AI/ML · PRODUCT DESIGN · ENTERPRISE SAAS · RESEARCH
Model Builder
Risk scores, eligibility checks, and pricing decisions powered the platform but the logic behind them was invisible to everyone except engineering. I designed the tool that changed that.
50%
Faster workflow creation
67%
Fewer configuration errors
30%
Reduction in task completion time
85%
Positive response to visual builder concept
/ PROBLEM
Experts spending their expertise on the wrong thing
Solution Architects at DecisionMines were spending 8+ hours on AI/ML model configurations that should take 2–4 hours. A 15% error rate meant frequent delivery delays and frustrated clients.
The core tension: the people best positioned to configure these models were domain experts, not developers. But the tooling forced a developer mindset on them.
PAIN 01
8+ hour configurations
Manual coding of each AI/ML model from scratch, relying on specialised skills unavailable to most domain experts.
PAIN 02
15% error rate
Cryptic error messages caused cascading failures during simulation, adding 2+ hour validation cycles per iteration.
PAIN 03
No system visibility
Workflows lived in code and disconnected spreadsheets. There was no way to see the whole picture — or show it to a client.
PAIN 04
Bottlenecked delivery
Business Analysts understood client needs but couldn't act. Every change required a developer detour, adding days to timelines.
/ RESEARCH
Twelve sessions to find the real problem
I ran stakeholder interviews for business context, user interviews with 12 system analysts and architects for pain points, ethnographic observations during live configurations to catch real behaviour, and workflow analysis to map inefficiencies end-to-end.
KEY PERSONAS
Senha Joshi
System Analyst · 5 years · Pune
PRIMARY PAIN
"I'll happily trust any tool that lets me see the entire workflow and tells me up-front if something will break."
GOALS
- Error free hand-offs
- Validate before delivery
- Neat audit trail
PAIN POINTS
- 2-3 hr manual validation
- Cryptic error logs
- No client-facing view
Satish R
Subject-Matter Expert · 8 years · Pune
PRIMARY PAIN
"Give me a canvas where I can drag nodes, hit simulate, and watch the metrics spike without writing another 100 lines of glue code."
GOALS
- Rapid iteration
- Reusable templates
- Client-ready dashboards
PAIN POINTS
- Slow experimentation
- Needs analyst to translate
- Can't visualise data flow
KEY INSIGHTS
Frame the problem before designing the solution.
Before any screens, I needed to understand how decision logic actually worked from both a business and a technical perspective.
Visual understanding is non-negotiable
Users needed to see workflow relationships, not just configure parameters. Without visibility, every change was a leap of faith.
"It's like flying blind."
— System Analyst
Prevent errors, don't just recover from them
Proactive guidance was far more valuable than sophisticated debugging. The cost of an error wasn't the fix — it was the hours lost to finding it.
ENGINEERING AUDIT · 4 SERVICE WALKTHROUGHS
Domain context drives model success
Banking and retail AI models behave differently. The tool needed to encode that knowledge, not expect users to carry it in their heads.
Configuration must be presentable
Stakeholders expected visual walkthroughs. A stack of config files wasn't a deliverable, it was a liability.
"All I have is code and config files."
— System Analyst
/ DESIGN
Three concepts.
One clear answer.
I explored three distinct directions before converging. Each was pressure-tested against both personas to understand what it actually solved and what it left broken.
Form based wizard:
Reasoning for partial consideration:
Addressed some usability issues but didn't solve core communication and understanding problems
Persona (Sneha response):
"Better than current, but still doesn't help me explain to clients
Persona (Satish response):
"Limits my flexibility, feels restrictive"
Decision:
Valuable for future phases, not core solution
Dashboard overview:
Reasoning for partial consideration:
Research showed users needed visual understanding, not better coding tools
Persona (Sneha response):
"Good overview, but how do I actually build workflows?"
Persona (Satish response):
"Nice for monitoring, but doesn't solve creation problem"
Decision:
Valuable for future phases, not core solution
Visual workflow builder
Reasoning for partial consideration:
Aligned with research insight that visual understanding was essential
Persona (Sneha response):
"Finally! I can see what I'm building and show it to others"
Persona (Satish response):
"This lets me experiment quickly and see relationships"
Decision:
Selected as core concept direction
/ DESIGN SOLUTIONS
Four integrated solutions, one seamless journey
Each solution addressed a specific research finding. Together they form a coherent journey from card delivery to first purchase.
Visual workflows over enhanced forms
Research showed that visual understanding wasn't just a personal preference, it was essential for stakeholder communication. Sneha needed to show clients something real.
Progressive disclosure architecture
Simple surfaces for Sneha, deep controls for Satish both in the same tool. Tabs for Properties → Input → Configure → Output let experts go deeper without overwhelming others.
Real-time simulation integration
The 2+ hour validation cycle was the biggest confidence-killer. Embedding simulation directly into the canvas meant users could test incrementally and catch errors early.
Domain-specific templates
Encoding industry best practices (banking vs retail vs insurance) into the starting point reduced setup time and prevented category errors before they happened.
/ THE OUTCOME
Numbers that rebuilt trust with clients
50%
Faster workflow creation
67%
Fewer configuration errors
30%
Reduction in avg task completion time