B2B SaaS 路 Enterprise AI/ML development platform 路 Template-first model-building workflow.
Back to work
Access to AI
Enterprise AI Ecosystem
A B2B SaaS case study about making AI development less intimidating through templates, plain language, progressive disclosure, and room for expert control.
50%reduction in time-to-value with templates and pre-trained algorithms40%usability performance improvement (Useberry, 25 participants)40%less user time to find data, algorithms, and solutions
Case snapshot
What I owned, shaped, and proved.
UX Designer owning user research, low-fidelity flows, high-fidelity design, usability testing, and handoff.
Solution owners, business analysts, data scientists, and ML engineers working at different levels of AI confidence.
2 moderated 45-minute interviews, affinity mapping, personas, journey mapping, and Useberry testing with 25 participants.
Product, ML engineering, data scientists, business analysts, and the Tredence dev team.
Reduce blank-slate intimidation, make selected assets visible, use business-language templates, and keep expert controls available.
Design judgment
My judgment was to make AI feel less like a room only experts were allowed to enter. That did not mean dumbing it down. It meant starting with the business problem, giving people a confident first step, and revealing technical depth only when it helped them move forward.
2 in-depth moderated interviews showed people were blocked less by AI itself and more by complex, unfamiliar workflows.
In testing, users could not tell which data, algorithm, or solution they had selected, so they froze on 'where do I click?'
Ready-to-use templates, pre-trained algorithms, and clearer selected-state feedback gave people a confident way in.
Before/after quantitative testing with 25 Useberry participants showed a 40% performance improvement.
Process evidence
The thinking I had to make visible.
Reconstructed artifacts that show what I was trying to understand, where trust felt fragile, and how the direction became safer to build.user map
Different users, different confidence gaps
I used this map to see where confidence broke differently for each user type, instead of designing one generic AI builder for everyone.
Data scientists
Model control / Advanced metrics / Experiment comparison
ML engineers
Deployment / API endpoints / Infrastructure fit
Business analysts
Use-case templates / Plain language / Confidence signals
Why it matteredPersonas kept the design honest to real user goals instead of one generic AI builder.
journey
Template-first creation
This journey reframed templates as a way to begin with intent, not as shortcuts for people who could not build from scratch.
Choose intent
User signalUsers arrived with a business problem, but the old entry point asked for technical model confidence first.
Design responseStart with use cases like churn prediction, sales forecasting, anomaly detection, and demand prediction.
Connect data
User signalSetup and data wrangling ate up time that should have gone to the business problem.
Design responseSupport common connectors and keep selected assets visible so choices never disappear.
Train
User signalNon-technical users did not know whether advanced metrics meant the model was usable.
Design responseShow Good, Fair, or Poor first, then reveal expert metrics when needed.
Deploy
User signalTeams stalled before production because deployment felt like a separate technical world.
Design responseProvide generated API endpoints, test prediction, and cloud deployment as one guided close.
Why it matteredTemplate-first entry let users start from a business problem instead of a blank slate.
iteration
Technical language to business language
This comparison made the accessibility problem obvious: technical precision was arriving before user confidence.
Entry point
BeforeXGBoost Regression Model.
AfterPredict Sales Revenue.
Training state
BeforeAdvanced metrics appeared before users knew whether the result was good enough.
AfterGood, Fair, or Poor appeared first, with advanced metrics available for experts.
Why it matteredPlain-language naming reduced hesitation before users had committed to anything.
decision matrix
Wizard vs dashboard vs workflow builder
This matrix helped choose a structure that offered direction without taking away agency.
Why it matteredThe guided card-based flow balanced structure with the freedom to revise.
progressive disclosure
Novice confidence, expert control
This board helped hold two truths at once: business analysts needed a confident path, and technical users still needed depth.
Default layer
Business use caseRecommended templateSimple training statusGuided deploy
Expert layer
Model familyFeature controlsAdvanced metricsInfrastructure settings
Confidence signal
I know where to startI understand what changedI know if this is good enoughI can test before handoff
Why it matteredBefore/after quantitative testing with 25 Useberry participants showed a 40% performance improvement.
Final design
Final Product Screens
Project artifact supplied by Savita for this case-study narrative.
Impact and results
What the work delivered.
Ready-to-use templates and pre-trained algorithms reduced time-to-value by 50%Quantitative usability testing with 25 Useberry participants showed a 40% performance improvementClearer asset access reduced user time to find data, algorithms, and solutions by 40%Simpler, more intuitive workflows supported higher platform adoption
Source: 2 moderated 1:1 user interviews (45 minutes each), plus before/after quantitative usability testing with 25 participants on Useberry. The Tredence dev team built the interaction design of the final product.