Faster debugging with a new Audit trail search

How I redesigned a 4-year-old debugging tool to cut investigation time and reduce support dependency.

Data transformation from one spreadsheet to another
Data transformation from one spreadsheet to another
Data transformation from one spreadsheet to another

👥

Lead UX Discovery

🎓

Redesign the eXperience

🤝

-50% investigation time

🎉

80% use cases covered

The problem

Support and Integration teams relied on a 4-year-old audit trail to debug customer issues but it has become a bottleneck on itself:

  • 50% of the ticket require a log investigation

  • No pagination —> all database loaded from the landing page

  • Outdated filters returned too many irrelevant results

  • Misleading labels causing confusion —> user apply one by one filter to try to find their log

Critical use case: Teams need quickly answer —> who made an action, when, on what item.

The challenge

How might we reduce investigation time while improving log therefor and fast resolution ?

1st insight : Pendo analysis revealed a mismatch between filters behavior and user mental models.

Design implication :
These insights led to a clear design priority & KPI → Refine and prioritize filter criteria

Short-term goal : Internal users

Mid-terme goal: Enable HR customers to self-serve their own log investigation.

Before

After work

My approach

Discovery

Discovery

  1. I shadowed log investigation (intern / external UT)

  2. Mapped the filter flow with Pendo

  3. Categorize each type of item to embed filters

  4. Standardize date format in UTC for alignment

  5. Partnered with copywriter to clarify labels

Caption: 48% of users immediately check "item impacted"

Key finding: Users need visibility before diving into payloads.

Validation

Tree testing for the 3+ filters research
Ran user test to check the use cases —> covered 80% of real-word debugging scenarios


Design decision: Require 3+ filters before showing results
✓ Reduces visual noise + prevents accidental full-database loads
✓ Guides progressive refinement (matches natural debugging flow)
✓ Improves precision (90%+ relevant results vs. <50% before)
✓ Compare only the same item regarding the retention period
✓ Remove unnecessary and misleading icons in a technical environment

Trade-off: Adds friction for edge cases, but 80% of use cases benefited from clearer, faster results and readibility.


Validation

Tree testing for the 3+ filters research
Ran user test to check the use cases —> covered 80% of real-word debugging scenarios


Design decision: Require 3+ filters before showing results
✓ Reduces visual noise + prevents accidental full-database loads
✓ Guides progressive refinement (matches natural debugging flow)
✓ Improves precision (90%+ relevant results vs. <50% before)
✓ Compare only the same item regarding the retention period
✓ Remove unnecessary and misleading icons in a technical environment

Trade-off: Adds friction for edge cases, but 80% of use cases benefited from clearer, faster results and readibility.

What we observed after 6months ?

Background

3

filters provide useful results

80%

of key use cases are covered with the new UI

50%

research time to find logs (from 9 to5min)

The solution

A focused search experience that guides users through progressive refinement:

Step 1: Empty State with Context

  1. Clear explanation: "Apply 3+ filters to see results"

  2. Link to Confluence documentation for advanced use cases

  3. No auto-load = faster landing experience

Decision rationale: Empty state as onboarding —> guide correct usage rather than overwhelming with data

Step 2: Refined Filter Logic

  1. Reordered filters to match mental model:
    Affected item → Employee name/ID → Date range → Action type

  2. Added "All actions" option

  3. Dual entry for employees (by name or UUID)

Decision rationale: Filter order = cognitive scaffolding

Step 3: Readable results

  1. Standardized attribute order: Who → What → When → Item

  2. "See changes" link surfaces payload diffs at a glance

  3. Export filtered results for sharing with customers

What I learned

  1. Data + observation beats assumptions
    Pendo analytics showed the problem, but shadowing revealed why (mental model mismatch)

  2. Constraints drive clarity
    The "3-filter minimum" felt risky, but testing proved guided constraints beat open-ended flexibility for high-frequency tasks.

  3. Onboarding through design
    The empty state + documentation guided users reducing aa learning phase.

Next use case

What we observed after 3 months ?

Background
Background

3

filters provide useful results

3

filters provide useful results

60%

research time to find logs (from 9min to 5'')

60%

research time to find logs (from 9min to 5'')

80%

Real scenarios validated in testing

80%

Real scenarios validated in testing

What I learned

  1. Data + observation beats assumptions
    Pendo analytics showed the problem, but shadowing revealed why (mental model mismatch)

  2. Constraints drive clarity
    The "3-filter minimum" felt risky, but testing proved guided constraints beat open-ended flexibility for high-frequency tasks.

  3. Onboarding through design
    The empty state + documentation guided users reducing aa learning phase.

Next use case

Next use case

Next use case

Create a free website with Framer, the website builder loved by startups, designers and agencies.