QoS means a guaranteed compute where one customer can not slow down another. Does this mean that the performance will vary based on global demand to some extent. We are getting results all over the place. Some days it takes minutes and some days it takes hours to process roughly 100 invoices.

Is there a practical business impact to it taking "hours to process roughly 100 invoices" for these customers running Smart Fields though backend process automation?

Separately, that's not the commonly used meaning of Quality of Service in a technical context, which is in almost all cases about networking. To quote Wikipedia's Quality of service article directly:

In the field of computer networking and other packet-switched telecommunication networks, quality of service refers to traffic prioritization and resource reservation control mechanisms rather than the achieved service quality. Quality of service is the ability to provide different priorities to different applications, users, or data flows, or to guarantee a certain level of performance to a data flow.

Abstracted to the application layer, QoS is a prioritization strategy. It doesn't prescribe a specific technical implementation like dedicated compute resources per tenant.

The strategy we use that Miruna outlined, auto-scaling and "fair queuing", is a modern, scalable, cloud-native approach. Within those there are various configuration levers - different scaling rules, fair queuing thresholds, etc. Fair queuing is the main mechanism to directly address "noisy neighbor" impact.

I know we actively service monitor metrics like average/max job time and adjust those configuration levers to try to keep relevant performance metrics within our service targets (which I don't know the specifics of).

As for the business case, I guess they are just trying to get the invoices out faster rather than waiting a couple hours.

The reason we are asking about QoS is because it is the measurement of the overall performance of a service through techniques in services that manage traffic and resources to guarantee performance. When we see massive varying in performance we can only guess that the impact is either one of the following:

• Something about the specific documents we are processing that takes a long time to analyze
• Someone else in the world who we are waiting in a queue behind

It's hard to say without seeing the whole picture, but I think you may be jumping to conclusions. While in most circumstances processing is expected to be within minutes, there may be other factors affecting processing speed. The upper bound for processing is about 4 hours before a processing request should be considered suspect/failed. 

There are various rate limits that could come into play for a given customer even without activity from other customers. So whether it was a batch of invoices or 100 invoices trickling through the day may make a difference. Whether workflow was involved may also make a difference. The number of pages could also be a factor.

If you want to look into this further, please open a case with Laserfiche Support and indicate the customer and the timeframe when processing occurred.