Customer signals — every finding

Some buses run at the wrong time. Gjirafa can fix this. One card per line — tap it to read what riders said.

Fixable lines come first. 4 of these 12 have a real time from a rider, so we can fix them now. The rest need the operator's answer. We trust the userfeedbacks report over a memory — it comes from a real trip. Reviews rank first, then both-source lines, then feedback-only ones. An operator with nothing bookable anywhere always ranks last.

How big is it? Each card compares riders who reported the problem to people who viewed that route's page. Reports are all-time; views cover the last 30 days — the two cover different windows, so read the percent as a size, never as a rate. It does not order this list.

How big is it? Complaints — reviews plus userfeedbacks, pooled — against passengers who bought a seat, all-time. This is not a complaint rate: it sizes the problem, it does not rank this list. A line nobody can book shows not reservable rather than disappearing.

The tags mark only 352 of these as schedule-related, but an AI read of all 4,470 pieces of rider text finds 1,199. Schedule problems are about 3.4× more common than the tagged dashboard shows.

Both free-text sources, counted together. Reviews: 4,323 comments read, 1,138 schedule-related, 798 never tagged with the schedule category. Post-trip feedback: 147 written messages read, 61 schedule-related, 49 carrying no schedule reason tag at all. Merged because the defect is identical — a schedule complaint the labels did not count.

What these schedule problems are

Action: treat schedule accuracy as a top platform issue — the tag-based view understates it about 3.4×. Most common failure: Ran late. (AI-screened leads; spot-check before quoting an exact figure.)

Each card is one line, one problem you can fix. Ranked by how fixable it is, not how bad the trip felt: a rejected ticket outranks a rude driver we can only report. Reviews and feedback tags keep separate counts, never summed. The only figure on a card that pools them is the reports-vs-views box described below, where a headcount is being sized against a headcount. Riders' own words sit behind the button.

Every line you can book comes before every line you cannot, always. An unbookable route shows not reservable. Within that group, a selling-nothing operator ranks last. Then: reviews first, both sources next, feedback alone last — severity breaks ties.

How big is it? Each card compares riders who reported the problem to people who viewed that route's page. Reports are all-time; views cover the last 30 days — the two cover different windows, so read the percent as a size, never as a rate. It does not order this list. 1 of these finding(s) carry no box: we have no page-view data for that route.

How big is it? Complaints — reviews plus userfeedbacks, pooled — against passengers who bought a seat, all-time. This is not a complaint rate: it sizes the problem, it does not rank this list. A line nobody can book shows not reservable rather than disappearing. 1 of these finding(s) carry no box — no sales or booking data.

Problems spanning several lines with none reaching the bar alone aren't here — no single line or departure to name. All 88 are in section 2, with the operators.