Redesigning cost estimation so advisors can have confident customer conversations
Problem Statement: The AI Core cost calculator existed online but was hard to use. Users had to guess what inputs meant, had no guidance on values to enter, and the output was in capacity units with no currency context. In one customer conversation, a miscalculation put the estimate at 1 million EUR per month. It was actually 1 EUR.
Business Needs: Presales engineers and solution advisors use cost estimates in every customer conversation. A tool that cannot produce a customer-ready number loses trust. The redesign had to fix the core usability problems first.
My Role: As UX Lead for AI Core, I owned the UX direction for the cost calculator from scratch. I set the research agenda, co-ran the study with a UX colleague, and drove the design from spec to dev handover.
Timeline: 3 months
Discovery Phase
Testing: Moderated usability study with 3 current users. 2 tasks: estimate costs for a RAG chatbot, then update for a production AI agent. A small sample, but enough to surface consistent critical patterns. Every issue below appeared across multiple participants.
Findings:
- Users had to guess inputs. Token counts, request volumes, and compute values were unfamiliar with no guidance to help.
- The output did not answer the key question. Capacity units required manual currency conversion, a step that had already caused serious miscalculations in real customer conversations.
- The compute section blocked confident presentation. One service dominated the total at ~60% with no explanation.
Impact: All 4 critical issues mapped directly to design decisions. Nothing in the spec was added without a finding behind it.
Design Decisions
- Template-based entry and autofill - select a scenario and size (S/M/L); the Orchestration section auto-populates from configured models. Eliminates the blank-form problem.
- Live cost panel with line-item breakdown - always-visible total, expandable by section. Directly addresses output comprehension.
- Contextual guidance throughout - pricing modes per model with explanatory copy, capacity unit definition inline, benchmark hints in compute fields.
Results and Next Steps
The redesign shipped. Telemetry is in place and will feed a continuous improvement cycle as real usage data comes in.
Credits
UX: Ana Luísa Cardoso Fonseca, Debarchan Sinha
Disclaimer: All images shown use mock data.
Next project
Redesigning process modeling for efficiency and adoption