Brand preference
Prefer169
Open60
Reject23
U&A and segmentation
“U&A is fine in total. The cuts the team wants are not.”
Make every U&A cut deeper.
A U&A or segmentation study lands, then the team asks for cuts the original sample cannot support: heavy users by region, lapsed loyalists by tenure, early adopters within Gen Z. Simulacra expands measured segments while preserving the relationships used to define them.
Crosstabs causing you problems?
The total sample holds. The activation cuts do not.
Activation
What do you do when activation cells are underpowered?
A segmentation that's defensible at the cluster level can still be unreliable for the within-cluster cuts marketing wants for activation.
Booster loss
Relationships lost in re-fielding
A booster sample fielded weeks later loses correlation structure with the original wave. Cuts on the merged data don't behave like the original.
Persona testing
Personas can't be tested
"Does this persona respond differently to claim X" needs a coherent in-segment dataset that booster panels rarely deliver.
Cuts the original sample can't support
Make every activation cut a viable read.
Segment expansion
Boost low-incidence measured segments up to 10× while preserving the relationships that defined them.
Stable reads inside each segment
Read claims, behaviors, or preferences inside the segments the original sample could not support, while preserving the measured relationships that define each segment.
Segment response under scenarios
Run measured-variable scenarios against a boosted segment to see how that segment responds under different conditions.
Segmentation workflow
Pick the segment. Set the boost. Read the cut.
Start with the U&A or segmentation study you already fielded, choose the segment the team needs, and preview the outcome read the original sample could not support.
Measured segment
Boost changes read stability and row count, not the segment definition.
Priority outcome
252 generated rows from 42 fielded rows in Heavy users, Midwest, for an activation cut the fielded study could not support.
Fielded segment42 rows
Generated rows252 rows
OutcomeBrand preference
SegmentHeavy users, Midwest
Claim response
Strong161
Mixed60
Weak31
Purchase intent
Likely154
Open63
Unlikely35
Segment cut
Brand preference share
Heavy users, Midwest
67%
Lapsed loyalists, 35-44
48%
Gen Z early adopters
70%
Premium-aware switchers
61%
- Readout
- Generated cut, outcome distributions, response by segment.
- Export
- CSV segment table, SPSS-ready cuts, generated-row flags, methodology appendix.
- Guardrail
- Unsupported intersections are not estimated. Generated rows stay inside the measured segmentation schema.
Workflow preview. Pre-computed example; no Causal Engine API call.
Research, validated
The validation question: do relationships survive expansion?
Twin-2K-500
Pairwise MAE 0.005 across 717 variables
Response-structure preservation on the hardest public benchmark: Twin-2K-500 tests whether generated rows preserve relationships across many survey variables, the property segment expansion depends on.
Data ReductionGeneralizes across consumers, products, and attributes
Random-sampling regime proves Simulacra isn't memorizing — it's learning the structure that makes new segment cuts viable.
Expand your segments
Bring a U&A or segmentation. We'll expand the cuts your sample can't support.
Simulacra fits the response structure of your sample, boosts low-incidence segments up to 10×, and returns cell-stable cuts inside the measured schema.