Where are riders getting stuck?
Not every machine issue matters equally. The point was to find the repeat failure steps most likely to break a purchase attempt.
UTA • BI Model
Failure-path review for rider flow, support visibility, and service recovery.
This page explains the UTA work in plain language. The goal was to understand where ticket purchases were failing, which issues the support alert stream was missing, and how to rank the problems in a way leadership could act on.
Fare context
$2.0M
Illustrative annual context for the machine network
Recovery view
$15K
Estimated annual opportunity tied to top failure paths
Fleet scope
250+
Machines in the analysis view
Dominant path
31%
Share of failures tied to one recurring path
What This Model Answers
The business questions behind the analysis.
What is support missing?
The alert stream did not always reflect the rider experience, so the comparison between machine behavior and support visibility was essential.
What should get fixed first?
The output had to support prioritization, not just description: which failure paths were common, costly, and worth escalating.
BI Flow
How the analysis worked in plain English.
Step 1
Collect the events
Pull the machine events, alert records, and support context into one place.
Step 2
Rebuild the rider path
Organize the event stream into readable purchase attempts from start to failure or completion.
Step 3
Compare with alerts
Check which real-world failures showed up in support reporting and which ones did not.
Step 4
Rank the recovery work
Prioritize the failure paths by frequency, impact, and likely value of fixing them.
Tools Used
What I actually used.
Excel + Power Query
- Quick shaping and review of machine data extracts.
- Readable validation during early analysis passes.
SQL + Python
- Joining event sources and organizing purchase attempts.
- Ranking repeated failure paths and common patterns.
Power BI
- Surfacing the top failure paths and alert gaps.
- Giving leadership a clearer remediation view.
Decision Output
Why the model mattered.
This was not a technical exercise for its own sake. The value was making the machine problem easier to explain, easier to challenge with the vendor, and easier to prioritize for service recovery. That is the standard I use for BI work in general.
All Raw Data Remains Proprietary