Nine years of public charging data from Palo Alto, turned into a seven-page Power BI report. The biggest site isn't the best one, pricing cost demand but freed up capacity, and the first version of my own model was wrong. All three are in here.
Filters, a site drill-through and the validation pages, recorded live. The headlines and chart titles you'll see change with the filters, because they're calculated in DAX rather than typed.
Captions available. Prefer YouTube? Watch it there ↗
Each one comes with the caveat that keeps it honest. A finding without its limits is just an opinion with a chart attached.
Webster's 3 ports deliver 62.9 kWh per port per day, almost twice the network's 33.9. The largest sites sit below the line. Bryant delivers the most energy overall, but it needs six ports to do it.
MPL ranks near the bottom as a site. Drill in, and a few of its ports deliver a fraction of the network rate while the rest sit close to average. The answer there isn't more ports. It's finding out what's wrong with the ones it has.
Fees arrived in August 2017. Sessions per port fell 26%, comparing 2015–16 with 2018–19, while average idle time dropped from 44 to 18 minutes per session. That's timing, not proof of cause, and the report says so.
Each page answers one question, and the next page picks up where it leaves off.
The most useful page in the report is the one nobody asks for. These are the corrections that changed what the report says.
My first utilisation measure treated every port as available from 2011. Sites that opened later looked idle, simply because they were being judged on years before they were built. Counting each port from its first session fixed it, and the top site changed: Webster, not Hamilton.
The original plan assumed fees hadn't reduced usage. Per port, they had. The report states the finding the data supports, and the blueprint was rewritten to match.
Five weekdays against two weekend days makes weekends look a third as busy. Per day, they run at about 78% of weekday volume, which is a capacity story, not a quiet one.
Missing user IDs were stored as empty text, which quietly turned 7,677 anonymous sessions into one very active "driver". Converting them to null took them out of every user measure.
Most networks do. Plenty of numbers, not many decisions. Six years in EV infrastructure taught me which questions are worth asking of it.