This depends on the form of delivery of ASR technology. Enterprises can purchase base ASR recognizers from Nuance, SpeechWorks, Philips, and IBM for telephony applications. However, enterprises can also sign up for speech recognition services from a hosted voice application service provider.
When you purchase base ASR technology substantial nonrecurring engineering costs are required to develop applications that run on top of the base ASR technology. In such circumstances it could take years to recoup the initial investment.
With a hosted ASR services, however, the ROI can be net positive much faster. Hosted ASR services usually ask their enterprises customers to pay a small upfront fee and then pay on a per-transaction basis. In this situation, the enterprise might pay 50 cents a call, or 20 cents a minute. This provides an immediate savings over equivalent call handling by a live operator. Typically there are some modest upfront fees even for the hosted solution, but these are much smaller than required for the purchase solution.