Marketers waited eleven days for data teams to build segments. The self-serve builder existed and was abandoned within ninety seconds because it exposed schema concepts nobody outside data understood.
STUDY 10 / 30 · ANONYMISED · NDA · MARTECH · US · SERIES C
MARTECH · US Segment building for people who do not write SQL.
RoleLead designer + systems
Timeline14 weeks
Team2 designers, 1 PM, 2 frontend, 1 data engineer
VerticalCustomer data platform · 600 marketing teams
Rebuilt segment construction around business language with a live population count that updates as conditions are added, so users learn the data model by watching it respond.
Segment self-service went from near-zero to the majority of segments. Data team ticket volume for segment requests effectively ended.
84%SEGMENTS BUILT SELF-SERVEfrom 6%
4 minTIME TO FIRST SEGMENTfrom 11 days
-91%DATA-TEAM SEGMENT TICKETSvs baseline
THE ARGUMENT
Why the obvious solution was wrong.
The study matters because the product problem was reframed before the interface was polished.
Self-serve segment builders fail for a specific reason: they are a SQL query builder with the SQL hidden, so they still require the user to hold the schema in their head. A marketer knows they want people who bought twice and have not opened an email in sixty days. They do not know that this requires joining an events table to a profile table on a hashed identifier, and exposing that join is exposing the thing that made them ask the data team in the first place.
The builder speaks in business objects - people, purchases, sessions, messages - and resolves the joins invisibly. The critical mechanism is the live count: every condition added updates the population immediately, so a user who adds something that drops the segment to zero learns instantly and undoes it. This turns the builder into a feedback loop rather than a form. Users learn the data model empirically in the first four minutes, which no amount of documentation had achieved in two years.
THE INTERFACE CRAFT
The interaction, rendered as a working product surface.
The specimen below is code-native and uses the study's own design logic. The client interface remains protected.
MARTECH
MODEL CONFIDENCE
Paid searchHIGH
EventsLOW
UnknownDECLARED
DESIGN DECISIONS
Positions we would defend.
Each decision names the principle and the product consequence, not a stylistic preference.
Hide the schema, not the logic
Users can reason about business logic. They cannot reason about a join path, and asking them to is the failure mode of every builder in this category.
The count is the teacher
A live population count teaches the data model faster than documentation, onboarding or training ever will.
Prevent the near-duplicate
Segment sprawl is the second-order failure of successful self-service. Design against it from the first release.
PRODUCT LEADER READOUT
What transfers, and what should remain specific to this product.
A case study is useful when its operating principle travels without turning the original interface into a template.
Read the operating condition
For Customer data platform · 600 marketing teams, the transferable lesson is not a copied screen. It is the condition the interface had to make legible: Conditions are written against people, purchases and messages. The join path is resolved by the system and never surfaced. Rebuild that visibility for your own roles, risk, terminology, and operating cadence.
Protect the design rule
Users can reason about business logic. They cannot reason about a join path, and asking them to is the failure mode of every builder in this category. Keep that rule in the acceptance criteria, component states, and production QA record so later visual cleanup cannot erase why the interaction exists.
Measure behaviour after ship
The evidence record is 84% for segments built self-serve, from 6%. Recreate the baseline and outcome window before rollout, segment the result by role and context, and state clearly what the measure cannot prove.
RESEARCH RECORD
The work behind the interface.
These artefacts connect the final interaction back to the evidence and product model that produced it.
Abandonment replay
Watched 40 sessions of the old builder being abandoned; median time to abandon was 87 seconds, always at the first schema concept.
Business-object model
Worked with the data team to define six user-facing objects covering 94% of historical segment requests.
Count latency test
Established that above 600ms the live count stopped functionin as a learning mechanism; engineering targeted 400ms.
Segment sprawl audit
g Analysed 4,000 existing segments at a design-partner account; 38% were near-duplicates of another segment.
“The data team got eleven days a month back. That was the actual deliverable.”
Head of Growth Marketing, consumer subscription · under NDA
NEXT STUDY · 11 / 30 · MARTECH