Weedmaps · Research strategy · Consumer segmentation
Defining the consumer behind the marketplace
The company had qualitative archetypes, behavioral data, and multiple teams making different assumptions about the same customer. I helped turn that uncertainty into a segmentation framework the organization could actually use.
Context
This work was for Weedmaps, a regulated marketplace where consumers browse, compare, and place orders through retailers.
Teams had plenty of inputs, but no shared model. Lifecycle saw the customer through CRM behavior. Marketing saw campaign audiences. Product saw flow decisions. Data Science saw what could be observed in the platform. Design saw the gaps between what people said they needed and what the interface asked them to do.
An earlier qualitative study had defined five consumer archetypes, but the model had not been tied cleanly to behavioral data. Two years later, the organization needed to know whether those segments were still useful — and whether they were actionable enough to guide product, growth, and lifecycle decisions.
The brief
What was asked for
Validate an existing five-segment consumer model with behavioral data and make the segments usable across product, marketing, and lifecycle planning.
What it actually was
Define what "validated" meant when motivation-led archetypes did not map cleanly to observable behavior. The design problem was not naming personas. It was deciding what evidence was strong enough to shape strategy.
Constraints
Every project has the constraints you negotiate and the ones you live inside. The ones that shaped this work:
- Qualitative archetypes are rich, but hard to measure. The original segments were based on motivation, confidence, and cannabis relationship — signals that do not always show up directly in clickstream or purchase behavior.
- Behavioral data is accurate, but ambiguous. A user who shops deals is not automatically a value-driven consumer. A new user is not automatically curious or inexperienced. The same behavior can come from different motivations.
- Regulated markets change the meaning of the same behavior. Product availability, delivery access, medical status, and local rules all shape what a customer can do. A jurisdiction-blind segmentation model would produce weak product decisions.
- Different teams needed different answers. Product needed prioritization signals. Marketing needed audience language. Lifecycle needed testable cohorts. Leadership needed a clear growth thesis. One model had to serve all of them without pretending their questions were the same.
Decisions
Start with the organization's questions, not the segment labels
I created a stakeholder survey for PMs and partner teams before the segment model was finalized. The goal was to understand what decisions the model needed to support, not just whether the segments looked believable on a slide.
Use the existing study as scaffolding, not scripture
Starting from scratch would have been slower and less useful than pressure-testing the qualitative model already in circulation. I helped frame the work as validation and refinement: keep what the data supported, challenge what it did not, and document the difference.
Treat data limits as findings
The analysis could support three of the five segments more cleanly than the others. Instead of smoothing over that gap, I pushed to make it visible. A partially confirmed model is not a failure if the organization understands what it can and cannot be used for.
Separate newness from curiosity
The most important methodological issue was a proxy problem: "new user" had been treated like a stand-in for curiosity, but every segment starts as new. That distinction changed the next research plan and prevented the team from over-targeting a cohort it had not actually defined.
Tradeoffs
- Existing model vs. blank slate — chose use the qualitative archetypes as a starting point because the organization already had language around them over treating every label as settled before the data could support it.
- Honest ambiguity vs. executive neatness — chose a three-confirmed, two-requiring-more-research story over a cleaner five-segment framework that would have been easier to present.
- Research architecture vs. premature activation — chose resetting the research questions so activation would not be built on weak proxies over moving straight into lifecycle experiments.
Working with the room
I worked across Data Science, Lifecycle, Marketing, Legal, Product, and Leadership to keep the model grounded in how it would actually be used. The key artifact was not a persona poster. It was a shared decision frame: which segments were supported, which were directional, which assumptions needed more research, and which teams were accountable for answering the next set of questions. That mattered because segmentation can become performative fast. If the model cannot change roadmap decisions, lifecycle tests, research priorities, or product messaging, it is just an internal taxonomy. This work kept the model connected to decisions.
Outcome
The project surfaced which parts of the existing model were behaviorally supportable, which parts needed qualitative follow-up, and where the organization had been relying on weak proxies. It also gave cross-functional teams a shared vocabulary for discussing the customer without overclaiming what the data could prove. The honest part: this work did not produce the satisfying finality of a polished segmentation model. It produced something more valuable — a clearer understanding of where the model was strong, where it was speculative, and what research needed to happen before the business acted on it at scale.
In hindsight
Most of the value was in refusing to pretend the model was more certain than it was. Research strategy is design work when it changes what the organization believes it knows. The most important design decision in this project was not how to visualize the segments. It was how to protect the quality of the decisions that would be made from them.
Next case study →
Making legal limits understandable before checkout
Weedmaps · Regulated marketplace · Compliance design