We have been having the exact same question concerning the revenue of Lokad. In my experience, tracking usage and forecasting revenues are very different matters.

As far usage is concerned, SaaS companies (cloud or not) have a significant advantage over desktop apps: you can track the usage of virtually any component. This could very handy to figure out what interests your customers most (based on what they do, not what they say).

Then, forecasting revenues is a different story. Unless those companies are going throw only a hand few massive B2B deals, I don't think that the CRM is the right place to do the forecasts. Indeed, pay-per-user can involve rapid changes in customer behavior that is best tracked directly at the app level (*), not the CRM level.

(*) You can use a monitoring infrastructure to do that.

Ex: at Lokad, customers frequently reverse-engineer our pricing to exclude low sales items from the forecasts. Thus, when the trial period ends, so does (frequently) the number of forecasting tasks (that's because of our pay-per-use pricing).

Thus, if you need to forecast your revenues, I would suggest to start by collecting key usage data indicators. Forecasting the usage (using a simple statistical model, say auto-regression) seems more robust as opposed to forecasting the revenues because cloud apps tend to frequently update their pricing, and customers respond by adapting their behavior too (to some extend). Then, once the usage forecasts is established, revenues can be inferred based on the pay-per-use pricing.