with thinking it through together.
Our team will outline a validation approach and cost estimate suited to your goals.
No need to attach real data.