Research FAQs
When Should I Create a Segment vs. Skip Segmentation Altogether?
Short Answer
Create a segment only when you need a clearly defined consumer cohort that will repeatedly power your research. Do not create a new segment for every run—segments must be created intentionally, because they determine the inputs that build your AI Twins, respondents, and ultimately the accuracy of your research.
Detailed Guidance
1. What a “Segment” Means in consumr.ai
Unlike demographic filters in traditional tools, a segment in consumr.ai is a formal audience definition used to generate:
AI Twins (qualitative personas)
Respondents (quantitative mini-twins powered by shard memories)
Insight reports (Behavior, Intent, Mentions)
Segments are foundational because all research begins with the correct definition of the cohort you want to study. consumr.ai uses the segment description to determine the exact combination of keywords, URLs, 1P/2P/3P audiences, job titles, and intent signals required to create clean insight assets → which then power the Twin → which then powers both qual and quant research.
If the segment is wrong, every downstream output becomes unreliable.
2. Why Segments Matter So Much
Over 4 years, consumr.ai has built proprietary technology that:
Accepts audiences, keywords, URLs, job titles, interest signals, and more
Connects to APIs across Meta, Google, Snapchat, Pinterest, TikTok, etc.
Produces Behavior Reports, Intent Reports, and Mentions Reports
These three reports form the intelligence backbone of an AI Twin:
Behavior → persona & personality
Intent → active mindsets
Mentions → real-world conversations & sentiment
Once these are fused with a conversion mindset, consumr.ai generates a Twin that represents hundreds of thousands of similar consumers, not outliers.
Twins then generate respondents for quant studies, ensuring:
4,000–10,000 representative respondents
Weighted to ACS / Meta / World Bank universes
< 1% MoE
Deff close to 1
True 95% confidence at scale
Because everything derives from the segment definition, accurate segmentation is non-negotiable.
3. When You SHOULD Create a Segment
Create a segment when:
✔You have a clear research objective that requires a stable, repeatedly addressable audience.
Example: Frontline Diabetes Care Physicians Seeking Practical, Patient-Centric Treatment Solutions Used by Pfizer’s diabetes division to build representative Twins and respondents.
✔ The cohort is distinct, meaningful, and materially different in needs or behaviors.
If a group has a unique mindset, purchasing journey, or decision logic, segment it.
✔ You require consistency across multiple studies.
Brand trackers, message testing, AEI, longitudinal behavioral studies, or multi-wave creative testing all rely on stable audience definitions.
4. When You Should NOT Create a Segment
Avoid creating a segment when:
✘ You are running a one-off question, brainstorm, quick meeting, or exploratory investigation.
These do not need a fully defined segment.
✘ You haven’t clarified the research objective yet.
Creating segments prematurely leads to incorrect inputs → incorrect insights → incorrect twins.
✘ The audience is not meaningfully different from an existing segment.
Redundant segments increase noise and fragmentation.
✘ You’re trying to capture every micro-variation in behavior.
A good segment covers ~85% of your audience type. The remaining 15% are acceptable outliers and do not require their own segment.
5. How Often Should Segments Be Updated?
Rarely!
Consumer landscapes shift, but not daily. You should update or rebuild a segment only when there is clear evidence that:
The market has dramatically moved
New behaviors or intents consistently appear
The segment is no longer representative
Internal segmentation frameworks have changed
Otherwise, segments are meant to be long-term assets, much like research panels.
6. The Key Principle
Segments are not operational objects. They are strategic objects. Every AI Twin, every respondent, and every insight report inherits the intelligence defined by your segment.
Create them carefully, use them consistently, and avoid unnecessary proliferation.
Practical Example
Good Use Case for a Segment: "Urban Gen-Z Female Shoppers Motivated by Sustainability & Value" Used repeatedly across creative testing, AEI, brand tracking, and journey diagnostics.
Bad Use Case for a Segment: “People curious about this ad I just uploaded today.” This should simply use existing segments; no new segment should be created.
Are consumr.ai survey results shown at an individual respondent level or at a cohort level? How do open-ended questions work?
By default, the results you get from consumr.ai are at the cohort level, not at the level of individual people. So if you define a cohort of 100 people, you won’t receive 100 individual answer rows. Instead, you’ll receive aggregated outputs such as percentages or distributions that describe how that group responds overall.
For example, if you ask a closed-ended question like, “How likely are you to buy this product?”, you won’t see 100 separate ratings. What you’ll see is something like: 45% likely, 30% neutral, and 25% unlikely. This tells you how the group behaves as a whole, which is what most business decisions are based on.
This approach matches how products and marketing campaigns are usually designed. They are built to appeal to the majority, not to extreme opinions. If you imagine all responses plotted on a bell curve, decisions are typically guided by responses near the middle of that curve. Those data points close to the mean represent how most people think and behave.
Because of this, outliers are not considered by default. Outliers tend to add noise and can distort the picture of what the majority believes. For example, if 95 people feel a product is reasonably priced and 5 people feel it should be free, including those extreme views in the main analysis would skew the outcome without helping a mass-market decision.
However, if your goal is to understand those edge cases, you can do that by defining them as a separate segment. For instance, if you specifically want to study highly price-sensitive consumers, you can create a cohort focused only on that group. As long as the audience size is statistically meaningful, the system will work with it. The important point is that outliers are handled intentionally, not mixed into a broader group where they introduce noise.
Open-ended questions work a bit differently but follow the same principle. You don’t get individual verbatim responses from each person. Instead, the system looks at how people in the cohort typically explain their thinking and summarizes those explanations into common themes.
For example, if you ask, “Why does this product not appeal to you?”, you won’t receive 100 different text answers. Instead, you might see insights such as: people feel the product is too expensive, the value is unclear, or it feels too complex to use. These explanations reflect the dominant reasoning patterns across the group, not one-off opinions.
So in simple terms, closed-ended questions give you percentages that describe the group, and open-ended questions give you summarized explanations of how the group thinks. Individual responses and outliers are filtered out by design, unless you explicitly choose to study them as their own cohort.
Last updated