TL;DR
- Context
- Frubana (B2B marketplace, YC 2019, LATAM). The WMS gave traceability, but the inventory reconciliation process (“saneamiento”, matching physical counts to system records) was failing: operators used mental maths and paper instead of the system.
- Role
- Product Manager & Lead Product Designer
- Team
- The warehouse squad: data analyst, tech lead, engineers, product designer and low-code specialists.
- Timeline
- 2 months
- Outcome
- −65%inventory leakage30%faster task completion
The problem
Digital and physical inventory did not match, and nobody knew exactly why: this was the “unknown leakage”. Operators did repetitive eight-hour tasks with mental maths, fragmented tools (pen, paper, calculators) and inconsistent supervisor feedback. They did not trust the system and found its screens overwhelming. Supervisors could not trace discrepancies or record decisions, relied on word of mouth, and had no history to spot recurring errors.
Business stake: every unexplained gap between theoretical and physical stock is lost money and an unreliable record.
The key reading from the research: the inefficiency was a usability problem, not a skills problem.
My role
- Owned: framing the problem, research with operators and supervisors, redesign of the reconciliation flow, the native app direction, the design system, and leading the designers working on it.
- Team did: WMS engineering and delivery.
- Not mine: WMS traceability itself, which already existed.
Key decisions
Automate the calculation or train the operators?
Training and clearer instructions
Cheap and fast.
Remove the calculation from the operator’s job
ChosenNeeds build effort and scale integration.
- Trade-off
- Training is cheap and fast but leaves human error and cognitive load in an eight-hour repetitive task; removing the calculation needs build effort and scale integration.
- Choice
- The second. Operators now enter only the weight or units read from the scale; the system does the maths.
- Why
- Research showed they reached for paper and calculators because the system asked them to compute.
Mirror the physical process, don’t mirror the database
Keep the form-based module and fix its layout
Restructure the flow to follow the operator’s physical steps
ChosenGuided, one step at a time.
- Trade-off
- Restructuring means retraining and a bigger redesign; patching leaves cognitive jumps.
- Choice
- A guided step-by-step flow, smart alerts when something is inconsistent, progressive disclosure (only what the current step needs), search and barcode scan. Copy co-created with operators in warehouse vocabulary.
- Why
- The old screen packed LPN, SKU, quantity, adjustment and waste fields into one form with raw system errors; the new one asks for one thing at a time and explains what a warning means.
Make supervisors part of the flow
Report and record inconsistencies in the platform
ChosenReplacing word of mouth.
- Choice
- Inconsistencies are reported and recorded in the platform, replacing word of mouth.
- Why
- Supervisors could not trace or act on discrepancies, and without history recurring errors stayed invisible.
Native app and a design system, not a one-off screen
Redesign the one module inside the existing web UI
Build a native interface and a reusable system
ChosenFor very large data sets and every future module.
- Trade-off
- More upfront work, but the module handled very large data sets and other WMS modules faced the same problems.
- Choice
- A native app optimised for large data sets, with a new design system that became the base for other WMS modules and was later adopted company-wide.
- Why
- Consistency, scale and speed for every future module.
What we shipped
A redesigned reconciliation module in the WMS: guided flow, automated calculations, real-time inconsistency reporting for supervisors, a native app for large data sets, and a new design system.
- 1Confirmation: “Correct quantity, you can continue with the reconciliation.”
- 2Origin LPN (pallet label): scan the barcode instead of typing it.
- 3Quantity: the operator enters what is physically there; the system does the maths.
- 4Adjustments by type: New SKU, Merma (shrinkage) and Desecho (waste).
- 5Cancel, or Sanear (reconcile).
Results
| Metric | Result | How it was measured |
|---|---|---|
| Inventory leakage | −65% | Frequent mismatches between physical and system data before; real-time visibility after. |
| Task time | 30% faster | Manual maths and paper notes before. |
Inventory leakage
−65%
- How it was measured
- Frequent mismatches between physical and system data before; real-time visibility after.
Task time
30% faster
- How it was measured
- Manual maths and paper notes before.
The two numbers are the only quantitative claims. The rest is qualitative: adoption went from low trust in the system to “high engagement” ; communication moved from word of mouth to centralised digital reporting; onboarding moved from peer training to a guided UI for new hires .
What I learned
- Redesigning the work (who computes, who reports) moved the numbers more than polishing the form would have.
- Co-writing copy with operators was as important as the layout.
Transferable to any reconciliation workflow where records must match reality (billing and payroll ledgers, timesheet vs. roster approvals), and to products for non-technical frontline users who need guided flows and supervisors who need to see and act on exceptions.

