It is important to evaluate the impact of hints on the accuracy of targeted high-value words and phrases, and not on global Word Error Rate (WER), which is unlikely to change significantly due to hinting. Look for significant improvements in analytics results, where categorization accuracy and key phrase counts are most important.

All words are not created equal. Consider, for example, that someone calls Joe's Hardware and says the following:

"Hi there, how are you. I'm fine thanks. Hey, I'm taking my cat to the vet later today and while I'm out I'd like to pick up a pair of Kobalt pliers. I've heard Kobalt makes great tools and I'm hoping you carry them."

While overall accuracy of transcription is important, it's clear that you would prefer to maximize the recognition accuracy of the term "Kobalt" and the phrase "Kobalt pliers", potentially at the expense of slightly lower accuracy during the initial pleasantries and less relevant information.

From the hardware store owner's perspective, there is little analytical value in knowing that this person is running the specific errand of visiting a veterinary clinic, or that the patient is a cat. What matters is that they have heard good things about the Kobalt brand and wish to purchase pliers from you, but only if you carry the Kobalt brand.