A streaming service worked with Gatestone to bring AI assistance into everyday support, with approved knowledge and agent review.
The AI demonstration retrieved useful answers. Using those answers during a live support conversation was more difficult.
At a streaming service, advisors had to open another tool, check the source and rewrite a suggestion before using it. The extra steps stood between the pilot and daily adoption. There was also no clear route for correcting outdated guidance or dealing with an unexpected response.
The service needed to establish where the assistant belonged in the work, including the moments when an advisor needed to set its suggestion aside.
Gatestone helped narrow the initial scope to common account and subscription questions supported by approved knowledge. The team defined what the assistant could suggest, which questions required a person and how source information would be maintained.
Suggestions were placed within the support workflow, with the supporting source available for review. Testing included representative questions, incomplete requests and known exceptions. This brought the focus back to the advisor's task: understanding the customer's question and finding guidance that applied to it.
Human review remained part of the process. The assistant did not independently approve refunds, change subscriptions or send customer responses. Account decisions and data-access rules stayed under the streaming service's control.
Gatestone prepared a trained group to use the assistant and established a fallback to normal support.
Incorrect suggestions, substantial edits and unresolved questions became inputs to a maintained improvement backlog.
Knowledge corrections and technical problems had defined routes for action. The review looked beyond whether advisors opened the tool.
Search effort, source accuracy, corrections and customer follow-through were considered for the same types of question. That kept the assessment connected to the work the assistant was meant to support.
The engagement gave the pilot a defined role in the service. Advisors had assistance they could review in context, a way to report problems and a familiar route when the tool could not help.
Broader use remained a decision for the client, informed by the experience of the people handling the conversations.