Those are great points - let me speak on what I know first.
The indoor gunshot detection system I work with has two features which give it the functional accuracy we're discussing:
It uses two different factors to determine if a gunshot has occurred: sound, and the IR flash of a gun. Both must be present to determine a gunshot has occurred. ShotSpotter relies solely on sound, which leads to a lot of false alerts (the main complaint of the system in places like Wilmington DE)
All of the processing is done at the sensor and integrates to a proprietary alerting interface that can use a phone, SMS, digital radio, email, etc to alert whoever you want. This eliminates the need for a human to monitor the system, which is ok because you've got a dual factor method to determine a legit gunshot.
This system was of course born out of an outside gunshot detection system, developed by DARPA and privatized. It has 17 million hours of operation with a false alert of either kind: it never missed a real gunshot, and never reported a fake one. But it has a limit, and that's a 40 ft range to assure its accuracy, and it needs to be inside. That's what this really comes down to: being honest about the limits of the tech you're pitching.
Their outdoor system was successful as well, but because of the method it was deployed:
Unlike ShotSpotter, this outdoor detection system was used in a perimeter for specific buildings or mobile patrol vehicles that were already experiencing incoming fire. As in, the bullet itself is moving through its sensor network. As a bullet of sufficient caliber, speed, etc traversed the sensor area, the sound of the travelling round, the air it displaced, trajectory, etc was captured and could determine the caliber of bullet, distance from the shooter and elevation. But the bullet had to cross the sensor field, coordinated among a group of individual sensors.
This detail is crucial because it's the exact opposite of the way ShotSpotter is deploying: placing them in a grid and just listening for a sound from any direction. This gives you the least accurate picture of what is occurring in relation to gunfire, basically 'loud bang in the average decibel range of a gunshot'.
Now, knowing this, here's what I think. ShotSpotter owns the data because it doesn't want it disseminated because it would paint a less than stellar picture of its performance. The technology just doesn't work the way they're marketing it, and if you judged officer response time with vs without ShotSpotter, I doubt you'll see any trend upward for any length of time that's statistically viable.
For it to be an effective tool, and if it worked the way it was marketed, it would need help: integration with a VMS with facial recog and high-quality cameras to support an evidentiary chain once the system detected an accurate gunshot. In that way, it can feed LEOs data about the scene before they arrive. The value of that, however, is honestly up for interpretation and is pretty contextual.
In short, it doesn't really work, but there's a huge incentive to a lot of people to make it seem like it does. ShotSpotter went public not too long ago, I believe.
Sorry for the wall of text, but I enjoy the discussion.