After a couple weeks of her manually designing pickup routes for the "fleet" I hacked together a demo using TIGER data and OR-tools that planned routes (with autogen'd tick-sheets) and calculated a rudimentary cost-per-mile. While it wasn't saving game-changing amounts of time over her initial route optimizations, removing the manual overhead of generating routes on a daily basis was a big win.
I would say it became a "fleet management" project (if you can call 2 cars, a trailer, and the occasional rental to pick up the slack a fleet) after one of the drivers suddenly heard a "grinding" sound while on route 5 miles from the nearest gas station. Based on the description I guessed that the transmission was dying and had them check the fluid. It didn't register on the dip stick so she dispatched another vehicle to tow back the trailer while the other one limped to the shop. The offending vehicle was one of the fraudulent VW TDI's that just had to get to the dealer's lot for the trade-in value (our goal had been to _barely_ get there under it's own power and it seemed we went a bit too far :) ). "Too far" was using it to tow ~1200# of compost 4 days a week for 6 months with what we determined to be one oil change at the beginning of that period. This experience led to adding a step to the route planning calendar events for maintenance checks for the vehicles in the "fleet" based on their current and expected mileage.
There were a bunch of things that I (in my night-time, engineering resource capacity) wanted to add to the system (tracking with GPS built into car, RFID readers in car and on bucket/totes to remove driver's manual annotations, CANbus diagnostic readout to identify early issues) but before any of those got off the ground she sold the company (hooray!) when we found out she was pregnant with twins and she realized she was way more interested in the physical composting work than the cost-center minimization of the drivers and the fleet anyway.
I'm aware of things like pgRouting but wondering if there is some basic FOSS software that does some rudimentary route assignment with multiple vehicles involved instead of leaving it up to the end user to batch destinations or vehicles before feeding them as batches into the routing engine. You can kind of get there yourself doing some basic clustering stuff, but again pretty manual. If I was doing it today at the same (small) scale my approach would probably be to create some drivetime polygons with my constraints around central areas and use to batch my stops before feeding them into pgRouting, but there's probably some way more intelligent way of doing it.
I count myself extremely lucky to have found an 'ideas guy' (and people person, and bookkeeper, money raiser, etc etc...) who came out of the world of the 90s and 2000's where he'd fly around the world implementing copies of high touch software.
From day 1 his vision was clear, we had an in-house subject matter expert, and the customers were there when our SaaS was ready.
The experience or knowledge of a problem seems like an extremely important aspect of starting a business, but it's often talked about as being "generated" or thought up.
The more I've worked on my own idea, the more I've realized that I needed to better understand the problem space and refine my vision.
Like I doubt anyone has ever had their own itch for “garbage truck fleet management software “, but I’m sure there is money in it. Or I suppose people have parents / family in the garbage truck industry, so you can “scratch a relatives itch”.
Personally my SaaS has nothing to do with myself nor anyone I know, but it’s an opportunity of weak software I found. I assume this is also a path people go down...